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"
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]";
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 };
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;
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;
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;
89 static unsigned long window_memory_limit = 0;
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;
99 /*
100 * stats
101 */
102 static uint32_t written, written_delta;
103 static uint32_t reused, reused_delta;
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;
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 }
127 static unsigned long do_compress(void **pptr, unsigned long size)
128 {
129 git_zstream stream;
130 void *in, *out;
131 unsigned long maxsize;
133 memset(&stream, 0, sizeof(stream));
134 git_deflate_init(&stream, pack_compression_level);
135 maxsize = git_deflate_bound(&stream, size);
137 in = *pptr;
138 out = xmalloc(maxsize);
139 *pptr = out;
141 stream.next_in = in;
142 stream.avail_in = size;
143 stream.next_out = out;
144 stream.avail_out = maxsize;
145 while (git_deflate(&stream, Z_FINISH) == Z_OK)
146 ; /* nothing */
147 git_deflate_end(&stream);
149 free(in);
150 return stream.total_out;
151 }
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 git_zstream stream;
164 unsigned char fakebuf[4096], *in;
165 int st;
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 }
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 long avail;
192 while (len) {
193 in = use_pack(p, w_curs, offset, &avail);
194 if (avail > len)
195 avail = (unsigned long)len;
196 sha1write(f, in, avail);
197 offset += avail;
198 len -= avail;
199 }
200 }
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;
213 if (!pack_to_stdout)
214 crc32_begin(f);
216 type = entry->type;
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;
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 */
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 */
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 }
284 if (entry->z_delta_size)
285 datalen = entry->z_delta_size;
286 else
287 datalen = do_compress(&buf, size);
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);
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;
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);
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 }
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 }
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 }
407 enum write_one_status {
408 WRITE_ONE_SKIP = -1, /* already written */
409 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
410 WRITE_ONE_WRITTEN = 1, /* normal */
411 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
412 };
414 static enum write_one_status write_one(struct sha1file *f,
415 struct object_entry *e,
416 off_t *offset)
417 {
418 unsigned long size;
419 int recursing;
421 /*
422 * we set offset to 1 (which is an impossible value) to mark
423 * the fact that this object is involved in "write its base
424 * first before writing a deltified object" recursion.
425 */
426 recursing = (e->idx.offset == 1);
427 if (recursing) {
428 warning("recursive delta detected for object %s",
429 sha1_to_hex(e->idx.sha1));
430 return WRITE_ONE_RECURSIVE;
431 } else if (e->idx.offset || e->preferred_base) {
432 /* offset is non zero if object is written already. */
433 return WRITE_ONE_SKIP;
434 }
436 /* if we are deltified, write out base object first. */
437 if (e->delta) {
438 e->idx.offset = 1; /* now recurse */
439 switch (write_one(f, e->delta, offset)) {
440 case WRITE_ONE_RECURSIVE:
441 /* we cannot depend on this one */
442 e->delta = NULL;
443 break;
444 default:
445 break;
446 case WRITE_ONE_BREAK:
447 e->idx.offset = recursing;
448 return WRITE_ONE_BREAK;
449 }
450 }
452 e->idx.offset = *offset;
453 size = write_object(f, e, *offset);
454 if (!size) {
455 e->idx.offset = recursing;
456 return WRITE_ONE_BREAK;
457 }
458 written_list[nr_written++] = &e->idx;
460 /* make sure off_t is sufficiently large not to wrap */
461 if (signed_add_overflows(*offset, size))
462 die("pack too large for current definition of off_t");
463 *offset += size;
464 return WRITE_ONE_WRITTEN;
465 }
467 static void write_pack_file(void)
468 {
469 uint32_t i = 0, j;
470 struct sha1file *f;
471 off_t offset;
472 struct pack_header hdr;
473 uint32_t nr_remaining = nr_result;
474 time_t last_mtime = 0;
476 if (progress > pack_to_stdout)
477 progress_state = start_progress("Writing objects", nr_result);
478 written_list = xmalloc(nr_objects * sizeof(*written_list));
480 do {
481 unsigned char sha1[20];
482 char *pack_tmp_name = NULL;
484 if (pack_to_stdout) {
485 f = sha1fd_throughput(1, "<stdout>", progress_state);
486 } else {
487 char tmpname[PATH_MAX];
488 int fd;
489 fd = odb_mkstemp(tmpname, sizeof(tmpname),
490 "pack/tmp_pack_XXXXXX");
491 pack_tmp_name = xstrdup(tmpname);
492 f = sha1fd(fd, pack_tmp_name);
493 }
495 hdr.hdr_signature = htonl(PACK_SIGNATURE);
496 hdr.hdr_version = htonl(PACK_VERSION);
497 hdr.hdr_entries = htonl(nr_remaining);
498 sha1write(f, &hdr, sizeof(hdr));
499 offset = sizeof(hdr);
500 nr_written = 0;
501 for (; i < nr_objects; i++) {
502 if (write_one(f, objects + i, &offset) == WRITE_ONE_BREAK)
503 break;
504 display_progress(progress_state, written);
505 }
507 /*
508 * Did we write the wrong # entries in the header?
509 * If so, rewrite it like in fast-import
510 */
511 if (pack_to_stdout) {
512 sha1close(f, sha1, CSUM_CLOSE);
513 } else if (nr_written == nr_remaining) {
514 sha1close(f, sha1, CSUM_FSYNC);
515 } else {
516 int fd = sha1close(f, sha1, 0);
517 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
518 nr_written, sha1, offset);
519 close(fd);
520 }
522 if (!pack_to_stdout) {
523 struct stat st;
524 const char *idx_tmp_name;
525 char tmpname[PATH_MAX];
527 idx_tmp_name = write_idx_file(NULL, written_list,
528 nr_written, sha1);
530 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
531 base_name, sha1_to_hex(sha1));
532 free_pack_by_name(tmpname);
533 if (adjust_shared_perm(pack_tmp_name))
534 die_errno("unable to make temporary pack file readable");
535 if (rename(pack_tmp_name, tmpname))
536 die_errno("unable to rename temporary pack file");
538 /*
539 * Packs are runtime accessed in their mtime
540 * order since newer packs are more likely to contain
541 * younger objects. So if we are creating multiple
542 * packs then we should modify the mtime of later ones
543 * to preserve this property.
544 */
545 if (stat(tmpname, &st) < 0) {
546 warning("failed to stat %s: %s",
547 tmpname, strerror(errno));
548 } else if (!last_mtime) {
549 last_mtime = st.st_mtime;
550 } else {
551 struct utimbuf utb;
552 utb.actime = st.st_atime;
553 utb.modtime = --last_mtime;
554 if (utime(tmpname, &utb) < 0)
555 warning("failed utime() on %s: %s",
556 tmpname, strerror(errno));
557 }
559 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
560 base_name, sha1_to_hex(sha1));
561 if (adjust_shared_perm(idx_tmp_name))
562 die_errno("unable to make temporary index file readable");
563 if (rename(idx_tmp_name, tmpname))
564 die_errno("unable to rename temporary index file");
566 free((void *) idx_tmp_name);
567 free(pack_tmp_name);
568 puts(sha1_to_hex(sha1));
569 }
571 /* mark written objects as written to previous pack */
572 for (j = 0; j < nr_written; j++) {
573 written_list[j]->offset = (off_t)-1;
574 }
575 nr_remaining -= nr_written;
576 } while (nr_remaining && i < nr_objects);
578 free(written_list);
579 stop_progress(&progress_state);
580 if (written != nr_result)
581 die("wrote %"PRIu32" objects while expecting %"PRIu32,
582 written, nr_result);
583 }
585 static int locate_object_entry_hash(const unsigned char *sha1)
586 {
587 int i;
588 unsigned int ui;
589 memcpy(&ui, sha1, sizeof(unsigned int));
590 i = ui % object_ix_hashsz;
591 while (0 < object_ix[i]) {
592 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
593 return i;
594 if (++i == object_ix_hashsz)
595 i = 0;
596 }
597 return -1 - i;
598 }
600 static struct object_entry *locate_object_entry(const unsigned char *sha1)
601 {
602 int i;
604 if (!object_ix_hashsz)
605 return NULL;
607 i = locate_object_entry_hash(sha1);
608 if (0 <= i)
609 return &objects[object_ix[i]-1];
610 return NULL;
611 }
613 static void rehash_objects(void)
614 {
615 uint32_t i;
616 struct object_entry *oe;
618 object_ix_hashsz = nr_objects * 3;
619 if (object_ix_hashsz < 1024)
620 object_ix_hashsz = 1024;
621 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
622 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
623 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
624 int ix = locate_object_entry_hash(oe->idx.sha1);
625 if (0 <= ix)
626 continue;
627 ix = -1 - ix;
628 object_ix[ix] = i + 1;
629 }
630 }
632 static unsigned name_hash(const char *name)
633 {
634 unsigned c, hash = 0;
636 if (!name)
637 return 0;
639 /*
640 * This effectively just creates a sortable number from the
641 * last sixteen non-whitespace characters. Last characters
642 * count "most", so things that end in ".c" sort together.
