1 #include "builtin.h"
2 #include "cache.h"
3 #include "attr.h"
4 #include "object.h"
5 #include "blob.h"
6 #include "commit.h"
7 #include "tag.h"
8 #include "tree.h"
9 #include "delta.h"
10 #include "pack.h"
11 #include "pack-revindex.h"
12 #include "csum-file.h"
13 #include "tree-walk.h"
14 #include "diff.h"
15 #include "revision.h"
16 #include "list-objects.h"
17 #include "progress.h"
18 #include "refs.h"
20 #ifdef THREADED_DELTA_SEARCH
21 #include "thread-utils.h"
22 #include <pthread.h>
23 #endif
25 static const char pack_usage[] =
26 "git pack-objects [{ -q | --progress | --all-progress }]\n"
27 " [--all-progress-implied]\n"
28 " [--max-pack-size=N] [--local] [--incremental]\n"
29 " [--window=N] [--window-memory=N] [--depth=N]\n"
30 " [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset]\n"
31 " [--threads=N] [--non-empty] [--revs [--unpacked | --all]*]\n"
32 " [--reflog] [--stdout | base-name] [--include-tag]\n"
33 " [--keep-unreachable | --unpack-unreachable \n"
34 " [<ref-list | <object-list]";
36 struct object_entry {
37 struct pack_idx_entry idx;
38 unsigned long size; /* uncompressed size */
39 struct packed_git *in_pack; /* already in pack */
40 off_t in_pack_offset;
41 struct object_entry *delta; /* delta base object */
42 struct object_entry *delta_child; /* deltified objects who bases me */
43 struct object_entry *delta_sibling; /* other deltified objects who
44 * uses the same base as me
45 */
46 void *delta_data; /* cached delta (uncompressed) */
47 unsigned long delta_size; /* delta data size (uncompressed) */
48 unsigned long z_delta_size; /* delta data size (compressed) */
49 unsigned int hash; /* name hint hash */
50 enum object_type type;
51 enum object_type in_pack_type; /* could be delta */
52 unsigned char in_pack_header_size;
53 unsigned char preferred_base; /* we do not pack this, but is available
54 * to be used as the base object to delta
55 * objects against.
56 */
57 unsigned char no_try_delta;
58 };
60 /*
61 * Objects we are going to pack are collected in objects array (dynamically
62 * expanded). nr_objects & nr_alloc controls this array. They are stored
63 * in the order we see -- typically rev-list --objects order that gives us
64 * nice "minimum seek" order.
65 */
66 static struct object_entry *objects;
67 static struct pack_idx_entry **written_list;
68 static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
70 static int non_empty;
71 static int reuse_delta = 1, reuse_object = 1;
72 static int keep_unreachable, unpack_unreachable, include_tag;
73 static int local;
74 static int incremental;
75 static int ignore_packed_keep;
76 static int allow_ofs_delta;
77 static const char *base_name;
78 static int progress = 1;
79 static int window = 10;
80 static uint32_t pack_size_limit, pack_size_limit_cfg;
81 static int depth = 50;
82 static int delta_search_threads;
83 static int pack_to_stdout;
84 static int num_preferred_base;
85 static struct progress *progress_state;
86 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
87 static int pack_compression_seen;
89 static unsigned long delta_cache_size = 0;
90 static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
91 static unsigned long cache_max_small_delta_size = 1000;
93 static unsigned long window_memory_limit = 0;
95 /*
96 * The object names in objects array are hashed with this hashtable,
97 * to help looking up the entry by object name.
98 * This hashtable is built after all the objects are seen.
