b0503b202afb4356caaca794518fb4e21082a48f
[git/git.git] / builtin / pack-objects.c
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"
19 #include "thread-utils.h"
20
21 static const char pack_usage[] =
22 "git pack-objects [ -q | --progress | --all-progress ]\n"
23 " [--all-progress-implied]\n"
24 " [--max-pack-size=<n>] [--local] [--incremental]\n"
25 " [--window=<n>] [--window-memory=<n>] [--depth=<n>]\n"
26 " [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset]\n"
27 " [--threads=<n>] [--non-empty] [--revs [--unpacked | --all]]\n"
28 " [--reflog] [--stdout | base-name] [--include-tag]\n"
29 " [--keep-unreachable | --unpack-unreachable]\n"
30 " [< ref-list | < object-list]";
31
32 struct object_entry {
33 struct pack_idx_entry idx;
34 unsigned long size; /* uncompressed size */
35 struct packed_git *in_pack; /* already in pack */
36 off_t in_pack_offset;
37 struct object_entry *delta; /* delta base object */
38 struct object_entry *delta_child; /* deltified objects who bases me */
39 struct object_entry *delta_sibling; /* other deltified objects who
40 * uses the same base as me
41 */
42 void *delta_data; /* cached delta (uncompressed) */
43 unsigned long delta_size; /* delta data size (uncompressed) */
44 unsigned long z_delta_size; /* delta data size (compressed) */
45 unsigned int hash; /* name hint hash */
46 enum object_type type;
47 enum object_type in_pack_type; /* could be delta */
48 unsigned char in_pack_header_size;
49 unsigned char preferred_base; /* we do not pack this, but is available
50 * to be used as the base object to delta
51 * objects against.
52 */
53 unsigned char no_try_delta;
54 };
55
56 /*
57 * Objects we are going to pack are collected in objects array (dynamically
58 * expanded). nr_objects & nr_alloc controls this array. They are stored
59 * in the order we see -- typically rev-list --objects order that gives us
60 * nice "minimum seek" order.
61 */
62 static struct object_entry *objects;
63 static struct pack_idx_entry **written_list;
64 static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
65
66 static int non_empty;
67 static int reuse_delta = 1, reuse_object = 1;
68 static int keep_unreachable, unpack_unreachable, include_tag;
69 static int local;
70 static int incremental;
71 static int ignore_packed_keep;
72 static int allow_ofs_delta;
73 static const char *base_name;
74 static int progress = 1;
75 static int window = 10;
76 static unsigned long pack_size_limit, pack_size_limit_cfg;
77 static int depth = 50;
78 static int delta_search_threads;
79 static int pack_to_stdout;
80 static int num_preferred_base;
81 static struct progress *progress_state;
82 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
83 static int pack_compression_seen;
84
85 static unsigned long delta_cache_size = 0;
86 static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
87 static unsigned long cache_max_small_delta_size = 1000;
88
89 static unsigned long window_memory_limit = 0;
90
91 /*
92 * The object names in objects array are hashed with this hashtable,
93 * to help looking up the entry by object name.
94 * This hashtable is built after all the objects are seen.
95 */
96 static int *object_ix;
97 static int object_ix_hashsz;
98
99 /*
100 * stats
101 */
102 static uint32_t written, written_delta;
103 static uint32_t reused, reused_delta;
104
105
106 static void *get_delta(struct object_entry *entry)
107 {
108 unsigned long size, base_size, delta_size;
109 void *buf, *base_buf, *delta_buf;
110 enum object_type type;
111
112 buf = read_sha1_file(entry->idx.sha1, &type, &size);
113 if (!buf)
114 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
115 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
116 if (!base_buf)
117 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
118 delta_buf = diff_delta(base_buf, base_size,
119 buf, size, &delta_size, 0);
120 if (!delta_buf || delta_size != entry->delta_size)
121 die("delta size changed");
122 free(buf);
123 free(base_buf);
124 return delta_buf;
125 }
126
127 static unsigned long do_compress(void **pptr, unsigned long size)
128 {
129 z_stream stream;
130 void *in, *out;
131 unsigned long maxsize;
132
133 memset(&stream, 0, sizeof(stream));
134 deflateInit(&stream, pack_compression_level);
135 maxsize = deflateBound(&stream, size);
136
137 in = *pptr;
138 out = xmalloc(maxsize);
139 *pptr = out;
140
141 stream.next_in = in;
142 stream.avail_in = size;
143 stream.next_out = out;
144 stream.avail_out = maxsize;
145 while (deflate(&stream, Z_FINISH) == Z_OK)
146 ; /* nothing */
147 deflateEnd(&stream);
148
149 free(in);
150 return stream.total_out;
151 }
152
153 /*
154 * we are going to reuse the existing object data as is. make
155 * sure it is not corrupt.
156 */
157 static int check_pack_inflate(struct packed_git *p,
158 struct pack_window **w_curs,
159 off_t offset,
160 off_t len,
161 unsigned long expect)
162 {
163 z_stream stream;
164 unsigned char fakebuf[4096], *in;
165 int st;
166
167 memset(&stream, 0, sizeof(stream));
168 git_inflate_init(&stream);
169 do {
170 in = use_pack(p, w_curs, offset, &stream.avail_in);
171 stream.next_in = in;
172 stream.next_out = fakebuf;
173 stream.avail_out = sizeof(fakebuf);
174 st = git_inflate(&stream, Z_FINISH);
175 offset += stream.next_in - in;
176 } while (st == Z_OK || st == Z_BUF_ERROR);
177 git_inflate_end(&stream);
178 return (st == Z_STREAM_END &&
179 stream.total_out == expect &&
180 stream.total_in == len) ? 0 : -1;
181 }
182
183 static void copy_pack_data(struct sha1file *f,
184 struct packed_git *p,
185 struct pack_window **w_curs,
186 off_t offset,
187 off_t len)
188 {
189 unsigned char *in;
190 unsigned int avail;
191
192 while (len) {
193 in = use_pack(p, w_curs, offset, &avail);
194 if (avail > len)
195 avail = (unsigned int)len;
196 sha1write(f, in, avail);
197 offset += avail;
198 len -= avail;
199 }
200 }
201
202 static unsigned long write_object(struct sha1file *f,
203 struct object_entry *entry,
204 off_t write_offset)
205 {
206 unsigned long size, limit, datalen;
207 void *buf;
208 unsigned char header[10], dheader[10];
209 unsigned hdrlen;
210 enum object_type type;
211 int usable_delta, to_reuse;
212
213 if (!pack_to_stdout)
214 crc32_begin(f);
215
216 type = entry->type;
217
218 /* apply size limit if limited packsize and not first object */
219 if (!pack_size_limit || !nr_written)
220 limit = 0;
221 else if (pack_size_limit <= write_offset)
222 /*
223 * the earlier object did not fit the limit; avoid
224 * mistaking this with unlimited (i.e. limit = 0).
225 */
226 limit = 1;
227 else
228 limit = pack_size_limit - write_offset;
229
230 if (!entry->delta)
231 usable_delta = 0; /* no delta */
232 else if (!pack_size_limit)
233 usable_delta = 1; /* unlimited packfile */
234 else if (entry->delta->idx.offset == (off_t)-1)
235 usable_delta = 0; /* base was written to another pack */
236 else if (entry->delta->idx.offset)
237 usable_delta = 1; /* base already exists in this pack */
238 else
239 usable_delta = 0; /* base could end up in another pack */
240
241 if (!reuse_object)
242 to_reuse = 0; /* explicit */
243 else if (!entry->in_pack)
244 to_reuse = 0; /* can't reuse what we don't have */
245 else if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA)
246 /* check_object() decided it for us ... */
247 to_reuse = usable_delta;
248 /* ... but pack split may override that */
249 else if (type != entry->in_pack_type)
250 to_reuse = 0; /* pack has delta which is unusable */
251 else if (entry->delta)
252 to_reuse = 0; /* we want to pack afresh */
253 else
254 to_reuse = 1; /* we have it in-pack undeltified,
255 * and we do not need to deltify it.
256 */
257
258 if (!to_reuse) {
259 no_reuse:
260 if (!usable_delta) {
261 buf = read_sha1_file(entry->idx.sha1, &type, &size);
262 if (!buf)
263 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
264 /*
265 * make sure no cached delta data remains from a
266 * previous attempt before a pack split occurred.
267 */
268 free(entry->delta_data);
269 entry->delta_data = NULL;
270 entry->z_delta_size = 0;
271 } else if (entry->delta_data) {
272 size = entry->delta_size;
273 buf = entry->delta_data;
274 entry->delta_data = NULL;
275 type = (allow_ofs_delta && entry->delta->idx.offset) ?
276 OBJ_OFS_DELTA : OBJ_REF_DELTA;
277 } else {
278 buf = get_delta(entry);
279 size = entry->delta_size;
280 type = (allow_ofs_delta && entry->delta->idx.offset) ?
281 OBJ_OFS_DELTA : OBJ_REF_DELTA;
282 }
283
284 if (entry->z_delta_size)
285 datalen = entry->z_delta_size;
286 else
287 datalen = do_compress(&buf, size);
288
289 /*
290 * The object header is a byte of 'type' followed by zero or
291 * more bytes of length.