643 */
644 while ((c = *name++) != 0) {
645 if (isspace(c))
646 continue;
647 hash = (hash >> 2) + (c << 24);
648 }
649 return hash;
650 }
652 static void setup_delta_attr_check(struct git_attr_check *check)
653 {
654 static struct git_attr *attr_delta;
656 if (!attr_delta)
657 attr_delta = git_attr("delta");
659 check[0].attr = attr_delta;
660 }
662 static int no_try_delta(const char *path)
663 {
664 struct git_attr_check check[1];
666 setup_delta_attr_check(check);
667 if (git_checkattr(path, ARRAY_SIZE(check), check))
668 return 0;
669 if (ATTR_FALSE(check->value))
670 return 1;
671 return 0;
672 }
674 static int add_object_entry(const unsigned char *sha1, enum object_type type,
675 const char *name, int exclude)
676 {
677 struct object_entry *entry;
678 struct packed_git *p, *found_pack = NULL;
679 off_t found_offset = 0;
680 int ix;
681 unsigned hash = name_hash(name);
683 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
684 if (ix >= 0) {
685 if (exclude) {
686 entry = objects + object_ix[ix] - 1;
687 if (!entry->preferred_base)
688 nr_result--;
689 entry->preferred_base = 1;
690 }
691 return 0;
692 }
694 if (!exclude && local && has_loose_object_nonlocal(sha1))
695 return 0;
697 for (p = packed_git; p; p = p->next) {
698 off_t offset = find_pack_entry_one(sha1, p);
699 if (offset) {
700 if (!found_pack) {
701 found_offset = offset;
702 found_pack = p;
703 }
704 if (exclude)
705 break;
706 if (incremental)
707 return 0;
708 if (local && !p->pack_local)
709 return 0;
710 if (ignore_packed_keep && p->pack_local && p->pack_keep)
711 return 0;
712 }
713 }
715 if (nr_objects >= nr_alloc) {
716 nr_alloc = (nr_alloc + 1024) * 3 / 2;
717 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
718 }
720 entry = objects + nr_objects++;
721 memset(entry, 0, sizeof(*entry));
722 hashcpy(entry->idx.sha1, sha1);
723 entry->hash = hash;
724 if (type)
725 entry->type = type;
726 if (exclude)
727 entry->preferred_base = 1;
728 else
729 nr_result++;
730 if (found_pack) {
731 entry->in_pack = found_pack;
732 entry->in_pack_offset = found_offset;
733 }
735 if (object_ix_hashsz * 3 <= nr_objects * 4)
736 rehash_objects();
737 else
738 object_ix[-1 - ix] = nr_objects;
740 display_progress(progress_state, nr_objects);
742 if (name && no_try_delta(name))
743 entry->no_try_delta = 1;
745 return 1;
746 }
748 struct pbase_tree_cache {
749 unsigned char sha1[20];
750 int ref;
751 int temporary;
752 void *tree_data;
753 unsigned long tree_size;
754 };
756 static struct pbase_tree_cache *(pbase_tree_cache[256]);
757 static int pbase_tree_cache_ix(const unsigned char *sha1)
758 {
759 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
760 }
761 static int pbase_tree_cache_ix_incr(int ix)
762 {
763 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
764 }
766 static struct pbase_tree {
767 struct pbase_tree *next;
768 /* This is a phony "cache" entry; we are not
769 * going to evict it nor find it through _get()
770 * mechanism -- this is for the toplevel node that
771 * would almost always change with any commit.
772 */
773 struct pbase_tree_cache pcache;
774 } *pbase_tree;
776 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
777 {
778 struct pbase_tree_cache *ent, *nent;
779 void *data;
780 unsigned long size;
781 enum object_type type;
782 int neigh;
783 int my_ix = pbase_tree_cache_ix(sha1);
784 int available_ix = -1;
786 /* pbase-tree-cache acts as a limited hashtable.
787 * your object will be found at your index or within a few
788 * slots after that slot if it is cached.
789 */
790 for (neigh = 0; neigh < 8; neigh++) {
791 ent = pbase_tree_cache[my_ix];
792 if (ent && !hashcmp(ent->sha1, sha1)) {
793 ent->ref++;
794 return ent;
795 }
796 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
797 ((0 <= available_ix) &&
798 (!ent && pbase_tree_cache[available_ix])))
799 available_ix = my_ix;
800 if (!ent)
801 break;
802 my_ix = pbase_tree_cache_ix_incr(my_ix);
803 }
805 /* Did not find one. Either we got a bogus request or
806 * we need to read and perhaps cache.