99 */
100 static int *object_ix;
101 static int object_ix_hashsz;
103 /*
104 * stats
105 */
106 static uint32_t written, written_delta;
107 static uint32_t reused, reused_delta;
110 static void *get_delta(struct object_entry *entry)
111 {
112 unsigned long size, base_size, delta_size;
113 void *buf, *base_buf, *delta_buf;
114 enum object_type type;
116 buf = read_sha1_file(entry->idx.sha1, &type, &size);
117 if (!buf)
118 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
119 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
120 if (!base_buf)
121 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
122 delta_buf = diff_delta(base_buf, base_size,
123 buf, size, &delta_size, 0);
124 if (!delta_buf || delta_size != entry->delta_size)
125 die("delta size changed");
126 free(buf);
127 free(base_buf);
128 return delta_buf;
129 }
131 static unsigned long do_compress(void **pptr, unsigned long size)
132 {
133 z_stream stream;
134 void *in, *out;
135 unsigned long maxsize;
137 memset(&stream, 0, sizeof(stream));
138 deflateInit(&stream, pack_compression_level);
139 maxsize = deflateBound(&stream, size);
141 in = *pptr;
142 out = xmalloc(maxsize);
143 *pptr = out;
145 stream.next_in = in;
146 stream.avail_in = size;
147 stream.next_out = out;
148 stream.avail_out = maxsize;
149 while (deflate(&stream, Z_FINISH) == Z_OK)
150 ; /* nothing */
151 deflateEnd(&stream);
153 free(in);
154 return stream.total_out;
155 }
157 /*
158 * The per-object header is a pretty dense thing, which is
159 * - first byte: low four bits are "size", then three bits of "type",
160 * and the high bit is "size continues".
161 * - each byte afterwards: low seven bits are size continuation,
162 * with the high bit being "size continues"
163 */
164 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
165 {
166 int n = 1;
167 unsigned char c;
169 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
170 die("bad type %d", type);
172 c = (type << 4) | (size & 15);
173 size >>= 4;
174 while (size) {
175 *hdr++ = c | 0x80;
176 c = size & 0x7f;
177 size >>= 7;
178 n++;
179 }
180 *hdr = c;
181 return n;
182 }
184 /*
185 * we are going to reuse the existing object data as is. make
186 * sure it is not corrupt.
187 */
188 static int check_pack_inflate(struct packed_git *p,
189 struct pack_window **w_curs,
190 off_t offset,
191 off_t len,
192 unsigned long expect)
193 {
194 z_stream stream;
195 unsigned char fakebuf[4096], *in;
196 int st;
198 memset(&stream, 0, sizeof(stream));
199 git_inflate_init(&stream);
200 do {
201 in = use_pack(p, w_curs, offset, &stream.avail_in);
202 stream.next_in = in;
203 stream.next_out = fakebuf;
204 stream.avail_out = sizeof(fakebuf);
205 st = git_inflate(&stream, Z_FINISH);
206 offset += stream.next_in - in;
207 } while (st == Z_OK || st == Z_BUF_ERROR);
208 git_inflate_end(&stream);
209 return (st == Z_STREAM_END &&
210 stream.total_out == expect &&
211 stream.total_in == len) ? 0 : -1;
212 }
214 static void copy_pack_data(struct sha1file *f,
215 struct packed_git *p,
216 struct pack_window **w_curs,
217 off_t offset,
218 off_t len)
219 {
220 unsigned char *in;
221 unsigned int avail;
223 while (len) {
224 in = use_pack(p, w_curs, offset, &avail);
225 if (avail > len)
226 avail = (unsigned int)len;
227 sha1write(f, in, avail);
228 offset += avail;
229 len -= avail;
230 }
231 }
233 static unsigned long write_object(struct sha1file *f,
234 struct object_entry *entry,
235 off_t write_offset)
236 {
237 unsigned long size, limit, datalen;
238 void *buf;
239 unsigned char header[10], dheader[10];
240 unsigned hdrlen;
241 enum object_type type;
242 int usable_delta, to_reuse;
244 if (!pack_to_stdout)
245 crc32_begin(f);
247 type = entry->type;
249 /* write limit if limited packsize and not first object */
250 if (!pack_size_limit || !nr_written)
251 limit = 0;
252 else if (pack_size_limit <= write_offset)
253 /*
254 * the earlier object did not fit the limit; avoid
255 * mistaking this with unlimited (i.e. limit = 0).