292 */
293 hdrlen = encode_in_pack_object_header(type, size, header);
294
295 if (type == OBJ_OFS_DELTA) {
296 /*
297 * Deltas with relative base contain an additional
298 * encoding of the relative offset for the delta
299 * base from this object's position in the pack.
300 */
301 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
302 unsigned pos = sizeof(dheader) - 1;
303 dheader[pos] = ofs & 127;
304 while (ofs >>= 7)
305 dheader[--pos] = 128 | (--ofs & 127);
306 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
307 free(buf);
308 return 0;
309 }
310 sha1write(f, header, hdrlen);
311 sha1write(f, dheader + pos, sizeof(dheader) - pos);
312 hdrlen += sizeof(dheader) - pos;
313 } else if (type == OBJ_REF_DELTA) {
314 /*
315 * Deltas with a base reference contain
316 * an additional 20 bytes for the base sha1.
317 */
318 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
319 free(buf);
320 return 0;
321 }
322 sha1write(f, header, hdrlen);
323 sha1write(f, entry->delta->idx.sha1, 20);
324 hdrlen += 20;
325 } else {
326 if (limit && hdrlen + datalen + 20 >= limit) {
327 free(buf);
328 return 0;
329 }
330 sha1write(f, header, hdrlen);
331 }
332 sha1write(f, buf, datalen);
333 free(buf);
334 }
335 else {
336 struct packed_git *p = entry->in_pack;
337 struct pack_window *w_curs = NULL;
338 struct revindex_entry *revidx;
339 off_t offset;
340
341 if (entry->delta)
342 type = (allow_ofs_delta && entry->delta->idx.offset) ?
343 OBJ_OFS_DELTA : OBJ_REF_DELTA;
344 hdrlen = encode_in_pack_object_header(type, entry->size, header);
345
346 offset = entry->in_pack_offset;
347 revidx = find_pack_revindex(p, offset);
348 datalen = revidx[1].offset - offset;
349 if (!pack_to_stdout && p->index_version > 1 &&
350 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
351 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
352 unuse_pack(&w_curs);
353 goto no_reuse;
354 }
355
356 offset += entry->in_pack_header_size;
357 datalen -= entry->in_pack_header_size;
358 if (!pack_to_stdout && p->index_version == 1 &&
359 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
360 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
361 unuse_pack(&w_curs);
362 goto no_reuse;
363 }
364
365 if (type == OBJ_OFS_DELTA) {
366 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
367 unsigned pos = sizeof(dheader) - 1;
368 dheader[pos] = ofs & 127;
369 while (ofs >>= 7)
370 dheader[--pos] = 128 | (--ofs & 127);
371 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
372 unuse_pack(&w_curs);
373 return 0;
374 }
375 sha1write(f, header, hdrlen);
376 sha1write(f, dheader + pos, sizeof(dheader) - pos);
377 hdrlen += sizeof(dheader) - pos;
378 reused_delta++;
379 } else if (type == OBJ_REF_DELTA) {
380 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
381 unuse_pack(&w_curs);
382 return 0;
383 }
384 sha1write(f, header, hdrlen);
385 sha1write(f, entry->delta->idx.sha1, 20);
386 hdrlen += 20;
387 reused_delta++;
388 } else {
389 if (limit && hdrlen + datalen + 20 >= limit) {
390 unuse_pack(&w_curs);
391 return 0;
392 }
393 sha1write(f, header, hdrlen);
394 }
395 copy_pack_data(f, p, &w_curs, offset, datalen);
396 unuse_pack(&w_curs);
397 reused++;
398 }
399 if (usable_delta)
400 written_delta++;
401 written++;
402 if (!pack_to_stdout)
403 entry->idx.crc32 = crc32_end(f);
404 return hdrlen + datalen;
405 }
406
407 static int write_one(struct sha1file *f,
408 struct object_entry *e,
409 off_t *offset)
410 {
411 unsigned long size;
412
413 /* offset is non zero if object is written already. */
414 if (e->idx.offset || e->preferred_base)
415 return -1;
416
417 /* if we are deltified, write out base object first. */
418 if (e->delta && !write_one(f, e->delta, offset))
419 return 0;
420
421 e->idx.offset = *offset;
422 size = write_object(f, e, *offset);
423 if (!size) {
424 e->idx.offset = 0;
425 return 0;
426 }
427 written_list[nr_written++] = &e->idx;
428
429 /* make sure off_t is sufficiently large not to wrap */
430 if (signed_add_overflows(*offset, size))
431 die("pack too large for current definition of off_t");
432 *offset += size;
433 return 1;
434 }
435
436 static void write_pack_file(void)
437 {
438 uint32_t i = 0, j;
439 struct sha1file *f;
440 off_t offset;
441 struct pack_header hdr;
442 uint32_t nr_remaining = nr_result;
443 time_t last_mtime = 0;
444
445 if (progress > pack_to_stdout)
446 progress_state = start_progress("Writing objects", nr_result);
447 written_list = xmalloc(nr_objects * sizeof(*written_list));
448
449 do {
450 unsigned char sha1[20];
451 char *pack_tmp_name = NULL;
452
453 if (pack_to_stdout) {
454 f = sha1fd_throughput(1, "<stdout>", progress_state);
455 } else {
456 char tmpname[PATH_MAX];
457 int fd;
458 fd = odb_mkstemp(tmpname, sizeof(tmpname),
459 "pack/tmp_pack_XXXXXX");
460 pack_tmp_name = xstrdup(tmpname);
461 f = sha1fd(fd, pack_tmp_name);
462 }
463
464 hdr.hdr_signature = htonl(PACK_SIGNATURE);
465 hdr.hdr_version = htonl(PACK_VERSION);
466 hdr.hdr_entries = htonl(nr_remaining);
467 sha1write(f, &hdr, sizeof(hdr));
468 offset = sizeof(hdr);
469 nr_written = 0;
470 for (; i < nr_objects; i++) {
471 if (!write_one(f, objects + i, &offset))
472 break;
473 display_progress(progress_state, written);
474 }
475
476 /*
477 * Did we write the wrong # entries in the header?
478 * If so, rewrite it like in fast-import
479 */
480 if (pack_to_stdout) {
481 sha1close(f, sha1, CSUM_CLOSE);
482 } else if (nr_written == nr_remaining) {
483 sha1close(f, sha1, CSUM_FSYNC);
484 } else {
485 int fd = sha1close(f, sha1, 0);
486 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
487 nr_written, sha1, offset);
488 close(fd);
489 }
490
491 if (!pack_to_stdout) {
492 struct stat st;
493 const char *idx_tmp_name;
494 char tmpname[PATH_MAX];
495
496 idx_tmp_name = write_idx_file(NULL, written_list,
497 nr_written, sha1);
498
499 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
500 base_name, sha1_to_hex(sha1));
501 free_pack_by_name(tmpname);
502 if (adjust_shared_perm(pack_tmp_name))
503 die_errno("unable to make temporary pack file readable");
504 if (rename(pack_tmp_name, tmpname))
505 die_errno("unable to rename temporary pack file");
506
507 /*
508 * Packs are runtime accessed in their mtime
509 * order since newer packs are more likely to contain
510 * younger objects. So if we are creating multiple
511 * packs then we should modify the mtime of later ones
512 * to preserve this property.
513 */
514 if (stat(tmpname, &st) < 0) {
515 warning("failed to stat %s: %s",
516 tmpname, strerror(errno));
517 } else if (!last_mtime) {
518 last_mtime = st.st_mtime;
519 } else {
520 struct utimbuf utb;
521 utb.actime = st.st_atime;
522 utb.modtime = --last_mtime;
523 if (utime(tmpname, &utb) < 0)
524 warning("failed utime() on %s: %s",
525 tmpname, strerror(errno));
526 }
527
528 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
529 base_name, sha1_to_hex(sha1));
530 if (adjust_shared_perm(idx_tmp_name))
531 die_errno("unable to make temporary index file readable");
532 if (rename(idx_tmp_name, tmpname))
533 die_errno("unable to rename temporary index file");
534
535 free((void *) idx_tmp_name);
536 free(pack_tmp_name);
537 puts(sha1_to_hex(sha1));
538 }
539
540 /* mark written objects as written to previous pack */
541 for (j = 0; j < nr_written; j++) {
542 written_list[j]->offset = (off_t)-1;
543 }
544 nr_remaining -= nr_written;
545 } while (nr_remaining && i < nr_objects);
546
547 free(written_list);
548 stop_progress(&progress_state);
549 if (written != nr_result)
550 die("wrote %"PRIu32" objects while expecting %"PRIu32,
551 written, nr_result);
552 }
553
554 static int locate_object_entry_hash(const unsigned char *sha1)
555 {
556 int i;
557 unsigned int ui;
558 memcpy(&ui, sha1, sizeof(unsigned int));
559 i = ui % object_ix_hashsz;
560 while (0 < object_ix[i]) {
561 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
562 return i;
563 if (++i == object_ix_hashsz)
564 i = 0;
565 }
566 return -1 - i;
567 }
568
569 static struct object_entry *locate_object_entry(const unsigned char *sha1)
570 {
571 int i;
572
573 if (!object_ix_hashsz)
574 return NULL;
575
576 i = locate_object_entry_hash(sha1);
577 if (0 <= i)
578 return &objects[object_ix[i]-1];
579 return NULL;
580 }
581
582 static void rehash_objects(void)
583 {
584 uint32_t i;
585 struct object_entry *oe;
586
587 object_ix_hashsz = nr_objects * 3;
588 if (object_ix_hashsz < 1024)
589 object_ix_hashsz = 1024;
590 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
591 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
592 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
593 int ix = locate_object_entry_hash(oe->idx.sha1);
594 if (0 <= ix)
595 continue;
596 ix = -1 - ix;
597 object_ix[ix] = i + 1;
598 }
599 }
600
601 static unsigned name_hash(const char *name)
602 {
603 unsigned c, hash = 0;
604
605 if (!name)
606 return 0;
607
608 /*
609 * This effectively just creates a sortable number from the
610 * last sixteen non-whitespace characters. Last characters
611 * count "most", so things that end in ".c" sort together.