807 */
808 data = read_sha1_file(sha1, &type, &size);
809 if (!data)
810 return NULL;
811 if (type != OBJ_TREE) {
812 free(data);
813 return NULL;
814 }
816 /* We need to either cache or return a throwaway copy */
818 if (available_ix < 0)
819 ent = NULL;
820 else {
821 ent = pbase_tree_cache[available_ix];
822 my_ix = available_ix;
823 }
825 if (!ent) {
826 nent = xmalloc(sizeof(*nent));
827 nent->temporary = (available_ix < 0);
828 }
829 else {
830 /* evict and reuse */
831 free(ent->tree_data);
832 nent = ent;
833 }
834 hashcpy(nent->sha1, sha1);
835 nent->tree_data = data;
836 nent->tree_size = size;
837 nent->ref = 1;
838 if (!nent->temporary)
839 pbase_tree_cache[my_ix] = nent;
840 return nent;
841 }
843 static void pbase_tree_put(struct pbase_tree_cache *cache)
844 {
845 if (!cache->temporary) {
846 cache->ref--;
847 return;
848 }
849 free(cache->tree_data);
850 free(cache);
851 }
853 static int name_cmp_len(const char *name)
854 {
855 int i;
856 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
857 ;
858 return i;
859 }
861 static void add_pbase_object(struct tree_desc *tree,
862 const char *name,
863 int cmplen,
864 const char *fullname)
865 {
866 struct name_entry entry;
867 int cmp;
869 while (tree_entry(tree,&entry)) {
870 if (S_ISGITLINK(entry.mode))
871 continue;
872 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
873 memcmp(name, entry.path, cmplen);
874 if (cmp > 0)
875 continue;
876 if (cmp < 0)
877 return;
878 if (name[cmplen] != '/') {
879 add_object_entry(entry.sha1,
880 object_type(entry.mode),
881 fullname, 1);
882 return;
883 }
884 if (S_ISDIR(entry.mode)) {
885 struct tree_desc sub;
886 struct pbase_tree_cache *tree;
887 const char *down = name+cmplen+1;
888 int downlen = name_cmp_len(down);
890 tree = pbase_tree_get(entry.sha1);
891 if (!tree)
892 return;
893 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
895 add_pbase_object(&sub, down, downlen, fullname);
896 pbase_tree_put(tree);
897 }
898 }
899 }
901 static unsigned *done_pbase_paths;
902 static int done_pbase_paths_num;
903 static int done_pbase_paths_alloc;
904 static int done_pbase_path_pos(unsigned hash)
905 {
906 int lo = 0;
907 int hi = done_pbase_paths_num;
908 while (lo < hi) {
909 int mi = (hi + lo) / 2;
910 if (done_pbase_paths[mi] == hash)
911 return mi;
912 if (done_pbase_paths[mi] < hash)
913 hi = mi;
914 else
915 lo = mi + 1;
916 }
917 return -lo-1;
918 }
920 static int check_pbase_path(unsigned hash)
921 {
922 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
923 if (0 <= pos)
924 return 1;
925 pos = -pos - 1;
926 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
927 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
928 done_pbase_paths = xrealloc(done_pbase_paths,
929 done_pbase_paths_alloc *
930 sizeof(unsigned));
931 }
932 done_pbase_paths_num++;
933 if (pos < done_pbase_paths_num)
934 memmove(done_pbase_paths + pos + 1,
935 done_pbase_paths + pos,
936 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
937 done_pbase_paths[pos] = hash;
938 return 0;
939 }
941 static void add_preferred_base_object(const char *name)
942 {
943 struct pbase_tree *it;
944 int cmplen;
945 unsigned hash = name_hash(name);
947 if (!num_preferred_base || check_pbase_path(hash))
948 return;
950 cmplen = name_cmp_len(name);
951 for (it = pbase_tree; it; it = it->next) {
952 if (cmplen == 0) {
953 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
954 }
955 else {
956 struct tree_desc tree;
957 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
958 add_pbase_object(&tree, name, cmplen, name);
959 }
960 }
961 }
963 static void add_preferred_base(unsigned char *sha1)
964 {
965 struct pbase_tree *it;
966 void *data;
967 unsigned long size;
968 unsigned char tree_sha1[20];
970 if (window <= num_preferred_base++)
971 return;
973 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
974 if (!data)
975 return;
977 for (it = pbase_tree; it; it = it->next) {
978 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
979 free(data);
980 return;
981 }
982 }
984 it = xcalloc(1, sizeof(*it));
985 it->next = pbase_tree;
986 pbase_tree = it;
988 hashcpy(it->pcache.sha1, tree_sha1);
989 it->pcache.tree_data = data;
990 it->pcache.tree_size = size;
991 }
993 static void cleanup_preferred_base(void)
994 {
995 struct pbase_tree *it;
996 unsigned i;
998 it = pbase_tree;
999 pbase_tree = NULL;
1000 while (it) {
1001 struct pbase_tree *this = it;
1002 it = this->next;
1003 free(this->pcache.tree_data);
1004 free(this);
1005 }
1007 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1008 if (!pbase_tree_cache[i])
1009 continue;
1010 free(pbase_tree_cache[i]->tree_data);
1011 free(pbase_tree_cache[i]);
1012 pbase_tree_cache[i] = NULL;
1013 }
1015 free(done_pbase_paths);
1016 done_pbase_paths = NULL;
1017 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1018 }
1020 static void check_object(struct object_entry *entry)
1021 {
1022 if (entry->in_pack) {
1023 struct packed_git *p = entry->in_pack;
1024 struct pack_window *w_curs = NULL;
1025 const unsigned char *base_ref = NULL;
1026 struct object_entry *base_entry;
1027 unsigned long used, used_0;
1028 unsigned long avail;
1029 off_t ofs;
1030 unsigned char *buf, c;
1032 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1034 /*
1035 * We want in_pack_type even if we do not reuse delta
1036 * since non-delta representations could still be reused.
1037 */
1038 used = unpack_object_header_buffer(buf, avail,
1039 &entry->in_pack_type,
1040 &entry->size);
1041 if (used == 0)
1042 goto give_up;
1044 /*
1045 * Determine if this is a delta and if so whether we can
1046 * reuse it or not. Otherwise let's find out as cheaply as
1047 * possible what the actual type and size for this object is.
1048 */
1049 switch (entry->in_pack_type) {
1050 default:
1051 /* Not a delta hence we've already got all we need. */
1052 entry->type = entry->in_pack_type;
1053 entry->in_pack_header_size = used;
1054 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1055 goto give_up;
1056 unuse_pack(&w_curs);
1057 return;
1058 case OBJ_REF_DELTA:
1059 if (reuse_delta && !entry->preferred_base)
1060 base_ref = use_pack(p, &w_curs,
1061 entry->in_pack_offset + used, NULL);
1062 entry->in_pack_header_size = used + 20;
1063 break;
1064 case OBJ_OFS_DELTA:
1065 buf = use_pack(p, &w_curs,
1066 entry->in_pack_offset + used, NULL);
1067 used_0 = 0;
1068 c = buf[used_0++];
1069 ofs = c & 127;
1070 while (c & 128) {
1071 ofs += 1;
1072 if (!ofs || MSB(ofs, 7)) {
1073 error("delta base offset overflow in pack for %s",
1074 sha1_to_hex(entry->idx.sha1));
1075 goto give_up;
1076 }
1077 c = buf[used_0++];
1078 ofs = (ofs << 7) + (c & 127);
1079 }
1080 ofs = entry->in_pack_offset - ofs;
1081 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1082 error("delta base offset out of bound for %s",
1083 sha1_to_hex(entry->idx.sha1));
1084 goto give_up;
1085 }
1086 if (reuse_delta && !entry->preferred_base) {
1087 struct revindex_entry *revidx;
1088 revidx = find_pack_revindex(p, ofs);
1089 if (!revidx)
1090 goto give_up;
1091 base_ref = nth_packed_object_sha1(p, revidx->nr);
1092 }
1093 entry->in_pack_header_size = used + used_0;
1094 break;
1095 }
1097 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1098 /*
1099 * If base_ref was set above that means we wish to
1100 * reuse delta data, and we even found that base
1101 * in the list of objects we want to pack. Goodie!
1102 *
1103 * Depth value does not matter - find_deltas() will
1104 * never consider reused delta as the base object to
1105 * deltify other objects against, in order to avoid
1106 * circular deltas.
1107 */
1108 entry->type = entry->in_pack_type;
1109 entry->delta = base_entry;
1110 entry->delta_size = entry->size;
1111 entry->delta_sibling = base_entry->delta_child;
1112 base_entry->delta_child = entry;
1113 unuse_pack(&w_curs);
1114 return;
1115 }
1117 if (entry->type) {
1118 /*
1119 * This must be a delta and we already know what the
1120 * final object type is. Let's extract the actual
1121 * object size from the delta header.
1122 */
1123 entry->size = get_size_from_delta(p, &w_curs,
1124 entry->in_pack_offset + entry->in_pack_header_size);
1125 if (entry->size == 0)
1126 goto give_up;
1127 unuse_pack(&w_curs);
1128 return;
1129 }
1131 /*
1132 * No choice but to fall back to the recursive delta walk
1133 * with sha1_object_info() to find about the object type
1134 * at this point...
1135 */
1136 give_up:
1137 unuse_pack(&w_curs);
1138 }
1140 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1141 /*
1142 * The error condition is checked in prepare_pack(). This is
1143 * to permit a missing preferred base object to be ignored
1144 * as a preferred base. Doing so can result in a larger
1145 * pack file, but the transfer will still take place.