256 */
257 limit = 1;
258 else
259 limit = pack_size_limit - write_offset;
261 if (!entry->delta)
262 usable_delta = 0; /* no delta */
263 else if (!pack_size_limit)
264 usable_delta = 1; /* unlimited packfile */
265 else if (entry->delta->idx.offset == (off_t)-1)
266 usable_delta = 0; /* base was written to another pack */
267 else if (entry->delta->idx.offset)
268 usable_delta = 1; /* base already exists in this pack */
269 else
270 usable_delta = 0; /* base could end up in another pack */
272 if (!reuse_object)
273 to_reuse = 0; /* explicit */
274 else if (!entry->in_pack)
275 to_reuse = 0; /* can't reuse what we don't have */
276 else if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA)
277 /* check_object() decided it for us ... */
278 to_reuse = usable_delta;
279 /* ... but pack split may override that */
280 else if (type != entry->in_pack_type)
281 to_reuse = 0; /* pack has delta which is unusable */
282 else if (entry->delta)
283 to_reuse = 0; /* we want to pack afresh */
284 else
285 to_reuse = 1; /* we have it in-pack undeltified,
286 * and we do not need to deltify it.
287 */
289 if (!to_reuse) {
290 no_reuse:
291 if (!usable_delta) {
292 buf = read_sha1_file(entry->idx.sha1, &type, &size);
293 if (!buf)
294 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
295 /*
296 * make sure no cached delta data remains from a
297 * previous attempt before a pack split occurred.
298 */
299 free(entry->delta_data);
300 entry->delta_data = NULL;
301 entry->z_delta_size = 0;
302 } else if (entry->delta_data) {
303 size = entry->delta_size;
304 buf = entry->delta_data;
305 entry->delta_data = NULL;
306 type = (allow_ofs_delta && entry->delta->idx.offset) ?
307 OBJ_OFS_DELTA : OBJ_REF_DELTA;
308 } else {
309 buf = get_delta(entry);
310 size = entry->delta_size;
311 type = (allow_ofs_delta && entry->delta->idx.offset) ?
312 OBJ_OFS_DELTA : OBJ_REF_DELTA;
313 }
315 if (entry->z_delta_size)
316 datalen = entry->z_delta_size;
317 else
318 datalen = do_compress(&buf, size);
320 /*
321 * The object header is a byte of 'type' followed by zero or
322 * more bytes of length.
323 */
324 hdrlen = encode_header(type, size, header);
326 if (type == OBJ_OFS_DELTA) {
327 /*
328 * Deltas with relative base contain an additional
329 * encoding of the relative offset for the delta
330 * base from this object's position in the pack.
331 */
332 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
333 unsigned pos = sizeof(dheader) - 1;
334 dheader[pos] = ofs & 127;
335 while (ofs >>= 7)
336 dheader[--pos] = 128 | (--ofs & 127);
337 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
338 free(buf);
339 return 0;
340 }
341 sha1write(f, header, hdrlen);
342 sha1write(f, dheader + pos, sizeof(dheader) - pos);
343 hdrlen += sizeof(dheader) - pos;
344 } else if (type == OBJ_REF_DELTA) {
345 /*
346 * Deltas with a base reference contain
347 * an additional 20 bytes for the base sha1.
348 */
349 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
350 free(buf);
351 return 0;
352 }
353 sha1write(f, header, hdrlen);
354 sha1write(f, entry->delta->idx.sha1, 20);
355 hdrlen += 20;
356 } else {
357 if (limit && hdrlen + datalen + 20 >= limit) {
358 free(buf);
359 return 0;
360 }
361 sha1write(f, header, hdrlen);
362 }
363 sha1write(f, buf, datalen);
364 free(buf);
365 }
366 else {
367 struct packed_git *p = entry->in_pack;
368 struct pack_window *w_curs = NULL;
369 struct revindex_entry *revidx;
370 off_t offset;
372 if (entry->delta)
373 type = (allow_ofs_delta && entry->delta->idx.offset) ?