612 */
613 while ((c = *name++) != 0) {
614 if (isspace(c))
615 continue;
616 hash = (hash >> 2) + (c << 24);
617 }
618 return hash;
619 }
620
621 static void setup_delta_attr_check(struct git_attr_check *check)
622 {
623 static struct git_attr *attr_delta;
624
625 if (!attr_delta)
626 attr_delta = git_attr("delta");
627
628 check[0].attr = attr_delta;
629 }
630
631 static int no_try_delta(const char *path)
632 {
633 struct git_attr_check check[1];
634
635 setup_delta_attr_check(check);
636 if (git_checkattr(path, ARRAY_SIZE(check), check))
637 return 0;
638 if (ATTR_FALSE(check->value))
639 return 1;
640 return 0;
641 }
642
643 static int add_object_entry(const unsigned char *sha1, enum object_type type,
644 const char *name, int exclude)
645 {
646 struct object_entry *entry;
647 struct packed_git *p, *found_pack = NULL;
648 off_t found_offset = 0;
649 int ix;
650 unsigned hash = name_hash(name);
651
652 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
653 if (ix >= 0) {
654 if (exclude) {
655 entry = objects + object_ix[ix] - 1;
656 if (!entry->preferred_base)
657 nr_result--;
658 entry->preferred_base = 1;
659 }
660 return 0;
661 }
662
663 if (!exclude && local && has_loose_object_nonlocal(sha1))
664 return 0;
665
666 for (p = packed_git; p; p = p->next) {
667 off_t offset = find_pack_entry_one(sha1, p);
668 if (offset) {
669 if (!found_pack) {
670 found_offset = offset;
671 found_pack = p;
672 }
673 if (exclude)
674 break;
675 if (incremental)
676 return 0;
677 if (local && !p->pack_local)
678 return 0;
679 if (ignore_packed_keep && p->pack_local && p->pack_keep)
680 return 0;
681 }
682 }
683
684 if (nr_objects >= nr_alloc) {
685 nr_alloc = (nr_alloc + 1024) * 3 / 2;
686 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
687 }
688
689 entry = objects + nr_objects++;
690 memset(entry, 0, sizeof(*entry));
691 hashcpy(entry->idx.sha1, sha1);
692 entry->hash = hash;
693 if (type)
694 entry->type = type;
695 if (exclude)
696 entry->preferred_base = 1;
697 else
698 nr_result++;
699 if (found_pack) {
700 entry->in_pack = found_pack;
701 entry->in_pack_offset = found_offset;
702 }
703
704 if (object_ix_hashsz * 3 <= nr_objects * 4)
705 rehash_objects();
706 else
707 object_ix[-1 - ix] = nr_objects;
708
709 display_progress(progress_state, nr_objects);
710
711 if (name && no_try_delta(name))
712 entry->no_try_delta = 1;
713
714 return 1;
715 }
716
717 struct pbase_tree_cache {
718 unsigned char sha1[20];
719 int ref;
720 int temporary;
721 void *tree_data;
722 unsigned long tree_size;
723 };
724
725 static struct pbase_tree_cache *(pbase_tree_cache[256]);
726 static int pbase_tree_cache_ix(const unsigned char *sha1)
727 {
728 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
729 }
730 static int pbase_tree_cache_ix_incr(int ix)
731 {
732 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
733 }
734
735 static struct pbase_tree {
736 struct pbase_tree *next;
737 /* This is a phony "cache" entry; we are not
738 * going to evict it nor find it through _get()
739 * mechanism -- this is for the toplevel node that
740 * would almost always change with any commit.
741 */
742 struct pbase_tree_cache pcache;
743 } *pbase_tree;
744
745 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
746 {
747 struct pbase_tree_cache *ent, *nent;
748 void *data;
749 unsigned long size;
750 enum object_type type;
751 int neigh;
752 int my_ix = pbase_tree_cache_ix(sha1);
753 int available_ix = -1;
754
755 /* pbase-tree-cache acts as a limited hashtable.
756 * your object will be found at your index or within a few
757 * slots after that slot if it is cached.
758 */
759 for (neigh = 0; neigh < 8; neigh++) {
760 ent = pbase_tree_cache[my_ix];
761 if (ent && !hashcmp(ent->sha1, sha1)) {
762 ent->ref++;
763 return ent;
764 }
765 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
766 ((0 <= available_ix) &&
767 (!ent && pbase_tree_cache[available_ix])))
768 available_ix = my_ix;
769 if (!ent)
770 break;
771 my_ix = pbase_tree_cache_ix_incr(my_ix);
772 }
773
774 /* Did not find one. Either we got a bogus request or
775 * we need to read and perhaps cache.
776 */
777 data = read_sha1_file(sha1, &type, &size);
778 if (!data)
779 return NULL;
780 if (type != OBJ_TREE) {
781 free(data);
782 return NULL;
783 }
784
785 /* We need to either cache or return a throwaway copy */
786
787 if (available_ix < 0)
788 ent = NULL;
789 else {
790 ent = pbase_tree_cache[available_ix];
791 my_ix = available_ix;
792 }
793
794 if (!ent) {
795 nent = xmalloc(sizeof(*nent));
796 nent->temporary = (available_ix < 0);
797 }
798 else {
799 /* evict and reuse */
800 free(ent->tree_data);
801 nent = ent;
802 }
803 hashcpy(nent->sha1, sha1);
804 nent->tree_data = data;
805 nent->tree_size = size;
806 nent->ref = 1;
807 if (!nent->temporary)
808 pbase_tree_cache[my_ix] = nent;
809 return nent;
810 }
811
812 static void pbase_tree_put(struct pbase_tree_cache *cache)
813 {
814 if (!cache->temporary) {
815 cache->ref--;
816 return;
817 }
818 free(cache->tree_data);
819 free(cache);
820 }
821
822 static int name_cmp_len(const char *name)
823 {
824 int i;
825 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
826 ;
827 return i;
828 }
829
830 static void add_pbase_object(struct tree_desc *tree,
831 const char *name,
832 int cmplen,
833 const char *fullname)
834 {
835 struct name_entry entry;
836 int cmp;
837
838 while (tree_entry(tree,&entry)) {
839 if (S_ISGITLINK(entry.mode))
840 continue;
841 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
842 memcmp(name, entry.path, cmplen);
843 if (cmp > 0)
844 continue;
845 if (cmp < 0)
846 return;
847 if (name[cmplen] != '/') {
848 add_object_entry(entry.sha1,
849 object_type(entry.mode),
850 fullname, 1);
851 return;
852 }
853 if (S_ISDIR(entry.mode)) {
854 struct tree_desc sub;
855 struct pbase_tree_cache *tree;
856 const char *down = name+cmplen+1;
857 int downlen = name_cmp_len(down);
858
859 tree = pbase_tree_get(entry.sha1);
860 if (!tree)
861 return;
862 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
863
864 add_pbase_object(&sub, down, downlen, fullname);
865 pbase_tree_put(tree);
866 }
867 }
868 }
869
870 static unsigned *done_pbase_paths;
871 static int done_pbase_paths_num;
872 static int done_pbase_paths_alloc;
873 static int done_pbase_path_pos(unsigned hash)
874 {
875 int lo = 0;
876 int hi = done_pbase_paths_num;
877 while (lo < hi) {
878 int mi = (hi + lo) / 2;
879 if (done_pbase_paths[mi] == hash)
880 return mi;
881 if (done_pbase_paths[mi] < hash)
882 hi = mi;
883 else
884 lo = mi + 1;
885 }
886 return -lo-1;
887 }
888
889 static int check_pbase_path(unsigned hash)
890 {
891 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
892 if (0 <= pos)
893 return 1;
894 pos = -pos - 1;
895 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
896 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
897 done_pbase_paths = xrealloc(done_pbase_paths,
898 done_pbase_paths_alloc *
899 sizeof(unsigned));
900 }
901 done_pbase_paths_num++;
902 if (pos < done_pbase_paths_num)
903 memmove(done_pbase_paths + pos + 1,
904 done_pbase_paths + pos,
905 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
906 done_pbase_paths[pos] = hash;
907 return 0;
908 }
909
910 static void add_preferred_base_object(const char *name)
911 {
912 struct pbase_tree *it;
913 int cmplen;
914 unsigned hash = name_hash(name);
915
916 if (!num_preferred_base || check_pbase_path(hash))
917 return;
918
919 cmplen = name_cmp_len(name);
920 for (it = pbase_tree; it; it = it->next) {
921 if (cmplen == 0) {
922 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
923 }
924 else {
925 struct tree_desc tree;
926 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
927 add_pbase_object(&tree, name, cmplen, name);
928 }
929 }
930 }
931
932 static void add_preferred_base(unsigned char *sha1)
933 {
934 struct pbase_tree *it;
935 void *data;
936 unsigned long size;
937 unsigned char tree_sha1[20];
938
939 if (window <= num_preferred_base++)
940 return;
941
942 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
943 if (!data)
944 return;
945
946 for (it = pbase_tree; it; it = it->next) {
947 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
948 free(data);
949 return;
950 }
951 }
952
953 it = xcalloc(1, sizeof(*it));
954 it->next = pbase_tree;
955 pbase_tree = it;
956
957 hashcpy(it->pcache.sha1, tree_sha1);
958 it->pcache.tree_data = data;
959 it->pcache.tree_size = size;
960 }
961
962 static void cleanup_preferred_base(void)
963 {
964 struct pbase_tree *it;
965 unsigned i;
966
967 it = pbase_tree;
968 pbase_tree = NULL;
969 while (it) {
970 struct pbase_tree *this = it;
971 it = this->next;
972 free(this->pcache.tree_data);
973 free(this);
974 }
975
976 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
977 if (!pbase_tree_cache[i])
978 continue;
979 free(pbase_tree_cache[i]->tree_data);
980 free(pbase_tree_cache[i]);
981 pbase_tree_cache[i] = NULL;
982 }
983
984 free(done_pbase_paths);
985 done_pbase_paths = NULL;
986 done_pbase_paths_num = done_pbase_paths_alloc = 0;
987 }
988
989 static void check_object(struct object_entry *entry)
990 {
991 if (entry->in_pack) {
992 struct packed_git *p = entry->in_pack;
993 struct pack_window *w_curs = NULL;
994 const unsigned char *base_ref = NULL;
995 struct object_entry *base_entry;
996 unsigned long used, used_0;
997 unsigned int avail;
998 off_t ofs;
999 unsigned char *buf, c;
1000
1001 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1002
1003 /*
1004 * We want in_pack_type even if we do not reuse delta
1005 * since non-delta representations could still be reused.