1146 */
1147 }
1149 static int pack_offset_sort(const void *_a, const void *_b)
1150 {
1151 const struct object_entry *a = *(struct object_entry **)_a;
1152 const struct object_entry *b = *(struct object_entry **)_b;
1154 /* avoid filesystem trashing with loose objects */
1155 if (!a->in_pack && !b->in_pack)
1156 return hashcmp(a->idx.sha1, b->idx.sha1);
1158 if (a->in_pack < b->in_pack)
1159 return -1;
1160 if (a->in_pack > b->in_pack)
1161 return 1;
1162 return a->in_pack_offset < b->in_pack_offset ? -1 :
1163 (a->in_pack_offset > b->in_pack_offset);
1164 }
1166 static void get_object_details(void)
1167 {
1168 uint32_t i;
1169 struct object_entry **sorted_by_offset;
1171 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1172 for (i = 0; i < nr_objects; i++)
1173 sorted_by_offset[i] = objects + i;
1174 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1176 for (i = 0; i < nr_objects; i++) {
1177 struct object_entry *entry = sorted_by_offset[i];
1178 check_object(entry);
1179 if (big_file_threshold <= entry->size)
1180 entry->no_try_delta = 1;
1181 }
1183 free(sorted_by_offset);
1184 }
1186 /*
1187 * We search for deltas in a list sorted by type, by filename hash, and then
1188 * by size, so that we see progressively smaller and smaller files.
1189 * That's because we prefer deltas to be from the bigger file
1190 * to the smaller -- deletes are potentially cheaper, but perhaps
1191 * more importantly, the bigger file is likely the more recent
1192 * one. The deepest deltas are therefore the oldest objects which are
1193 * less susceptible to be accessed often.
1194 */
1195 static int type_size_sort(const void *_a, const void *_b)
1196 {
1197 const struct object_entry *a = *(struct object_entry **)_a;
1198 const struct object_entry *b = *(struct object_entry **)_b;
1200 if (a->type > b->type)
1201 return -1;
1202 if (a->type < b->type)
1203 return 1;
1204 if (a->hash > b->hash)
1205 return -1;
1206 if (a->hash < b->hash)
1207 return 1;
1208 if (a->preferred_base > b->preferred_base)
1209 return -1;
1210 if (a->preferred_base < b->preferred_base)
1211 return 1;
1212 if (a->size > b->size)
1213 return -1;
1214 if (a->size < b->size)
1215 return 1;
1216 return a < b ? -1 : (a > b); /* newest first */
1217 }
1219 struct unpacked {
1220 struct object_entry *entry;
1221 void *data;
1222 struct delta_index *index;
1223 unsigned depth;
1224 };
1226 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1227 unsigned long delta_size)
1228 {
1229 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1230 return 0;
1232 if (delta_size < cache_max_small_delta_size)
1233 return 1;
1235 /* cache delta, if objects are large enough compared to delta size */
1236 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1237 return 1;
1239 return 0;
1240 }
1242 #ifndef NO_PTHREADS
1244 static pthread_mutex_t read_mutex;
1245 #define read_lock() pthread_mutex_lock(&read_mutex)
1246 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1248 static pthread_mutex_t cache_mutex;
1249 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1250 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1252 static pthread_mutex_t progress_mutex;
1253 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1254 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1256 #else
1258 #define read_lock() (void)0
1259 #define read_unlock() (void)0
1260 #define cache_lock() (void)0
1261 #define cache_unlock() (void)0
1262 #define progress_lock() (void)0
1263 #define progress_unlock() (void)0
1265 #endif
1267 static int try_delta(struct unpacked *trg, struct unpacked *src,
1268 unsigned max_depth, unsigned long *mem_usage)
1269 {
1270 struct object_entry *trg_entry = trg->entry;
1271 struct object_entry *src_entry = src->entry;
1272 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1273 unsigned ref_depth;
1274 enum object_type type;
1275 void *delta_buf;
1277 /* Don't bother doing diffs between different types */
1278 if (trg_entry->type != src_entry->type)
1279 return -1;
1281 /*
1282 * We do not bother to try a delta that we discarded
1283 * on an earlier try, but only when reusing delta data.
1284 */
1285 if (reuse_delta && trg_entry->in_pack &&
1286 trg_entry->in_pack == src_entry->in_pack &&
1287 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1288 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1289 return 0;
1291 /* Let's not bust the allowed depth. */
1292 if (src->depth >= max_depth)
1293 return 0;
1295 /* Now some size filtering heuristics. */
1296 trg_size = trg_entry->size;
1297 if (!trg_entry->delta) {
1298 max_size = trg_size/2 - 20;
1299 ref_depth = 1;
1300 } else {
1301 max_size = trg_entry->delta_size;
1302 ref_depth = trg->depth;
1303 }
1304 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1305 (max_depth - ref_depth + 1);
1306 if (max_size == 0)
1307 return 0;
1308 src_size = src_entry->size;
1309 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1310 if (sizediff >= max_size)
1311 return 0;
1312 if (trg_size < src_size / 32)
1313 return 0;
1315 /* Load data if not already done */
1316 if (!trg->data) {
1317 read_lock();
1318 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1319 read_unlock();
1320 if (!trg->data)
1321 die("object %s cannot be read",
1322 sha1_to_hex(trg_entry->idx.sha1));
1323 if (sz != trg_size)
1324 die("object %s inconsistent object length (%lu vs %lu)",
1325 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1326 *mem_usage += sz;
1327 }
1328 if (!src->data) {
1329 read_lock();
1330 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1331 read_unlock();
1332 if (!src->data) {
1333 if (src_entry->preferred_base) {
1334 static int warned = 0;
1335 if (!warned++)
1336 warning("object %s cannot be read",
1337 sha1_to_hex(src_entry->idx.sha1));
1338 /*
1339 * Those objects are not included in the
1340 * resulting pack. Be resilient and ignore
1341 * them if they can't be read, in case the
1342 * pack could be created nevertheless.
1343 */
1344 return 0;
1345 }
1346 die("object %s cannot be read",
1347 sha1_to_hex(src_entry->idx.sha1));
1348 }
1349 if (sz != src_size)
1350 die("object %s inconsistent object length (%lu vs %lu)",
1351 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1352 *mem_usage += sz;
1353 }
1354 if (!src->index) {
1355 src->index = create_delta_index(src->data, src_size);
1356 if (!src->index) {
1357 static int warned = 0;
1358 if (!warned++)
1359 warning("suboptimal pack - out of memory");
1360 return 0;
1361 }
1362 *mem_usage += sizeof_delta_index(src->index);
1363 }
1365 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1366 if (!delta_buf)
1367 return 0;
1369 if (trg_entry->delta) {
1370 /* Prefer only shallower same-sized deltas. */
1371 if (delta_size == trg_entry->delta_size &&
1372 src->depth + 1 >= trg->depth) {
1373 free(delta_buf);
1374 return 0;
1375 }
1376 }
1378 /*
1379 * Handle memory allocation outside of the cache
1380 * accounting lock. Compiler will optimize the strangeness
1381 * away when NO_PTHREADS is defined.