374 OBJ_OFS_DELTA : OBJ_REF_DELTA;
375 hdrlen = encode_header(type, entry->size, header);
377 offset = entry->in_pack_offset;
378 revidx = find_pack_revindex(p, offset);
379 datalen = revidx[1].offset - offset;
380 if (!pack_to_stdout && p->index_version > 1 &&
381 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
382 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
383 unuse_pack(&w_curs);
384 goto no_reuse;
385 }
387 offset += entry->in_pack_header_size;
388 datalen -= entry->in_pack_header_size;
389 if (!pack_to_stdout && p->index_version == 1 &&
390 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
391 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
392 unuse_pack(&w_curs);
393 goto no_reuse;
394 }
396 if (type == OBJ_OFS_DELTA) {
397 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
398 unsigned pos = sizeof(dheader) - 1;
399 dheader[pos] = ofs & 127;
400 while (ofs >>= 7)
401 dheader[--pos] = 128 | (--ofs & 127);
402 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
403 unuse_pack(&w_curs);
404 return 0;
405 }
406 sha1write(f, header, hdrlen);
407 sha1write(f, dheader + pos, sizeof(dheader) - pos);
408 hdrlen += sizeof(dheader) - pos;
409 reused_delta++;
410 } else if (type == OBJ_REF_DELTA) {
411 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
412 unuse_pack(&w_curs);
413 return 0;
414 }
415 sha1write(f, header, hdrlen);
416 sha1write(f, entry->delta->idx.sha1, 20);
417 hdrlen += 20;
418 reused_delta++;
419 } else {
420 if (limit && hdrlen + datalen + 20 >= limit) {
421 unuse_pack(&w_curs);
422 return 0;
423 }
424 sha1write(f, header, hdrlen);
425 }
426 copy_pack_data(f, p, &w_curs, offset, datalen);
427 unuse_pack(&w_curs);
428 reused++;
429 }
430 if (usable_delta)
431 written_delta++;
432 written++;
433 if (!pack_to_stdout)
434 entry->idx.crc32 = crc32_end(f);
435 return hdrlen + datalen;
436 }
438 static int write_one(struct sha1file *f,
439 struct object_entry *e,
440 off_t *offset)
441 {
442 unsigned long size;
444 /* offset is non zero if object is written already. */
445 if (e->idx.offset || e->preferred_base)
446 return 1;
448 /* if we are deltified, write out base object first. */
449 if (e->delta && !write_one(f, e->delta, offset))
450 return 0;
452 e->idx.offset = *offset;
453 size = write_object(f, e, *offset);
454 if (!size) {
455 e->idx.offset = 0;
456 return 0;
457 }
458 written_list[nr_written++] = &e->idx;
460 /* make sure off_t is sufficiently large not to wrap */
461 if (*offset > *offset + size)
462 die("pack too large for current definition of off_t");
463 *offset += size;
464 return 1;
465 }
467 /* forward declaration for write_pack_file */
468 static int adjust_perm(const char *path, mode_t mode);
470 static void write_pack_file(void)
471 {
472 uint32_t i = 0, j;
473 struct sha1file *f;
474 off_t offset;
475 struct pack_header hdr;
476 uint32_t nr_remaining = nr_result;
477 time_t last_mtime = 0;
479 if (progress > pack_to_stdout)
480 progress_state = start_progress("Writing objects", nr_result);
481 written_list = xmalloc(nr_objects * sizeof(*written_list));
483 do {
484 unsigned char sha1[20];
485 char *pack_tmp_name = NULL;
487 if (pack_to_stdout) {
488 f = sha1fd_throughput(1, "<stdout>", progress_state);
489 } else {
490 char tmpname[PATH_MAX];
491 int fd;
492 fd = odb_mkstemp(tmpname, sizeof(tmpname),
493 "pack/tmp_pack_XXXXXX");
494 pack_tmp_name = xstrdup(tmpname);
495 f = sha1fd(fd, pack_tmp_name);
496 }
498 hdr.hdr_signature = htonl(PACK_SIGNATURE);
499 hdr.hdr_version = htonl(PACK_VERSION);
500 hdr.hdr_entries = htonl(nr_remaining);
501 sha1write(f, &hdr, sizeof(hdr));
502 offset = sizeof(hdr);
503 nr_written = 0;
504 for (; i < nr_objects; i++) {
505 if (!write_one(f, objects + i, &offset))
506 break;
507 display_progress(progress_state, written);
508 }
510 /*
511 * Did we write the wrong # entries in the header?