1006 */
1007 used = unpack_object_header_buffer(buf, avail,
1008 &entry->in_pack_type,
1009 &entry->size);
1010 if (used == 0)
1011 goto give_up;
1012
1013 /*
1014 * Determine if this is a delta and if so whether we can
1015 * reuse it or not. Otherwise let's find out as cheaply as
1016 * possible what the actual type and size for this object is.
1017 */
1018 switch (entry->in_pack_type) {
1019 default:
1020 /* Not a delta hence we've already got all we need. */
1021 entry->type = entry->in_pack_type;
1022 entry->in_pack_header_size = used;
1023 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1024 goto give_up;
1025 unuse_pack(&w_curs);
1026 return;
1027 case OBJ_REF_DELTA:
1028 if (reuse_delta && !entry->preferred_base)
1029 base_ref = use_pack(p, &w_curs,
1030 entry->in_pack_offset + used, NULL);
1031 entry->in_pack_header_size = used + 20;
1032 break;
1033 case OBJ_OFS_DELTA:
1034 buf = use_pack(p, &w_curs,
1035 entry->in_pack_offset + used, NULL);
1036 used_0 = 0;
1037 c = buf[used_0++];
1038 ofs = c & 127;
1039 while (c & 128) {
1040 ofs += 1;
1041 if (!ofs || MSB(ofs, 7)) {
1042 error("delta base offset overflow in pack for %s",
1043 sha1_to_hex(entry->idx.sha1));
1044 goto give_up;
1045 }
1046 c = buf[used_0++];
1047 ofs = (ofs << 7) + (c & 127);
1048 }
1049 ofs = entry->in_pack_offset - ofs;
1050 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1051 error("delta base offset out of bound for %s",
1052 sha1_to_hex(entry->idx.sha1));
1053 goto give_up;
1054 }
1055 if (reuse_delta && !entry->preferred_base) {
1056 struct revindex_entry *revidx;
1057 revidx = find_pack_revindex(p, ofs);
1058 if (!revidx)
1059 goto give_up;
1060 base_ref = nth_packed_object_sha1(p, revidx->nr);
1061 }
1062 entry->in_pack_header_size = used + used_0;
1063 break;
1064 }
1065
1066 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1067 /*
1068 * If base_ref was set above that means we wish to
1069 * reuse delta data, and we even found that base
1070 * in the list of objects we want to pack. Goodie!
1071 *
1072 * Depth value does not matter - find_deltas() will
1073 * never consider reused delta as the base object to
1074 * deltify other objects against, in order to avoid
1075 * circular deltas.
1076 */
1077 entry->type = entry->in_pack_type;
1078 entry->delta = base_entry;
1079 entry->delta_size = entry->size;
1080 entry->delta_sibling = base_entry->delta_child;
1081 base_entry->delta_child = entry;
1082 unuse_pack(&w_curs);
1083 return;
1084 }
1085
1086 if (entry->type) {
1087 /*
1088 * This must be a delta and we already know what the
1089 * final object type is. Let's extract the actual
1090 * object size from the delta header.
1091 */
1092 entry->size = get_size_from_delta(p, &w_curs,
1093 entry->in_pack_offset + entry->in_pack_header_size);
1094 if (entry->size == 0)
1095 goto give_up;
1096 unuse_pack(&w_curs);
1097 return;
1098 }
1099
1100 /*
1101 * No choice but to fall back to the recursive delta walk
1102 * with sha1_object_info() to find about the object type
1103 * at this point...
1104 */
1105 give_up:
1106 unuse_pack(&w_curs);
1107 }
1108
1109 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1110 /*
1111 * The error condition is checked in prepare_pack(). This is
1112 * to permit a missing preferred base object to be ignored
1113 * as a preferred base. Doing so can result in a larger
1114 * pack file, but the transfer will still take place.
1115 */
1116 }
1117
1118 static int pack_offset_sort(const void *_a, const void *_b)
1119 {
1120 const struct object_entry *a = *(struct object_entry **)_a;
1121 const struct object_entry *b = *(struct object_entry **)_b;
1122
1123 /* avoid filesystem trashing with loose objects */
1124 if (!a->in_pack && !b->in_pack)
1125 return hashcmp(a->idx.sha1, b->idx.sha1);
1126
1127 if (a->in_pack < b->in_pack)
1128 return -1;
1129 if (a->in_pack > b->in_pack)
1130 return 1;
1131 return a->in_pack_offset < b->in_pack_offset ? -1 :
1132 (a->in_pack_offset > b->in_pack_offset);
1133 }
1134
1135 static void get_object_details(void)
1136 {
1137 uint32_t i;
1138 struct object_entry **sorted_by_offset;
1139
1140 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1141 for (i = 0; i < nr_objects; i++)
1142 sorted_by_offset[i] = objects + i;
1143 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1144
1145 for (i = 0; i < nr_objects; i++)
1146 check_object(sorted_by_offset[i]);
1147
1148 free(sorted_by_offset);
1149 }
1150
1151 /*
1152 * We search for deltas in a list sorted by type, by filename hash, and then
1153 * by size, so that we see progressively smaller and smaller files.
1154 * That's because we prefer deltas to be from the bigger file
1155 * to the smaller -- deletes are potentially cheaper, but perhaps
1156 * more importantly, the bigger file is likely the more recent
1157 * one. The deepest deltas are therefore the oldest objects which are
1158 * less susceptible to be accessed often.
1159 */
1160 static int type_size_sort(const void *_a, const void *_b)
1161 {
1162 const struct object_entry *a = *(struct object_entry **)_a;
1163 const struct object_entry *b = *(struct object_entry **)_b;
1164
1165 if (a->type > b->type)
1166 return -1;
1167 if (a->type < b->type)
1168 return 1;
1169 if (a->hash > b->hash)
1170 return -1;
1171 if (a->hash < b->hash)
1172 return 1;
1173 if (a->preferred_base > b->preferred_base)
1174 return -1;
1175 if (a->preferred_base < b->preferred_base)
1176 return 1;
1177 if (a->size > b->size)
1178 return -1;
1179 if (a->size < b->size)
1180 return 1;
1181 return a < b ? -1 : (a > b); /* newest first */
1182 }
1183
1184 struct unpacked {
1185 struct object_entry *entry;
1186 void *data;
1187 struct delta_index *index;
1188 unsigned depth;
1189 };
1190
1191 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1192 unsigned long delta_size)
1193 {
1194 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1195 return 0;
1196
1197 if (delta_size < cache_max_small_delta_size)
1198 return 1;
1199
1200 /* cache delta, if objects are large enough compared to delta size */
1201 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1202 return 1;
1203
1204 return 0;
1205 }
1206
1207 #ifndef NO_PTHREADS
1208
1209 static pthread_mutex_t read_mutex;
1210 #define read_lock() pthread_mutex_lock(&read_mutex)
1211 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1212
1213 static pthread_mutex_t cache_mutex;
1214 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1215 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1216
1217 static pthread_mutex_t progress_mutex;
1218 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1219 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1220
1221 #else
1222
1223 #define read_lock() (void)0
1224 #define read_unlock() (void)0
1225 #define cache_lock() (void)0
1226 #define cache_unlock() (void)0
1227 #define progress_lock() (void)0
1228 #define progress_unlock() (void)0
1229
1230 #endif
1231
1232 static int try_delta(struct unpacked *trg, struct unpacked *src,
1233 unsigned max_depth, unsigned long *mem_usage)
1234 {
1235 struct object_entry *trg_entry = trg->entry;
1236 struct object_entry *src_entry = src->entry;
1237 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1238 unsigned ref_depth;
1239 enum object_type type;
1240 void *delta_buf;
1241
1242 /* Don't bother doing diffs between different types */
1243 if (trg_entry->type != src_entry->type)
1244 return -1;
1245
1246 /*
1247 * We do not bother to try a delta that we discarded
1248 * on an earlier try, but only when reusing delta data.