1382 */
1383 free(trg_entry->delta_data);
1384 cache_lock();
1385 if (trg_entry->delta_data) {
1386 delta_cache_size -= trg_entry->delta_size;
1387 trg_entry->delta_data = NULL;
1388 }
1389 if (delta_cacheable(src_size, trg_size, delta_size)) {
1390 delta_cache_size += delta_size;
1391 cache_unlock();
1392 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1393 } else {
1394 cache_unlock();
1395 free(delta_buf);
1396 }
1398 trg_entry->delta = src_entry;
1399 trg_entry->delta_size = delta_size;
1400 trg->depth = src->depth + 1;
1402 return 1;
1403 }
1405 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1406 {
1407 struct object_entry *child = me->delta_child;
1408 unsigned int m = n;
1409 while (child) {
1410 unsigned int c = check_delta_limit(child, n + 1);
1411 if (m < c)
1412 m = c;
1413 child = child->delta_sibling;
1414 }
1415 return m;
1416 }
1418 static unsigned long free_unpacked(struct unpacked *n)
1419 {
1420 unsigned long freed_mem = sizeof_delta_index(n->index);
1421 free_delta_index(n->index);
1422 n->index = NULL;
1423 if (n->data) {
1424 freed_mem += n->entry->size;
1425 free(n->data);
1426 n->data = NULL;
1427 }
1428 n->entry = NULL;
1429 n->depth = 0;
1430 return freed_mem;
1431 }
1433 static void find_deltas(struct object_entry **list, unsigned *list_size,
1434 int window, int depth, unsigned *processed)
1435 {
1436 uint32_t i, idx = 0, count = 0;
1437 struct unpacked *array;
1438 unsigned long mem_usage = 0;
1440 array = xcalloc(window, sizeof(struct unpacked));
1442 for (;;) {
1443 struct object_entry *entry;
1444 struct unpacked *n = array + idx;
1445 int j, max_depth, best_base = -1;
1447 progress_lock();
1448 if (!*list_size) {
1449 progress_unlock();
1450 break;
1451 }
1452 entry = *list++;
1453 (*list_size)--;
1454 if (!entry->preferred_base) {
1455 (*processed)++;
1456 display_progress(progress_state, *processed);
1457 }
1458 progress_unlock();
1460 mem_usage -= free_unpacked(n);
1461 n->entry = entry;
1463 while (window_memory_limit &&
1464 mem_usage > window_memory_limit &&
1465 count > 1) {
1466 uint32_t tail = (idx + window - count) % window;
1467 mem_usage -= free_unpacked(array + tail);
1468 count--;
1469 }
1471 /* We do not compute delta to *create* objects we are not
1472 * going to pack.
1473 */
1474 if (entry->preferred_base)
1475 goto next;
1477 /*
1478 * If the current object is at pack edge, take the depth the
1479 * objects that depend on the current object into account
1480 * otherwise they would become too deep.
1481 */
1482 max_depth = depth;
1483 if (entry->delta_child) {
1484 max_depth -= check_delta_limit(entry, 0);
1485 if (max_depth <= 0)
1486 goto next;
1487 }
1489 j = window;
1490 while (--j > 0) {
1491 int ret;
1492 uint32_t other_idx = idx + j;
1493 struct unpacked *m;
1494 if (other_idx >= window)
1495 other_idx -= window;
1496 m = array + other_idx;
1497 if (!m->entry)
1498 break;
1499 ret = try_delta(n, m, max_depth, &mem_usage);
1500 if (ret < 0)
1501 break;
1502 else if (ret > 0)
1503 best_base = other_idx;
1504 }
1506 /*
1507 * If we decided to cache the delta data, then it is best
1508 * to compress it right away. First because we have to do
1509 * it anyway, and doing it here while we're threaded will
1510 * save a lot of time in the non threaded write phase,
1511 * as well as allow for caching more deltas within
1512 * the same cache size limit.
1513 * ...
1514 * But only if not writing to stdout, since in that case
1515 * the network is most likely throttling writes anyway,
1516 * and therefore it is best to go to the write phase ASAP
1517 * instead, as we can afford spending more time compressing
1518 * between writes at that moment.
1519 */
1520 if (entry->delta_data && !pack_to_stdout) {
1521 entry->z_delta_size = do_compress(&entry->delta_data,
1522 entry->delta_size);
1523 cache_lock();
1524 delta_cache_size -= entry->delta_size;
1525 delta_cache_size += entry->z_delta_size;
1526 cache_unlock();
1527 }
1529 /* if we made n a delta, and if n is already at max
1530 * depth, leaving it in the window is pointless. we
1531 * should evict it first.
1532 */
1533 if (entry->delta && max_depth <= n->depth)
1534 continue;
1536 /*
1537 * Move the best delta base up in the window, after the
1538 * currently deltified object, to keep it longer. It will
1539 * be the first base object to be attempted next.
1540 */
1541 if (entry->delta) {
1542 struct unpacked swap = array[best_base];
1543 int dist = (window + idx - best_base) % window;
1544 int dst = best_base;
1545 while (dist--) {
1546 int src = (dst + 1) % window;
1547 array[dst] = array[src];
1548 dst = src;
1549 }
1550 array[dst] = swap;
1551 }
1553 next:
1554 idx++;
1555 if (count + 1 < window)
1556 count++;
1557 if (idx >= window)
1558 idx = 0;
1559 }
1561 for (i = 0; i < window; ++i) {
1562 free_delta_index(array[i].index);
1563 free(array[i].data);
1564 }
1565 free(array);
1566 }
1568 #ifndef NO_PTHREADS
1570 static void try_to_free_from_threads(size_t size)
1571 {
1572 read_lock();
1573 release_pack_memory(size, -1);
1574 read_unlock();
1575 }
1577 static try_to_free_t old_try_to_free_routine;
1579 /*
1580 * The main thread waits on the condition that (at least) one of the workers
1581 * has stopped working (which is indicated in the .working member of
1582 * struct thread_params).
1583 * When a work thread has completed its work, it sets .working to 0 and
1584 * signals the main thread and waits on the condition that .data_ready
1585 * becomes 1.
1586 */
1588 struct thread_params {
1589 pthread_t thread;
1590 struct object_entry **list;
1591 unsigned list_size;
1592 unsigned remaining;
1593 int window;
1594 int depth;
1595 int working;
1596 int data_ready;
1597 pthread_mutex_t mutex;
1598 pthread_cond_t cond;
1599 unsigned *processed;
1600 };
1602 static pthread_cond_t progress_cond;
1604 /*
1605 * Mutex and conditional variable can't be statically-initialized on Windows.
1606 */
1607 static void init_threaded_search(void)
1608 {
1609 init_recursive_mutex(&read_mutex);
1610 pthread_mutex_init(&cache_mutex, NULL);
1611 pthread_mutex_init(&progress_mutex, NULL);
1612 pthread_cond_init(&progress_cond, NULL);
1613 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
1614 }
1616 static void cleanup_threaded_search(void)
1617 {
1618 set_try_to_free_routine(old_try_to_free_routine);
1619 pthread_cond_destroy(&progress_cond);
1620 pthread_mutex_destroy(&read_mutex);
1621 pthread_mutex_destroy(&cache_mutex);
1622 pthread_mutex_destroy(&progress_mutex);
1623 }
1625 static void *threaded_find_deltas(void *arg)
1626 {
1627 struct thread_params *me = arg;
1629 while (me->remaining) {
1630 find_deltas(me->list, &me->remaining,
1631 me->window, me->depth, me->processed);
1633 progress_lock();
1634 me->working = 0;
1635 pthread_cond_signal(&progress_cond);
1636 progress_unlock();
1638 /*
1639 * We must not set ->data_ready before we wait on the
1640 * condition because the main thread may have set it to 1
1641 * before we get here. In order to be sure that new
1642 * work is available if we see 1 in ->data_ready, it
1643 * was initialized to 0 before this thread was spawned
1644 * and we reset it to 0 right away.