512 * If so, rewrite it like in fast-import
513 */
514 if (pack_to_stdout) {
515 sha1close(f, sha1, CSUM_CLOSE);
516 } else if (nr_written == nr_remaining) {
517 sha1close(f, sha1, CSUM_FSYNC);
518 } else {
519 int fd = sha1close(f, sha1, 0);
520 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
521 nr_written, sha1, offset);
522 close(fd);
523 }
525 if (!pack_to_stdout) {
526 mode_t mode = umask(0);
527 struct stat st;
528 const char *idx_tmp_name;
529 char tmpname[PATH_MAX];
531 umask(mode);
532 mode = 0444 & ~mode;
534 idx_tmp_name = write_idx_file(NULL, written_list,
535 nr_written, sha1);
537 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
538 base_name, sha1_to_hex(sha1));
539 free_pack_by_name(tmpname);
540 if (adjust_perm(pack_tmp_name, mode))
541 die_errno("unable to make temporary pack file readable");
542 if (rename(pack_tmp_name, tmpname))
543 die_errno("unable to rename temporary pack file");
545 /*
546 * Packs are runtime accessed in their mtime
547 * order since newer packs are more likely to contain
548 * younger objects. So if we are creating multiple
549 * packs then we should modify the mtime of later ones
550 * to preserve this property.
551 */
552 if (stat(tmpname, &st) < 0) {
553 warning("failed to stat %s: %s",
554 tmpname, strerror(errno));
555 } else if (!last_mtime) {
556 last_mtime = st.st_mtime;
557 } else {
558 struct utimbuf utb;
559 utb.actime = st.st_atime;
560 utb.modtime = --last_mtime;
561 if (utime(tmpname, &utb) < 0)
562 warning("failed utime() on %s: %s",
563 tmpname, strerror(errno));
564 }
566 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
567 base_name, sha1_to_hex(sha1));
568 if (adjust_perm(idx_tmp_name, mode))
569 die_errno("unable to make temporary index file readable");
570 if (rename(idx_tmp_name, tmpname))
571 die_errno("unable to rename temporary index file");
573 free((void *) idx_tmp_name);
574 free(pack_tmp_name);
575 puts(sha1_to_hex(sha1));
576 }
578 /* mark written objects as written to previous pack */
579 for (j = 0; j < nr_written; j++) {
580 written_list[j]->offset = (off_t)-1;
581 }
582 nr_remaining -= nr_written;
583 } while (nr_remaining && i < nr_objects);
585 free(written_list);
586 stop_progress(&progress_state);
587 if (written != nr_result)
588 die("wrote %"PRIu32" objects while expecting %"PRIu32,
589 written, nr_result);
590 /*
591 * We have scanned through [0 ... i). Since we have written
592 * the correct number of objects, the remaining [i ... nr_objects)
593 * items must be either already written (due to out-of-order delta base)
594 * or a preferred base. Count those which are neither and complain if any.