1249 */
1250 if (reuse_delta && trg_entry->in_pack &&
1251 trg_entry->in_pack == src_entry->in_pack &&
1252 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1253 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1254 return 0;
1255
1256 /* Let's not bust the allowed depth. */
1257 if (src->depth >= max_depth)
1258 return 0;
1259
1260 /* Now some size filtering heuristics. */
1261 trg_size = trg_entry->size;
1262 if (!trg_entry->delta) {
1263 max_size = trg_size/2 - 20;
1264 ref_depth = 1;
1265 } else {
1266 max_size = trg_entry->delta_size;
1267 ref_depth = trg->depth;
1268 }
1269 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1270 (max_depth - ref_depth + 1);
1271 if (max_size == 0)
1272 return 0;
1273 src_size = src_entry->size;
1274 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1275 if (sizediff >= max_size)
1276 return 0;
1277 if (trg_size < src_size / 32)
1278 return 0;
1279
1280 /* Load data if not already done */
1281 if (!trg->data) {
1282 read_lock();
1283 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1284 read_unlock();
1285 if (!trg->data)
1286 die("object %s cannot be read",
1287 sha1_to_hex(trg_entry->idx.sha1));
1288 if (sz != trg_size)
1289 die("object %s inconsistent object length (%lu vs %lu)",
1290 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1291 *mem_usage += sz;
1292 }
1293 if (!src->data) {
1294 read_lock();
1295 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1296 read_unlock();
1297 if (!src->data) {
1298 if (src_entry->preferred_base) {
1299 static int warned = 0;
1300 if (!warned++)
1301 warning("object %s cannot be read",
1302 sha1_to_hex(src_entry->idx.sha1));
1303 /*
1304 * Those objects are not included in the
1305 * resulting pack. Be resilient and ignore
1306 * them if they can't be read, in case the
1307 * pack could be created nevertheless.
1308 */
1309 return 0;
1310 }
1311 die("object %s cannot be read",
1312 sha1_to_hex(src_entry->idx.sha1));
1313 }
1314 if (sz != src_size)
1315 die("object %s inconsistent object length (%lu vs %lu)",
1316 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1317 *mem_usage += sz;
1318 }
1319 if (!src->index) {
1320 src->index = create_delta_index(src->data, src_size);
1321 if (!src->index) {
1322 static int warned = 0;
1323 if (!warned++)
1324 warning("suboptimal pack - out of memory");
1325 return 0;
1326 }
1327 *mem_usage += sizeof_delta_index(src->index);
1328 }
1329
1330 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1331 if (!delta_buf)
1332 return 0;
1333
1334 if (trg_entry->delta) {
1335 /* Prefer only shallower same-sized deltas. */
1336 if (delta_size == trg_entry->delta_size &&
1337 src->depth + 1 >= trg->depth) {
1338 free(delta_buf);
1339 return 0;
1340 }
1341 }
1342
1343 /*
1344 * Handle memory allocation outside of the cache
1345 * accounting lock. Compiler will optimize the strangeness
1346 * away when NO_PTHREADS is defined.
1347 */
1348 free(trg_entry->delta_data);
1349 cache_lock();
1350 if (trg_entry->delta_data) {
1351 delta_cache_size -= trg_entry->delta_size;
1352 trg_entry->delta_data = NULL;
1353 }
1354 if (delta_cacheable(src_size, trg_size, delta_size)) {
1355 delta_cache_size += delta_size;
1356 cache_unlock();
1357 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1358 } else {
1359 cache_unlock();
1360 free(delta_buf);
1361 }
1362
1363 trg_entry->delta = src_entry;
1364 trg_entry->delta_size = delta_size;
1365 trg->depth = src->depth + 1;
1366
1367 return 1;
1368 }
1369
1370 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1371 {
1372 struct object_entry *child = me->delta_child;
1373 unsigned int m = n;
1374 while (child) {
1375 unsigned int c = check_delta_limit(child, n + 1);
1376 if (m < c)
1377 m = c;
1378 child = child->delta_sibling;
1379 }
1380 return m;
1381 }
1382
1383 static unsigned long free_unpacked(struct unpacked *n)
1384 {
1385 unsigned long freed_mem = sizeof_delta_index(n->index);
1386 free_delta_index(n->index);
1387 n->index = NULL;
1388 if (n->data) {
1389 freed_mem += n->entry->size;
1390 free(n->data);
1391 n->data = NULL;
1392 }
1393 n->entry = NULL;
1394 n->depth = 0;
1395 return freed_mem;
1396 }
1397
1398 static void find_deltas(struct object_entry **list, unsigned *list_size,
1399 int window, int depth, unsigned *processed)
1400 {
1401 uint32_t i, idx = 0, count = 0;
1402 struct unpacked *array;
1403 unsigned long mem_usage = 0;
1404
1405 array = xcalloc(window, sizeof(struct unpacked));
1406
1407 for (;;) {
1408 struct object_entry *entry;
1409 struct unpacked *n = array + idx;
1410 int j, max_depth, best_base = -1;
1411
1412 progress_lock();
1413 if (!*list_size) {
1414 progress_unlock();
1415 break;
1416 }
1417 entry = *list++;
1418 (*list_size)--;
1419 if (!entry->preferred_base) {
1420 (*processed)++;
1421 display_progress(progress_state, *processed);
1422 }
1423 progress_unlock();
1424
1425 mem_usage -= free_unpacked(n);
1426 n->entry = entry;
1427
1428 while (window_memory_limit &&
1429 mem_usage > window_memory_limit &&
1430 count > 1) {
1431 uint32_t tail = (idx + window - count) % window;
1432 mem_usage -= free_unpacked(array + tail);
1433 count--;
1434 }
1435
1436 /* We do not compute delta to *create* objects we are not
1437 * going to pack.
1438 */
1439 if (entry->preferred_base)
1440 goto next;
1441
1442 /*
1443 * If the current object is at pack edge, take the depth the
1444 * objects that depend on the current object into account
1445 * otherwise they would become too deep.
1446 */
1447 max_depth = depth;
1448 if (entry->delta_child) {
1449 max_depth -= check_delta_limit(entry, 0);
1450 if (max_depth <= 0)
1451 goto next;
1452 }
1453
1454 j = window;
1455 while (--j > 0) {
1456 int ret;
1457 uint32_t other_idx = idx + j;
1458 struct unpacked *m;
1459 if (other_idx >= window)
1460 other_idx -= window;
1461 m = array + other_idx;
1462 if (!m->entry)
1463 break;
1464 ret = try_delta(n, m, max_depth, &mem_usage);
1465 if (ret < 0)
1466 break;
1467 else if (ret > 0)
1468 best_base = other_idx;
1469 }
1470
1471 /*
1472 * If we decided to cache the delta data, then it is best
1473 * to compress it right away. First because we have to do
1474 * it anyway, and doing it here while we're threaded will
1475 * save a lot of time in the non threaded write phase,
1476 * as well as allow for caching more deltas within
1477 * the same cache size limit.
1478 * ...
1479 * But only if not writing to stdout, since in that case
1480 * the network is most likely throttling writes anyway,
1481 * and therefore it is best to go to the write phase ASAP
1482 * instead, as we can afford spending more time compressing
1483 * between writes at that moment.
1484 */
1485 if (entry->delta_data && !pack_to_stdout) {
1486 entry->z_delta_size = do_compress(&entry->delta_data,
1487 entry->delta_size);
1488 cache_lock();
1489 delta_cache_size -= entry->delta_size;
1490 delta_cache_size += entry->z_delta_size;
1491 cache_unlock();
1492 }
1493
1494 /* if we made n a delta, and if n is already at max
1495 * depth, leaving it in the window is pointless. we
1496 * should evict it first.
1497 */
1498 if (entry->delta && max_depth <= n->depth)
1499 continue;
1500
1501 /*
1502 * Move the best delta base up in the window, after the
1503 * currently deltified object, to keep it longer. It will
1504 * be the first base object to be attempted next.