1645 */
1646 pthread_mutex_lock(&me->mutex);
1647 while (!me->data_ready)
1648 pthread_cond_wait(&me->cond, &me->mutex);
1649 me->data_ready = 0;
1650 pthread_mutex_unlock(&me->mutex);
1651 }
1652 /* leave ->working 1 so that this doesn't get more work assigned */
1653 return NULL;
1654 }
1656 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1657 int window, int depth, unsigned *processed)
1658 {
1659 struct thread_params *p;
1660 int i, ret, active_threads = 0;
1662 init_threaded_search();
1664 if (!delta_search_threads) /* --threads=0 means autodetect */
1665 delta_search_threads = online_cpus();
1666 if (delta_search_threads <= 1) {
1667 find_deltas(list, &list_size, window, depth, processed);
1668 cleanup_threaded_search();
1669 return;
1670 }
1671 if (progress > pack_to_stdout)
1672 fprintf(stderr, "Delta compression using up to %d threads.\n",
1673 delta_search_threads);
1674 p = xcalloc(delta_search_threads, sizeof(*p));
1676 /* Partition the work amongst work threads. */
1677 for (i = 0; i < delta_search_threads; i++) {
1678 unsigned sub_size = list_size / (delta_search_threads - i);
1680 /* don't use too small segments or no deltas will be found */
1681 if (sub_size < 2*window && i+1 < delta_search_threads)
1682 sub_size = 0;
1684 p[i].window = window;
1685 p[i].depth = depth;
1686 p[i].processed = processed;
1687 p[i].working = 1;
1688 p[i].data_ready = 0;
1690 /* try to split chunks on "path" boundaries */
1691 while (sub_size && sub_size < list_size &&
1692 list[sub_size]->hash &&
1693 list[sub_size]->hash == list[sub_size-1]->hash)
1694 sub_size++;
1696 p[i].list = list;
1697 p[i].list_size = sub_size;
1698 p[i].remaining = sub_size;
1700 list += sub_size;
1701 list_size -= sub_size;
1702 }
1704 /* Start work threads. */
1705 for (i = 0; i < delta_search_threads; i++) {
1706 if (!p[i].list_size)
1707 continue;
1708 pthread_mutex_init(&p[i].mutex, NULL);
1709 pthread_cond_init(&p[i].cond, NULL);
1710 ret = pthread_create(&p[i].thread, NULL,
1711 threaded_find_deltas, &p[i]);
1712 if (ret)
1713 die("unable to create thread: %s", strerror(ret));
1714 active_threads++;
1715 }
1717 /*
1718 * Now let's wait for work completion. Each time a thread is done
1719 * with its work, we steal half of the remaining work from the
1720 * thread with the largest number of unprocessed objects and give
1721 * it to that newly idle thread. This ensure good load balancing
1722 * until the remaining object list segments are simply too short
1723 * to be worth splitting anymore.
1724 */
1725 while (active_threads) {
1726 struct thread_params *target = NULL;
1727 struct thread_params *victim = NULL;
1728 unsigned sub_size = 0;
1730 progress_lock();
1731 for (;;) {
1732 for (i = 0; !target && i < delta_search_threads; i++)
1733 if (!p[i].working)
1734 target = &p[i];
1735 if (target)
1736 break;
1737 pthread_cond_wait(&progress_cond, &progress_mutex);
1738 }
1740 for (i = 0; i < delta_search_threads; i++)
1741 if (p[i].remaining > 2*window &&
1742 (!victim || victim->remaining < p[i].remaining))
1743 victim = &p[i];
1744 if (victim) {
1745 sub_size = victim->remaining / 2;
1746 list = victim->list + victim->list_size - sub_size;
1747 while (sub_size && list[0]->hash &&
1748 list[0]->hash == list[-1]->hash) {
1749 list++;
1750 sub_size--;
1751 }
1752 if (!sub_size) {
1753 /*
1754 * It is possible for some "paths" to have
1755 * so many objects that no hash boundary
1756 * might be found. Let's just steal the
1757 * exact half in that case.
1758 */
1759 sub_size = victim->remaining / 2;
1760 list -= sub_size;
1761 }
1762 target->list = list;
1763 victim->list_size -= sub_size;
1764 victim->remaining -= sub_size;
1765 }
1766 target->list_size = sub_size;
1767 target->remaining = sub_size;
1768 target->working = 1;
1769 progress_unlock();
1771 pthread_mutex_lock(&target->mutex);
1772 target->data_ready = 1;
1773 pthread_cond_signal(&target->cond);
1774 pthread_mutex_unlock(&target->mutex);
1776 if (!sub_size) {
1777 pthread_join(target->thread, NULL);
1778 pthread_cond_destroy(&target->cond);
1779 pthread_mutex_destroy(&target->mutex);
1780 active_threads--;
1781 }
1782 }
1783 cleanup_threaded_search();
1784 free(p);
1785 }
1787 #else
1788 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1789 #endif
1791 static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1792 {
1793 unsigned char peeled[20];
1795 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1796 !peel_ref(path, peeled) && /* peelable? */
1797 !is_null_sha1(peeled) && /* annotated tag? */
1798 locate_object_entry(peeled)) /* object packed? */
1799 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1800 return 0;
1801 }
1803 static void prepare_pack(int window, int depth)
1804 {
1805 struct object_entry **delta_list;
1806 uint32_t i, nr_deltas;
1807 unsigned n;
1809 get_object_details();
1811 /*
1812 * If we're locally repacking then we need to be doubly careful
1813 * from now on in order to make sure no stealth corruption gets
1814 * propagated to the new pack. Clients receiving streamed packs
1815 * should validate everything they get anyway so no need to incur
1816 * the additional cost here in that case.
1817 */
1818 if (!pack_to_stdout)
1819 do_check_packed_object_crc = 1;
1821 if (!nr_objects || !window || !depth)
1822 return;
1824 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1825 nr_deltas = n = 0;
1827 for (i = 0; i < nr_objects; i++) {
1828 struct object_entry *entry = objects + i;
1830 if (entry->delta)
1831 /* This happens if we decided to reuse existing
1832 * delta from a pack. "reuse_delta &&" is implied.
1833 */
1834 continue;
1836 if (entry->size < 50)
1837 continue;
1839 if (entry->no_try_delta)
1840 continue;
1842 if (!entry->preferred_base) {
1843 nr_deltas++;
1844 if (entry->type < 0)
1845 die("unable to get type of object %s",
1846 sha1_to_hex(entry->idx.sha1));
1847 } else {
1848 if (entry->type < 0) {
1849 /*
1850 * This object is not found, but we
1851 * don't have to include it anyway.