595 */
596 for (j = 0; i < nr_objects; i++) {
597 struct object_entry *e = objects + i;
598 j += !e->idx.offset && !e->preferred_base;
599 }
600 if (j)
601 die("wrote %"PRIu32" objects as expected but %"PRIu32
602 " unwritten", written, j);
603 }
605 static int locate_object_entry_hash(const unsigned char *sha1)
606 {
607 int i;
608 unsigned int ui;
609 memcpy(&ui, sha1, sizeof(unsigned int));
610 i = ui % object_ix_hashsz;
611 while (0 < object_ix[i]) {
612 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
613 return i;
614 if (++i == object_ix_hashsz)
615 i = 0;
616 }
617 return -1 - i;
618 }
620 static struct object_entry *locate_object_entry(const unsigned char *sha1)
621 {
622 int i;
624 if (!object_ix_hashsz)
625 return NULL;
627 i = locate_object_entry_hash(sha1);
628 if (0 <= i)
629 return &objects[object_ix[i]-1];
630 return NULL;
631 }
633 static void rehash_objects(void)
634 {
635 uint32_t i;
636 struct object_entry *oe;
638 object_ix_hashsz = nr_objects * 3;
639 if (object_ix_hashsz < 1024)
640 object_ix_hashsz = 1024;
641 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
642 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
643 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
644 int ix = locate_object_entry_hash(oe->idx.sha1);
645 if (0 <= ix)
646 continue;
647 ix = -1 - ix;
648 object_ix[ix] = i + 1;
649 }
650 }
652 static unsigned name_hash(const char *name)
653 {
654 unsigned c, hash = 0;
656 if (!name)
657 return 0;
659 /*
660 * This effectively just creates a sortable number from the
661 * last sixteen non-whitespace characters. Last characters
662 * count "most", so things that end in ".c" sort together.
663 */
664 while ((c = *name++) != 0) {
665 if (isspace(c))
666 continue;
667 hash = (hash >> 2) + (c << 24);
668 }
669 return hash;
670 }
672 static void setup_delta_attr_check(struct git_attr_check *check)
673 {
674 static struct git_attr *attr_delta;
676 if (!attr_delta)
677 attr_delta = git_attr("delta");
679 check[0].attr = attr_delta;
680 }
682 static int no_try_delta(const char *path)
683 {
684 struct git_attr_check check[1];
686 setup_delta_attr_check(check);
687 if (git_checkattr(path, ARRAY_SIZE(check), check))
688 return 0;
689 if (ATTR_FALSE(check->value))
690 return 1;
691 return 0;
692 }
694 static int add_object_entry(const unsigned char *sha1, enum object_type type,
695 const char *name, int exclude)
696 {
697 struct object_entry *entry;
698 struct packed_git *p, *found_pack = NULL;
699 off_t found_offset = 0;
700 int ix;
701 unsigned hash = name_hash(name);
703 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
704 if (ix >= 0) {
705 if (exclude) {
706 entry = objects + object_ix[ix] - 1;
707 if (!entry->preferred_base)
708 nr_result--;
709 entry->preferred_base = 1;
710 }
711 return 0;
712 }
714 if (!exclude && local && has_loose_object_nonlocal(sha1))
715 return 0;
717 for (p = packed_git; p; p = p->next) {
718 off_t offset = find_pack_entry_one(sha1, p);
719 if (offset) {
720 if (!found_pack) {
721 found_offset = offset;
722 found_pack = p;
723 }
724 if (exclude)
725 break;
726 if (incremental)
727 return 0;
728 if (local && !p->pack_local)
729 return 0;
730 if (ignore_packed_keep && p->pack_local && p->pack_keep)
731 return 0;
732 }
733 }
735 if (nr_objects >= nr_alloc) {
736 nr_alloc = (nr_alloc + 1024) * 3 / 2;
737 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
738 }
740 entry = objects + nr_objects++;
741 memset(entry, 0, sizeof(*entry));
742 hashcpy(entry->idx.sha1, sha1);
743 entry->hash = hash;
744 if (type)
745 entry->type = type;
746 if (exclude)
747 entry->preferred_base = 1;
748 else
749 nr_result++;
750 if (found_pack) {
751 entry->in_pack = found_pack;
752 entry->in_pack_offset = found_offset;
753 }
755 if (object_ix_hashsz * 3 <= nr_objects * 4)
756 rehash_objects();
757 else
758 object_ix[-1 - ix] = nr_objects;
760 display_progress(progress_state, nr_objects);
762 if (name && no_try_delta(name))
763 entry->no_try_delta = 1;
765 return 1;
766 }
768 struct pbase_tree_cache {
769 unsigned char sha1[20];
770 int ref;
771 int temporary;
772 void *tree_data;
773 unsigned long tree_size;
774 };
776 static struct pbase_tree_cache *(pbase_tree_cache[256]);
777 static int pbase_tree_cache_ix(const unsigned char *sha1)
778 {
779 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
780 }
781 static int pbase_tree_cache_ix_incr(int ix)
782 {
783 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
784 }
786 static struct pbase_tree {
787 struct pbase_tree *next;
788 /* This is a phony "cache" entry; we are not
789 * going to evict it nor find it through _get()
790 * mechanism -- this is for the toplevel node that
791 * would almost always change with any commit.