1505 */
1506 if (entry->delta) {
1507 struct unpacked swap = array[best_base];
1508 int dist = (window + idx - best_base) % window;
1509 int dst = best_base;
1510 while (dist--) {
1511 int src = (dst + 1) % window;
1512 array[dst] = array[src];
1513 dst = src;
1514 }
1515 array[dst] = swap;
1516 }
1517
1518 next:
1519 idx++;
1520 if (count + 1 < window)
1521 count++;
1522 if (idx >= window)
1523 idx = 0;
1524 }
1525
1526 for (i = 0; i < window; ++i) {
1527 free_delta_index(array[i].index);
1528 free(array[i].data);
1529 }
1530 free(array);
1531 }
1532
1533 #ifndef NO_PTHREADS
1534
1535 static void try_to_free_from_threads(size_t size)
1536 {
1537 read_lock();
1538 release_pack_memory(size, -1);
1539 read_unlock();
1540 }
1541
1542 static try_to_free_t old_try_to_free_routine;
1543
1544 /*
1545 * The main thread waits on the condition that (at least) one of the workers
1546 * has stopped working (which is indicated in the .working member of
1547 * struct thread_params).
1548 * When a work thread has completed its work, it sets .working to 0 and
1549 * signals the main thread and waits on the condition that .data_ready
1550 * becomes 1.
1551 */
1552
1553 struct thread_params {
1554 pthread_t thread;
1555 struct object_entry **list;
1556 unsigned list_size;
1557 unsigned remaining;
1558 int window;
1559 int depth;
1560 int working;
1561 int data_ready;
1562 pthread_mutex_t mutex;
1563 pthread_cond_t cond;
1564 unsigned *processed;
1565 };
1566
1567 static pthread_cond_t progress_cond;
1568
1569 /*
1570 * Mutex and conditional variable can't be statically-initialized on Windows.
1571 */
1572 static void init_threaded_search(void)
1573 {
1574 init_recursive_mutex(&read_mutex);
1575 pthread_mutex_init(&cache_mutex, NULL);
1576 pthread_mutex_init(&progress_mutex, NULL);
1577 pthread_cond_init(&progress_cond, NULL);
1578 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
1579 }
1580
1581 static void cleanup_threaded_search(void)
1582 {
1583 set_try_to_free_routine(old_try_to_free_routine);
1584 pthread_cond_destroy(&progress_cond);
1585 pthread_mutex_destroy(&read_mutex);
1586 pthread_mutex_destroy(&cache_mutex);
1587 pthread_mutex_destroy(&progress_mutex);
1588 }
1589
1590 static void *threaded_find_deltas(void *arg)
1591 {
1592 struct thread_params *me = arg;
1593
1594 while (me->remaining) {
1595 find_deltas(me->list, &me->remaining,
1596 me->window, me->depth, me->processed);
1597
1598 progress_lock();
1599 me->working = 0;
1600 pthread_cond_signal(&progress_cond);
1601 progress_unlock();
1602
1603 /*
1604 * We must not set ->data_ready before we wait on the
1605 * condition because the main thread may have set it to 1
1606 * before we get here. In order to be sure that new
1607 * work is available if we see 1 in ->data_ready, it
1608 * was initialized to 0 before this thread was spawned
1609 * and we reset it to 0 right away.
1610 */
1611 pthread_mutex_lock(&me->mutex);
1612 while (!me->data_ready)
1613 pthread_cond_wait(&me->cond, &me->mutex);
1614 me->data_ready = 0;
1615 pthread_mutex_unlock(&me->mutex);
1616 }
1617 /* leave ->working 1 so that this doesn't get more work assigned */
1618 return NULL;
1619 }
1620
1621 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1622 int window, int depth, unsigned *processed)
1623 {
1624 struct thread_params *p;
1625 int i, ret, active_threads = 0;
1626
1627 init_threaded_search();
1628
1629 if (!delta_search_threads) /* --threads=0 means autodetect */
1630 delta_search_threads = online_cpus();
1631 if (delta_search_threads <= 1) {
1632 find_deltas(list, &list_size, window, depth, processed);
1633 cleanup_threaded_search();
1634 return;
1635 }
1636 if (progress > pack_to_stdout)
1637 fprintf(stderr, "Delta compression using up to %d threads.\n",
1638 delta_search_threads);
1639 p = xcalloc(delta_search_threads, sizeof(*p));
1640
1641 /* Partition the work amongst work threads. */
1642 for (i = 0; i < delta_search_threads; i++) {
1643 unsigned sub_size = list_size / (delta_search_threads - i);
1644
1645 /* don't use too small segments or no deltas will be found */
1646 if (sub_size < 2*window && i+1 < delta_search_threads)
1647 sub_size = 0;
1648
1649 p[i].window = window;
1650 p[i].depth = depth;
1651 p[i].processed = processed;
1652 p[i].working = 1;
1653 p[i].data_ready = 0;
1654
1655 /* try to split chunks on "path" boundaries */
1656 while (sub_size && sub_size < list_size &&
1657 list[sub_size]->hash &&
1658 list[sub_size]->hash == list[sub_size-1]->hash)
1659 sub_size++;
1660
1661 p[i].list = list;
1662 p[i].list_size = sub_size;
1663 p[i].remaining = sub_size;
1664
1665 list += sub_size;
1666 list_size -= sub_size;
1667 }
1668
1669 /* Start work threads. */
1670 for (i = 0; i < delta_search_threads; i++) {
1671 if (!p[i].list_size)
1672 continue;
1673 pthread_mutex_init(&p[i].mutex, NULL);
1674 pthread_cond_init(&p[i].cond, NULL);
1675 ret = pthread_create(&p[i].thread, NULL,
1676 threaded_find_deltas, &p[i]);
1677 if (ret)
1678 die("unable to create thread: %s", strerror(ret));
1679 active_threads++;
1680 }
1681
1682 /*
1683 * Now let's wait for work completion. Each time a thread is done
1684 * with its work, we steal half of the remaining work from the
1685 * thread with the largest number of unprocessed objects and give
1686 * it to that newly idle thread. This ensure good load balancing
1687 * until the remaining object list segments are simply too short
1688 * to be worth splitting anymore.
1689 */
1690 while (active_threads) {
1691 struct thread_params *target = NULL;
1692 struct thread_params *victim = NULL;
1693 unsigned sub_size = 0;
1694
1695 progress_lock();
1696 for (;;) {
1697 for (i = 0; !target && i < delta_search_threads; i++)
1698 if (!p[i].working)
1699 target = &p[i];
1700 if (target)
1701 break;
1702 pthread_cond_wait(&progress_cond, &progress_mutex);
1703 }
1704
1705 for (i = 0; i < delta_search_threads; i++)
1706 if (p[i].remaining > 2*window &&
1707 (!victim || victim->remaining < p[i].remaining))
1708 victim = &p[i];
1709 if (victim) {
1710 sub_size = victim->remaining / 2;
1711 list = victim->list + victim->list_size - sub_size;
1712 while (sub_size && list[0]->hash &&
1713 list[0]->hash == list[-1]->hash) {
1714 list++;
1715 sub_size--;
1716 }
1717 if (!sub_size) {
1718 /*
1719 * It is possible for some "paths" to have
1720 * so many objects that no hash boundary
1721 * might be found. Let's just steal the
1722 * exact half in that case.
1723 */
1724 sub_size = victim->remaining / 2;
1725 list -= sub_size;
1726 }
1727 target->list = list;
1728 victim->list_size -= sub_size;
1729 victim->remaining -= sub_size;
1730 }
1731 target->list_size = sub_size;
1732 target->remaining = sub_size;
1733 target->working = 1;
1734 progress_unlock();
1735
1736 pthread_mutex_lock(&target->mutex);
1737 target->data_ready = 1;
1738 pthread_cond_signal(&target->cond);
1739 pthread_mutex_unlock(&target->mutex);
1740
1741 if (!sub_size) {
1742 pthread_join(target->thread, NULL);
1743 pthread_cond_destroy(&target->cond);
1744 pthread_mutex_destroy(&target->mutex);
1745 active_threads--;
1746 }
1747 }
1748 cleanup_threaded_search();
1749 free(p);
1750 }
1751
1752 #else
1753 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1754 #endif
1755
1756 static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1757 {
1758 unsigned char peeled[20];
1759
1760 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1761 !peel_ref(path, peeled) && /* peelable? */
1762 !is_null_sha1(peeled) && /* annotated tag? */
1763 locate_object_entry(peeled)) /* object packed? */
1764 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1765 return 0;
1766 }
1767
1768 static void prepare_pack(int window, int depth)
1769 {
1770 struct object_entry **delta_list;
1771 uint32_t i, nr_deltas;
1772 unsigned n;
1773
1774 get_object_details();
1775
1776 /*
1777 * If we're locally repacking then we need to be doubly careful
1778 * from now on in order to make sure no stealth corruption gets
1779 * propagated to the new pack. Clients receiving streamed packs
1780 * should validate everything they get anyway so no need to incur
1781 * the additional cost here in that case.
1782 */
1783 if (!pack_to_stdout)
1784 do_check_packed_object_crc = 1;
1785
1786 if (!nr_objects || !window || !depth)
1787 return;
1788
1789 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1790 nr_deltas = n = 0;
1791
1792 for (i = 0; i < nr_objects; i++) {
1793 struct object_entry *entry = objects + i;
1794
1795 if (entry->delta)
1796 /* This happens if we decided to reuse existing
1797 * delta from a pack. "reuse_delta &&" is implied.