1852 */
1853 continue;
1854 }
1855 }
1857 delta_list[n++] = entry;
1858 }
1860 if (nr_deltas && n > 1) {
1861 unsigned nr_done = 0;
1862 if (progress)
1863 progress_state = start_progress("Compressing objects",
1864 nr_deltas);
1865 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
1866 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
1867 stop_progress(&progress_state);
1868 if (nr_done != nr_deltas)
1869 die("inconsistency with delta count");
1870 }
1871 free(delta_list);
1872 }
1874 static int git_pack_config(const char *k, const char *v, void *cb)
1875 {
1876 if (!strcmp(k, "pack.window")) {
1877 window = git_config_int(k, v);
1878 return 0;
1879 }
1880 if (!strcmp(k, "pack.windowmemory")) {
1881 window_memory_limit = git_config_ulong(k, v);
1882 return 0;
1883 }
1884 if (!strcmp(k, "pack.depth")) {
1885 depth = git_config_int(k, v);
1886 return 0;
1887 }
1888 if (!strcmp(k, "pack.compression")) {
1889 int level = git_config_int(k, v);
1890 if (level == -1)
1891 level = Z_DEFAULT_COMPRESSION;
1892 else if (level < 0 || level > Z_BEST_COMPRESSION)
1893 die("bad pack compression level %d", level);
1894 pack_compression_level = level;
1895 pack_compression_seen = 1;
1896 return 0;
1897 }
1898 if (!strcmp(k, "pack.deltacachesize")) {
1899 max_delta_cache_size = git_config_int(k, v);
1900 return 0;
1901 }
1902 if (!strcmp(k, "pack.deltacachelimit")) {
1903 cache_max_small_delta_size = git_config_int(k, v);
1904 return 0;
1905 }
1906 if (!strcmp(k, "pack.threads")) {
1907 delta_search_threads = git_config_int(k, v);
1908 if (delta_search_threads < 0)
1909 die("invalid number of threads specified (%d)",
1910 delta_search_threads);
1911 #ifdef NO_PTHREADS
1912 if (delta_search_threads != 1)
1913 warning("no threads support, ignoring %s", k);
1914 #endif
1915 return 0;
1916 }
1917 if (!strcmp(k, "pack.indexversion")) {
1918 pack_idx_default_version = git_config_int(k, v);
1919 if (pack_idx_default_version > 2)
1920 die("bad pack.indexversion=%"PRIu32,
1921 pack_idx_default_version);
1922 return 0;
1923 }
1924 if (!strcmp(k, "pack.packsizelimit")) {
1925 pack_size_limit_cfg = git_config_ulong(k, v);
1926 return 0;
1927 }
1928 return git_default_config(k, v, cb);
1929 }
1931 static void read_object_list_from_stdin(void)
1932 {
1933 char line[40 + 1 + PATH_MAX + 2];
1934 unsigned char sha1[20];
1936 for (;;) {
1937 if (!fgets(line, sizeof(line), stdin)) {
1938 if (feof(stdin))
1939 break;
1940 if (!ferror(stdin))
1941 die("fgets returned NULL, not EOF, not error!");
1942 if (errno != EINTR)
1943 die_errno("fgets");
1944 clearerr(stdin);
1945 continue;
1946 }
1947 if (line[0] == '-') {
1948 if (get_sha1_hex(line+1, sha1))
1949 die("expected edge sha1, got garbage:\n %s",
1950 line);
1951 add_preferred_base(sha1);
1952 continue;
1953 }
1954 if (get_sha1_hex(line, sha1))
1955 die("expected sha1, got garbage:\n %s", line);
1957 add_preferred_base_object(line+41);
1958 add_object_entry(sha1, 0, line+41, 0);
1959 }
1960 }
1962 #define OBJECT_ADDED (1u<<20)
1964 static void show_commit(struct commit *commit, void *data)
1965 {
1966 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
1967 commit->object.flags |= OBJECT_ADDED;
1968 }
1970 static void show_object(struct object *obj, const struct name_path *path, const char *last)
1971 {
1972 char *name = path_name(path, last);
1974 add_preferred_base_object(name);
1975 add_object_entry(obj->sha1, obj->type, name, 0);
1976 obj->flags |= OBJECT_ADDED;
1978 /*
1979 * We will have generated the hash from the name,
1980 * but not saved a pointer to it - we can free it
1981 */
1982 free((char *)name);
1983 }
1985 static void show_edge(struct commit *commit)
1986 {
1987 add_preferred_base(commit->object.sha1);
1988 }
1990 struct in_pack_object {
1991 off_t offset;
1992 struct object *object;
1993 };
1995 struct in_pack {
1996 int alloc;
1997 int nr;
1998 struct in_pack_object *array;
1999 };
2001 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2002 {
2003 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
2004 in_pack->array[in_pack->nr].object = object;
2005 in_pack->nr++;
2006 }
2008 /*
2009 * Compare the objects in the offset order, in order to emulate the
2010 * "git rev-list --objects" output that produced the pack originally.
2011 */
2012 static int ofscmp(const void *a_, const void *b_)
2013 {
2014 struct in_pack_object *a = (struct in_pack_object *)a_;
2015 struct in_pack_object *b = (struct in_pack_object *)b_;
2017 if (a->offset < b->offset)
2018 return -1;
2019 else if (a->offset > b->offset)
2020 return 1;
2021 else
2022 return hashcmp(a->object->sha1, b->object->sha1);
2023 }
2025 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2026 {
2027 struct packed_git *p;
2028 struct in_pack in_pack;
2029 uint32_t i;
2031 memset(&in_pack, 0, sizeof(in_pack));
2033 for (p = packed_git; p; p = p->next) {
2034 const unsigned char *sha1;
2035 struct object *o;
2037 if (!p->pack_local || p->pack_keep)
2038 continue;
2039 if (open_pack_index(p))
2040 die("cannot open pack index");
2042 ALLOC_GROW(in_pack.array,
2043 in_pack.nr + p->num_objects,
2044 in_pack.alloc);
2046 for (i = 0; i < p->num_objects; i++) {
2047 sha1 = nth_packed_object_sha1(p, i);
2048 o = lookup_unknown_object(sha1);
2049 if (!(o->flags & OBJECT_ADDED))
2050 mark_in_pack_object(o, p, &in_pack);
2051 o->flags |= OBJECT_ADDED;
2052 }
2053 }
2055 if (in_pack.nr) {
2056 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2057 ofscmp);
2058 for (i = 0; i < in_pack.nr; i++) {
2059 struct object *o = in_pack.array[i].object;
2060 add_object_entry(o->sha1, o->type, "", 0);
2061 }
2062 }
2063 free(in_pack.array);
2064 }
2066 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2067 {
2068 static struct packed_git *last_found = (void *)1;
2069 struct packed_git *p;
2071 p = (last_found != (void *)1) ? last_found : packed_git;
2073 while (p) {
2074 if ((!p->pack_local || p->pack_keep) &&
2075 find_pack_entry_one(sha1, p)) {
2076 last_found = p;
2077 return 1;
2078 }
2079 if (p == last_found)
2080 p = packed_git;
2081 else
2082 p = p->next;
2083 if (p == last_found)
2084 p = p->next;
2085 }
2086 return 0;
2087 }
2089 static void loosen_unused_packed_objects(struct rev_info *revs)
2090 {
2091 struct packed_git *p;
2092 uint32_t i;
2093 const unsigned char *sha1;
2095 for (p = packed_git; p; p = p->next) {
2096 if (!p->pack_local || p->pack_keep)
2097 continue;
2099 if (open_pack_index(p))
2100 die("cannot open pack index");
2102 for (i = 0; i < p->num_objects; i++) {
2103 sha1 = nth_packed_object_sha1(p, i);
2104 if (!locate_object_entry(sha1) &&
2105 !has_sha1_pack_kept_or_nonlocal(sha1))
2106 if (force_object_loose(sha1, p->mtime))
2107 die("unable to force loose object");
2108 }
2109 }
2110 }
2112 static void get_object_list(int ac, const char **av)
2113 {
2114 struct rev_info revs;
2115 char line[1000];
2116 int flags = 0;
2118 init_revisions(&revs, NULL);
2119 save_commit_buffer = 0;
2120 setup_revisions(ac, av, &revs, NULL);
2122 while (fgets(line, sizeof(line), stdin) != NULL) {
2123 int len = strlen(line);
2124 if (len && line[len - 1] == '\n')