792 */
793 struct pbase_tree_cache pcache;
794 } *pbase_tree;
796 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
797 {
798 struct pbase_tree_cache *ent, *nent;
799 void *data;
800 unsigned long size;
801 enum object_type type;
802 int neigh;
803 int my_ix = pbase_tree_cache_ix(sha1);
804 int available_ix = -1;
806 /* pbase-tree-cache acts as a limited hashtable.
807 * your object will be found at your index or within a few
808 * slots after that slot if it is cached.
809 */
810 for (neigh = 0; neigh < 8; neigh++) {
811 ent = pbase_tree_cache[my_ix];
812 if (ent && !hashcmp(ent->sha1, sha1)) {
813 ent->ref++;
814 return ent;
815 }
816 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
817 ((0 <= available_ix) &&
818 (!ent && pbase_tree_cache[available_ix])))
819 available_ix = my_ix;
820 if (!ent)
821 break;
822 my_ix = pbase_tree_cache_ix_incr(my_ix);
823 }
825 /* Did not find one. Either we got a bogus request or
826 * we need to read and perhaps cache.
827 */
828 data = read_sha1_file(sha1, &type, &size);
829 if (!data)
830 return NULL;
831 if (type != OBJ_TREE) {
832 free(data);
833 return NULL;
834 }
836 /* We need to either cache or return a throwaway copy */
838 if (available_ix < 0)
839 ent = NULL;
840 else {
841 ent = pbase_tree_cache[available_ix];
842 my_ix = available_ix;
843 }
845 if (!ent) {
846 nent = xmalloc(sizeof(*nent));
847 nent->temporary = (available_ix < 0);
848 }
849 else {
850 /* evict and reuse */
851 free(ent->tree_data);
852 nent = ent;
853 }
854 hashcpy(nent->sha1, sha1);
855 nent->tree_data = data;
856 nent->tree_size = size;
857 nent->ref = 1;
858 if (!nent->temporary)
859 pbase_tree_cache[my_ix] = nent;
860 return nent;
861 }
863 static void pbase_tree_put(struct pbase_tree_cache *cache)
864 {
865 if (!cache->temporary) {
866 cache->ref--;
867 return;
868 }
869 free(cache->tree_data);
870 free(cache);
871 }
873 static int name_cmp_len(const char *name)
874 {
875 int i;
876 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
877 ;
878 return i;
879 }
881 static void add_pbase_object(struct tree_desc *tree,
882 const char *name,
883 int cmplen,
884 const char *fullname)
885 {
886 struct name_entry entry;
887 int cmp;
889 while (tree_entry(tree,&entry)) {
890 if (S_ISGITLINK(entry.mode))
891 continue;
892 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
893 memcmp(name, entry.path, cmplen);
894 if (cmp > 0)
895 continue;
896 if (cmp < 0)
897 return;
898 if (name[cmplen] != '/') {
899 add_object_entry(entry.sha1,
900 object_type(entry.mode),
901 fullname, 1);
902 return;
903 }
904 if (S_ISDIR(entry.mode)) {
905 struct tree_desc sub;
906 struct pbase_tree_cache *tree;
907 const char *down = name+cmplen+1;
908 int downlen = name_cmp_len(down);
910 tree = pbase_tree_get(entry.sha1);
911 if (!tree)
912 return;
913 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
915 add_pbase_object(&sub, down, downlen, fullname);
916 pbase_tree_put(tree);
917 }
918 }
919 }
921 static unsigned *done_pbase_paths;
922 static int done_pbase_paths_num;
923 static int done_pbase_paths_alloc;
924 static int done_pbase_path_pos(unsigned hash)
925 {
926 int lo = 0;
927 int hi = done_pbase_paths_num;