1798 */
1799 continue;
1800
1801 if (entry->size < 50)
1802 continue;
1803
1804 if (entry->no_try_delta)
1805 continue;
1806
1807 if (!entry->preferred_base) {
1808 nr_deltas++;
1809 if (entry->type < 0)
1810 die("unable to get type of object %s",
1811 sha1_to_hex(entry->idx.sha1));
1812 } else {
1813 if (entry->type < 0) {
1814 /*
1815 * This object is not found, but we
1816 * don't have to include it anyway.
1817 */
1818 continue;
1819 }
1820 }
1821
1822 delta_list[n++] = entry;
1823 }
1824
1825 if (nr_deltas && n > 1) {
1826 unsigned nr_done = 0;
1827 if (progress)
1828 progress_state = start_progress("Compressing objects",
1829 nr_deltas);
1830 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
1831 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
1832 stop_progress(&progress_state);
1833 if (nr_done != nr_deltas)
1834 die("inconsistency with delta count");
1835 }
1836 free(delta_list);
1837 }
1838
1839 static int git_pack_config(const char *k, const char *v, void *cb)
1840 {
1841 if (!strcmp(k, "pack.window")) {
1842 window = git_config_int(k, v);
1843 return 0;
1844 }
1845 if (!strcmp(k, "pack.windowmemory")) {
1846 window_memory_limit = git_config_ulong(k, v);
1847 return 0;
1848 }
1849 if (!strcmp(k, "pack.depth")) {
1850 depth = git_config_int(k, v);
1851 return 0;
1852 }
1853 if (!strcmp(k, "pack.compression")) {
1854 int level = git_config_int(k, v);
1855 if (level == -1)
1856 level = Z_DEFAULT_COMPRESSION;
1857 else if (level < 0 || level > Z_BEST_COMPRESSION)
1858 die("bad pack compression level %d", level);
1859 pack_compression_level = level;
1860 pack_compression_seen = 1;
1861 return 0;
1862 }
1863 if (!strcmp(k, "pack.deltacachesize")) {
1864 max_delta_cache_size = git_config_int(k, v);
1865 return 0;
1866 }
1867 if (!strcmp(k, "pack.deltacachelimit")) {
1868 cache_max_small_delta_size = git_config_int(k, v);
1869 return 0;
1870 }
1871 if (!strcmp(k, "pack.threads")) {
1872 delta_search_threads = git_config_int(k, v);
1873 if (delta_search_threads < 0)
1874 die("invalid number of threads specified (%d)",
1875 delta_search_threads);
1876 #ifdef NO_PTHREADS
1877 if (delta_search_threads != 1)
1878 warning("no threads support, ignoring %s", k);
1879 #endif
1880 return 0;
1881 }
1882 if (!strcmp(k, "pack.indexversion")) {
1883 pack_idx_default_version = git_config_int(k, v);
1884 if (pack_idx_default_version > 2)
1885 die("bad pack.indexversion=%"PRIu32,
1886 pack_idx_default_version);
1887 return 0;
1888 }
1889 if (!strcmp(k, "pack.packsizelimit")) {
1890 pack_size_limit_cfg = git_config_ulong(k, v);
1891 return 0;
1892 }
1893 return git_default_config(k, v, cb);
1894 }
1895
1896 static void read_object_list_from_stdin(void)
1897 {
1898 char line[40 + 1 + PATH_MAX + 2];
1899 unsigned char sha1[20];
1900
1901 for (;;) {
1902 if (!fgets(line, sizeof(line), stdin)) {
1903 if (feof(stdin))
1904 break;
1905 if (!ferror(stdin))
1906 die("fgets returned NULL, not EOF, not error!");
1907 if (errno != EINTR)
1908 die_errno("fgets");
1909 clearerr(stdin);
1910 continue;
1911 }
1912 if (line[0] == '-') {
1913 if (get_sha1_hex(line+1, sha1))
1914 die("expected edge sha1, got garbage:\n %s",
1915 line);
1916 add_preferred_base(sha1);
1917 continue;
1918 }
1919 if (get_sha1_hex(line, sha1))
1920 die("expected sha1, got garbage:\n %s", line);
1921
1922 add_preferred_base_object(line+41);
1923 add_object_entry(sha1, 0, line+41, 0);
1924 }
1925 }
1926
1927 #define OBJECT_ADDED (1u<<20)
1928
1929 static void show_commit(struct commit *commit, void *data)
1930 {
1931 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
1932 commit->object.flags |= OBJECT_ADDED;
1933 }
1934
1935 static void show_object(struct object *obj, const struct name_path *path, const char *last)
1936 {
1937 char *name = path_name(path, last);
1938
1939 add_preferred_base_object(name);
1940 add_object_entry(obj->sha1, obj->type, name, 0);
1941 obj->flags |= OBJECT_ADDED;
1942
1943 /*
1944 * We will have generated the hash from the name,
1945 * but not saved a pointer to it - we can free it
1946 */
1947 free((char *)name);
1948 }
1949
1950 static void show_edge(struct commit *commit)
1951 {
1952 add_preferred_base(commit->object.sha1);
1953 }
1954
1955 struct in_pack_object {
1956 off_t offset;
1957 struct object *object;
1958 };
1959
1960 struct in_pack {
1961 int alloc;
1962 int nr;
1963 struct in_pack_object *array;
1964 };
1965
1966 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
1967 {
1968 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
1969 in_pack->array[in_pack->nr].object = object;
1970 in_pack->nr++;
1971 }
1972
1973 /*
1974 * Compare the objects in the offset order, in order to emulate the
1975 * "git rev-list --objects" output that produced the pack originally.
1976 */
1977 static int ofscmp(const void *a_, const void *b_)
1978 {
1979 struct in_pack_object *a = (struct in_pack_object *)a_;
1980 struct in_pack_object *b = (struct in_pack_object *)b_;
1981
1982 if (a->offset < b->offset)
1983 return -1;
1984 else if (a->offset > b->offset)
1985 return 1;
1986 else
1987 return hashcmp(a->object->sha1, b->object->sha1);
1988 }
1989
1990 static void add_objects_in_unpacked_packs(struct rev_info *revs)
1991 {
1992 struct packed_git *p;
1993 struct in_pack in_pack;
1994 uint32_t i;
1995
1996 memset(&in_pack, 0, sizeof(in_pack));
1997
1998 for (p = packed_git; p; p = p->next) {
1999 const unsigned char *sha1;
2000 struct object *o;
2001
2002 if (!p->pack_local || p->pack_keep)
2003 continue;
2004 if (open_pack_index(p))
2005 die("cannot open pack index");
2006
2007 ALLOC_GROW(in_pack.array,
2008 in_pack.nr + p->num_objects,
2009 in_pack.alloc);
2010
2011 for (i = 0; i < p->num_objects; i++) {
2012 sha1 = nth_packed_object_sha1(p, i);
2013 o = lookup_unknown_object(sha1);
2014 if (!(o->flags & OBJECT_ADDED))
2015 mark_in_pack_object(o, p, &in_pack);
2016 o->flags |= OBJECT_ADDED;
2017 }
2018 }
2019
2020 if (in_pack.nr) {
2021 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2022 ofscmp);
2023 for (i = 0; i < in_pack.nr; i++) {
2024 struct object *o = in_pack.array[i].object;
2025 add_object_entry(o->sha1, o->type, "", 0);
2026 }
2027 }
2028 free(in_pack.array);
2029 }
2030
2031 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2032 {
2033 static struct packed_git *last_found = (void *)1;
2034 struct packed_git *p;
2035
2036 p = (last_found != (void *)1) ? last_found : packed_git;
2037
2038 while (p) {
2039 if ((!p->pack_local || p->pack_keep) &&
2040 find_pack_entry_one(sha1, p)) {
2041 last_found = p;
2042 return 1;
2043 }
2044 if (p == last_found)
2045 p = packed_git;
2046 else
2047 p = p->next;
2048 if (p == last_found)
2049 p = p->next;
2050 }
2051 return 0;
2052 }
2053
2054 static void loosen_unused_packed_objects(struct rev_info *revs)
2055 {
2056 struct packed_git *p;
2057 uint32_t i;
2058 const unsigned char *sha1;
2059
2060 for (p = packed_git; p; p = p->next) {
2061 if (!p->pack_local || p->pack_keep)
2062 continue;
2063
2064 if (open_pack_index(p))
2065 die("cannot open pack index");
2066
2067 for (i = 0; i < p->num_objects; i++) {
2068 sha1 = nth_packed_object_sha1(p, i);
2069 if (!locate_object_entry(sha1) &&
2070 !has_sha1_pack_kept_or_nonlocal(sha1))
2071 if (force_object_loose(sha1, p->mtime))
2072 die("unable to force loose object");
2073 }
2074 }
2075 }
2076
2077 static void get_object_list(int ac, const char **av)
2078 {
2079 struct rev_info revs;
2080 char line[1000];
2081 int flags = 0;