2125 line[--len] = 0;
2126 if (!len)
2127 break;
2128 if (*line == '-') {
2129 if (!strcmp(line, "--not")) {
2130 flags ^= UNINTERESTING;
2131 continue;
2132 }
2133 die("not a rev '%s'", line);
2134 }
2135 if (handle_revision_arg(line, &revs, flags, 1))
2136 die("bad revision '%s'", line);
2137 }
2139 if (prepare_revision_walk(&revs))
2140 die("revision walk setup failed");
2141 mark_edges_uninteresting(revs.commits, &revs, show_edge);
2142 traverse_commit_list(&revs, show_commit, show_object, NULL);
2144 if (keep_unreachable)
2145 add_objects_in_unpacked_packs(&revs);
2146 if (unpack_unreachable)
2147 loosen_unused_packed_objects(&revs);
2148 }
2150 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2151 {
2152 int use_internal_rev_list = 0;
2153 int thin = 0;
2154 int all_progress_implied = 0;
2155 uint32_t i;
2156 const char **rp_av;
2157 int rp_ac_alloc = 64;
2158 int rp_ac;
2160 read_replace_refs = 0;
2162 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2164 rp_av[0] = "pack-objects";
2165 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2166 rp_ac = 2;
2168 git_config(git_pack_config, NULL);
2169 if (!pack_compression_seen && core_compression_seen)
2170 pack_compression_level = core_compression_level;
2172 progress = isatty(2);
2173 for (i = 1; i < argc; i++) {
2174 const char *arg = argv[i];
2176 if (*arg != '-')
2177 break;
2179 if (!strcmp("--non-empty", arg)) {
2180 non_empty = 1;
2181 continue;
2182 }
2183 if (!strcmp("--local", arg)) {
2184 local = 1;
2185 continue;
2186 }
2187 if (!strcmp("--incremental", arg)) {
2188 incremental = 1;
2189 continue;
2190 }
2191 if (!strcmp("--honor-pack-keep", arg)) {
2192 ignore_packed_keep = 1;
2193 continue;
2194 }
2195 if (!prefixcmp(arg, "--compression=")) {
2196 char *end;
2197 int level = strtoul(arg+14, &end, 0);
2198 if (!arg[14] || *end)
2199 usage(pack_usage);
2200 if (level == -1)
2201 level = Z_DEFAULT_COMPRESSION;
2202 else if (level < 0 || level > Z_BEST_COMPRESSION)
2203 die("bad pack compression level %d", level);
2204 pack_compression_level = level;
2205 continue;
2206 }
2207 if (!prefixcmp(arg, "--max-pack-size=")) {
2208 pack_size_limit_cfg = 0;
2209 if (!git_parse_ulong(arg+16, &pack_size_limit))
2210 usage(pack_usage);
2211 continue;
2212 }
2213 if (!prefixcmp(arg, "--window=")) {
2214 char *end;
2215 window = strtoul(arg+9, &end, 0);
2216 if (!arg[9] || *end)
2217 usage(pack_usage);
2218 continue;
2219 }
2220 if (!prefixcmp(arg, "--window-memory=")) {
2221 if (!git_parse_ulong(arg+16, &window_memory_limit))
2222 usage(pack_usage);
2223 continue;
2224 }
2225 if (!prefixcmp(arg, "--threads=")) {
2226 char *end;
2227 delta_search_threads = strtoul(arg+10, &end, 0);
2228 if (!arg[10] || *end || delta_search_threads < 0)
2229 usage(pack_usage);
2230 #ifdef NO_PTHREADS
2231 if (delta_search_threads != 1)
2232 warning("no threads support, "
2233 "ignoring %s", arg);
2234 #endif
2235 continue;
2236 }
2237 if (!prefixcmp(arg, "--depth=")) {
2238 char *end;
2239 depth = strtoul(arg+8, &end, 0);
2240 if (!arg[8] || *end)
2241 usage(pack_usage);
2242 continue;
2243 }
2244 if (!strcmp("--progress", arg)) {
2245 progress = 1;
2246 continue;
2247 }
2248 if (!strcmp("--all-progress", arg)) {
2249 progress = 2;
2250 continue;
2251 }
2252 if (!strcmp("--all-progress-implied", arg)) {
2253 all_progress_implied = 1;
2254 continue;
2255 }
2256 if (!strcmp("-q", arg)) {
2257 progress = 0;
2258 continue;
2259 }
2260 if (!strcmp("--no-reuse-delta", arg)) {
2261 reuse_delta = 0;
2262 continue;
2263 }
2264 if (!strcmp("--no-reuse-object", arg)) {
2265 reuse_object = reuse_delta = 0;
2266 continue;
2267 }
2268 if (!strcmp("--delta-base-offset", arg)) {
2269 allow_ofs_delta = 1;
2270 continue;
2271 }
2272 if (!strcmp("--stdout", arg)) {
2273 pack_to_stdout = 1;
2274 continue;
2275 }
2276 if (!strcmp("--revs", arg)) {
2277 use_internal_rev_list = 1;
2278 continue;
2279 }
2280 if (!strcmp("--keep-unreachable", arg)) {
2281 keep_unreachable = 1;
2282 continue;
2283 }
2284 if (!strcmp("--unpack-unreachable", arg)) {
2285 unpack_unreachable = 1;
2286 continue;
2287 }
2288 if (!strcmp("--include-tag", arg)) {
2289 include_tag = 1;
2290 continue;
2291 }
2292 if (!strcmp("--unpacked", arg) ||
2293 !strcmp("--reflog", arg) ||
2294 !strcmp("--all", arg)) {
2295 use_internal_rev_list = 1;
2296 if (rp_ac >= rp_ac_alloc - 1) {
2297 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2298 rp_av = xrealloc(rp_av,
2299 rp_ac_alloc * sizeof(*rp_av));
2300 }
2301 rp_av[rp_ac++] = arg;
2302 continue;
2303 }
2304 if (!strcmp("--thin", arg)) {
2305 use_internal_rev_list = 1;
2306 thin = 1;
2307 rp_av[1] = "--objects-edge";
2308 continue;
2309 }
2310 if (!prefixcmp(arg, "--index-version=")) {
2311 char *c;
2312 pack_idx_default_version = strtoul(arg + 16, &c, 10);
2313 if (pack_idx_default_version > 2)
2314 die("bad %s", arg);
2315 if (*c == ',')
2316 pack_idx_off32_limit = strtoul(c+1, &c, 0);
2317 if (*c || pack_idx_off32_limit & 0x80000000)
2318 die("bad %s", arg);
2319 continue;
2320 }
2321 if (!strcmp(arg, "--keep-true-parents")) {
2322 grafts_replace_parents = 0;
2323 continue;
2324 }
2325 usage(pack_usage);
2326 }
2328 /* Traditionally "pack-objects [options] base extra" failed;
2329 * we would however want to take refs parameter that would
2330 * have been given to upstream rev-list ourselves, which means
2331 * we somehow want to say what the base name is. So the
2332 * syntax would be:
2333 *
2334 * pack-objects [options] base <refs...>
2335 *
2336 * in other words, we would treat the first non-option as the
2337 * base_name and send everything else to the internal revision
2338 * walker.
2339 */
2341 if (!pack_to_stdout)
2342 base_name = argv[i++];
2344 if (pack_to_stdout != !base_name)
2345 usage(pack_usage);
2347 if (!pack_to_stdout && !pack_size_limit)
2348 pack_size_limit = pack_size_limit_cfg;
2349 if (pack_to_stdout && pack_size_limit)
2350 die("--max-pack-size cannot be used to build a pack for transfer.");
2351 if (pack_size_limit && pack_size_limit < 1024*1024) {
2352 warning("minimum pack size limit is 1 MiB");
2353 pack_size_limit = 1024*1024;
2354 }
2356 if (!pack_to_stdout && thin)
2357 die("--thin cannot be used to build an indexable pack.");
2359 if (keep_unreachable && unpack_unreachable)
2360 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2362 if (progress && all_progress_implied)
2363 progress = 2;
2365 prepare_packed_git();
2367 if (progress)
2368 progress_state = start_progress("Counting objects", 0);
2369 if (!use_internal_rev_list)
2370 read_object_list_from_stdin();
2371 else {
2372 rp_av[rp_ac] = NULL;
2373 get_object_list(rp_ac, rp_av);
2374 }
2375 cleanup_preferred_base();
2376 if (include_tag && nr_result)
2377 for_each_ref(add_ref_tag, NULL);
2378 stop_progress(&progress_state);
2380 if (non_empty && !nr_result)
2381 return 0;
2382 if (nr_result)
2383 prepare_pack(window, depth);
2384 write_pack_file();
2385 if (progress)
2386 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2387 " reused %"PRIu32" (delta %"PRIu32")\n",
2388 written, written_delta, reused, reused_delta);
2389 return 0;
2390 }