2082
2083 init_revisions(&revs, NULL);
2084 save_commit_buffer = 0;
2085 setup_revisions(ac, av, &revs, NULL);
2086
2087 while (fgets(line, sizeof(line), stdin) != NULL) {
2088 int len = strlen(line);
2089 if (len && line[len - 1] == '\n')
2090 line[--len] = 0;
2091 if (!len)
2092 break;
2093 if (*line == '-') {
2094 if (!strcmp(line, "--not")) {
2095 flags ^= UNINTERESTING;
2096 continue;
2097 }
2098 die("not a rev '%s'", line);
2099 }
2100 if (handle_revision_arg(line, &revs, flags, 1))
2101 die("bad revision '%s'", line);
2102 }
2103
2104 if (prepare_revision_walk(&revs))
2105 die("revision walk setup failed");
2106 mark_edges_uninteresting(revs.commits, &revs, show_edge);
2107 traverse_commit_list(&revs, show_commit, show_object, NULL);
2108
2109 if (keep_unreachable)
2110 add_objects_in_unpacked_packs(&revs);
2111 if (unpack_unreachable)
2112 loosen_unused_packed_objects(&revs);
2113 }
2114
2115 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2116 {
2117 int use_internal_rev_list = 0;
2118 int thin = 0;
2119 int all_progress_implied = 0;
2120 uint32_t i;
2121 const char **rp_av;
2122 int rp_ac_alloc = 64;
2123 int rp_ac;
2124
2125 read_replace_refs = 0;
2126
2127 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2128
2129 rp_av[0] = "pack-objects";
2130 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2131 rp_ac = 2;
2132
2133 git_config(git_pack_config, NULL);
2134 if (!pack_compression_seen && core_compression_seen)
2135 pack_compression_level = core_compression_level;
2136
2137 progress = isatty(2);
2138 for (i = 1; i < argc; i++) {
2139 const char *arg = argv[i];
2140
2141 if (*arg != '-')
2142 break;
2143
2144 if (!strcmp("--non-empty", arg)) {
2145 non_empty = 1;
2146 continue;
2147 }
2148 if (!strcmp("--local", arg)) {
2149 local = 1;
2150 continue;
2151 }
2152 if (!strcmp("--incremental", arg)) {
2153 incremental = 1;
2154 continue;
2155 }
2156 if (!strcmp("--honor-pack-keep", arg)) {
2157 ignore_packed_keep = 1;
2158 continue;
2159 }
2160 if (!prefixcmp(arg, "--compression=")) {
2161 char *end;
2162 int level = strtoul(arg+14, &end, 0);
2163 if (!arg[14] || *end)
2164 usage(pack_usage);
2165 if (level == -1)
2166 level = Z_DEFAULT_COMPRESSION;
2167 else if (level < 0 || level > Z_BEST_COMPRESSION)
2168 die("bad pack compression level %d", level);
2169 pack_compression_level = level;
2170 continue;
2171 }
2172 if (!prefixcmp(arg, "--max-pack-size=")) {
2173 pack_size_limit_cfg = 0;
2174 if (!git_parse_ulong(arg+16, &pack_size_limit))
2175 usage(pack_usage);
2176 continue;
2177 }
2178 if (!prefixcmp(arg, "--window=")) {
2179 char *end;
2180 window = strtoul(arg+9, &end, 0);
2181 if (!arg[9] || *end)
2182 usage(pack_usage);
2183 continue;
2184 }
2185 if (!prefixcmp(arg, "--window-memory=")) {
2186 if (!git_parse_ulong(arg+16, &window_memory_limit))
2187 usage(pack_usage);
2188 continue;
2189 }
2190 if (!prefixcmp(arg, "--threads=")) {
2191 char *end;
2192 delta_search_threads = strtoul(arg+10, &end, 0);
2193 if (!arg[10] || *end || delta_search_threads < 0)
2194 usage(pack_usage);
2195 #ifdef NO_PTHREADS
2196 if (delta_search_threads != 1)
2197 warning("no threads support, "
2198 "ignoring %s", arg);
2199 #endif
2200 continue;
2201 }
2202 if (!prefixcmp(arg, "--depth=")) {
2203 char *end;
2204 depth = strtoul(arg+8, &end, 0);
2205 if (!arg[8] || *end)
2206 usage(pack_usage);
2207 continue;
2208 }
2209 if (!strcmp("--progress", arg)) {
2210 progress = 1;
2211 continue;
2212 }
2213 if (!strcmp("--all-progress", arg)) {
2214 progress = 2;
2215 continue;
2216 }
2217 if (!strcmp("--all-progress-implied", arg)) {
2218 all_progress_implied = 1;
2219 continue;
2220 }
2221 if (!strcmp("-q", arg)) {
2222 progress = 0;
2223 continue;
2224 }
2225 if (!strcmp("--no-reuse-delta", arg)) {
2226 reuse_delta = 0;
2227 continue;
2228 }
2229 if (!strcmp("--no-reuse-object", arg)) {
2230 reuse_object = reuse_delta = 0;
2231 continue;
2232 }
2233 if (!strcmp("--delta-base-offset", arg)) {
2234 allow_ofs_delta = 1;
2235 continue;
2236 }
2237 if (!strcmp("--stdout", arg)) {
2238 pack_to_stdout = 1;
2239 continue;
2240 }
2241 if (!strcmp("--revs", arg)) {
2242 use_internal_rev_list = 1;
2243 continue;
2244 }
2245 if (!strcmp("--keep-unreachable", arg)) {
2246 keep_unreachable = 1;
2247 continue;
2248 }
2249 if (!strcmp("--unpack-unreachable", arg)) {
2250 unpack_unreachable = 1;
2251 continue;
2252 }
2253 if (!strcmp("--include-tag", arg)) {
2254 include_tag = 1;
2255 continue;
2256 }
2257 if (!strcmp("--unpacked", arg) ||
2258 !strcmp("--reflog", arg) ||
2259 !strcmp("--all", arg)) {
2260 use_internal_rev_list = 1;
2261 if (rp_ac >= rp_ac_alloc - 1) {
2262 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2263 rp_av = xrealloc(rp_av,
2264 rp_ac_alloc * sizeof(*rp_av));
2265 }
2266 rp_av[rp_ac++] = arg;
2267 continue;
2268 }
2269 if (!strcmp("--thin", arg)) {
2270 use_internal_rev_list = 1;
2271 thin = 1;
2272 rp_av[1] = "--objects-edge";
2273 continue;
2274 }
2275 if (!prefixcmp(arg, "--index-version=")) {
2276 char *c;
2277 pack_idx_default_version = strtoul(arg + 16, &c, 10);
2278 if (pack_idx_default_version > 2)
2279 die("bad %s", arg);
2280 if (*c == ',')
2281 pack_idx_off32_limit = strtoul(c+1, &c, 0);
2282 if (*c || pack_idx_off32_limit & 0x80000000)
2283 die("bad %s", arg);
2284 continue;
2285 }
2286 if (!strcmp(arg, "--keep-true-parents")) {
2287 grafts_replace_parents = 0;
2288 continue;
2289 }
2290 usage(pack_usage);
2291 }
2292
2293 /* Traditionally "pack-objects [options] base extra" failed;
2294 * we would however want to take refs parameter that would
2295 * have been given to upstream rev-list ourselves, which means
2296 * we somehow want to say what the base name is. So the
2297 * syntax would be:
2298 *
2299 * pack-objects [options] base <refs...>
2300 *
2301 * in other words, we would treat the first non-option as the
2302 * base_name and send everything else to the internal revision
2303 * walker.
2304 */
2305
2306 if (!pack_to_stdout)
2307 base_name = argv[i++];
2308
2309 if (pack_to_stdout != !base_name)
2310 usage(pack_usage);
2311
2312 if (!pack_to_stdout && !pack_size_limit)
2313 pack_size_limit = pack_size_limit_cfg;
2314 if (pack_to_stdout && pack_size_limit)
2315 die("--max-pack-size cannot be used to build a pack for transfer.");
2316 if (pack_size_limit && pack_size_limit < 1024*1024) {
2317 warning("minimum pack size limit is 1 MiB");
2318 pack_size_limit = 1024*1024;
2319 }
2320
2321 if (!pack_to_stdout && thin)
2322 die("--thin cannot be used to build an indexable pack.");
2323
2324 if (keep_unreachable && unpack_unreachable)
2325 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2326
2327 if (progress && all_progress_implied)
2328 progress = 2;
2329
2330 prepare_packed_git();
2331
2332 if (progress)
2333 progress_state = start_progress("Counting objects", 0);
2334 if (!use_internal_rev_list)
2335 read_object_list_from_stdin();
2336 else {
2337 rp_av[rp_ac] = NULL;
2338 get_object_list(rp_ac, rp_av);
2339 }
2340 cleanup_preferred_base();
2341 if (include_tag && nr_result)
2342 for_each_ref(add_ref_tag, NULL);
2343 stop_progress(&progress_state);
2344
2345 if (non_empty && !nr_result)
2346 return 0;
2347 if (nr_result)
2348 prepare_pack(window, depth);
2349 write_pack_file();
2350 if (progress)
2351 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2352 " reused %"PRIu32" (delta %"PRIu32")\n",
2353 written, written_delta, reused, reused_delta);
2354 return 0;
2355 }