928 while (lo < hi) {
929 int mi = (hi + lo) / 2;
930 if (done_pbase_paths[mi] == hash)
931 return mi;
932 if (done_pbase_paths[mi] < hash)
933 hi = mi;
934 else
935 lo = mi + 1;
936 }
937 return -lo-1;
938 }
940 static int check_pbase_path(unsigned hash)
941 {
942 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
943 if (0 <= pos)
944 return 1;
945 pos = -pos - 1;
946 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
947 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
948 done_pbase_paths = xrealloc(done_pbase_paths,
949 done_pbase_paths_alloc *
950 sizeof(unsigned));
951 }
952 done_pbase_paths_num++;
953 if (pos < done_pbase_paths_num)
954 memmove(done_pbase_paths + pos + 1,
955 done_pbase_paths + pos,
956 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
957 done_pbase_paths[pos] = hash;
958 return 0;
959 }
961 static void add_preferred_base_object(const char *name)
962 {
963 struct pbase_tree *it;
964 int cmplen;
965 unsigned hash = name_hash(name);
967 if (!num_preferred_base || check_pbase_path(hash))
968 return;
970 cmplen = name_cmp_len(name);
971 for (it = pbase_tree; it; it = it->next) {
972 if (cmplen == 0) {
973 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
974 }
975 else {
976 struct tree_desc tree;
977 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
978 add_pbase_object(&tree, name, cmplen, name);
979 }
980 }
981 }
983 static void add_preferred_base(unsigned char *sha1)
984 {
985 struct pbase_tree *it;
986 void *data;
987 unsigned long size;
988 unsigned char tree_sha1[20];
990 if (window <= num_preferred_base++)
991 return;
993 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
994 if (!data)
995 return;
997 for (it = pbase_tree; it; it = it->next) {
998 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
999 free(data);
1000 return;
1001 }
1002 }
1004 it = xcalloc(1, sizeof(*it));
1005 it->next = pbase_tree;
1006 pbase_tree = it;
1008 hashcpy(it->pcache.sha1, tree_sha1);
1009 it->pcache.tree_data = data;
1010 it->pcache.tree_size = size;
1011 }
1013 static void cleanup_preferred_base(void)
1014 {
1015 struct pbase_tree *it;
1016 unsigned i;
1018 it = pbase_tree;
1019 pbase_tree = NULL;
1020 while (it) {
1021 struct pbase_tree *this = it;
1022 it = this->next;
1023 free(this->pcache.tree_data);
1024 free(this);
1025 }
1027 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1028 if (!pbase_tree_cache[i])
1029 continue;
1030 free(pbase_tree_cache[i]->tree_data);
1031 free(pbase_tree_cache[i]);
1032 pbase_tree_cache[i] = NULL;
1033 }
1035 free(done_pbase_paths);
1036 done_pbase_paths = NULL;
1037 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1038 }
1040 static void check_object(struct object_entry *entry)
1041 {
1042 if (entry->in_pack) {
1043 struct packed_git *p = entry->in_pack;
1044 struct pack_window *w_curs = NULL;
1045 const unsigned char *base_ref = NULL;
1046 struct object_entry *base_entry;
1047 unsigned long used, used_0;
1048 unsigned int avail;
1049 off_t ofs;
1050 unsigned char *buf, c;
1052 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1054 /*
1055 * We want in_pack_type even if we do not reuse delta
1056 * since non-delta representations could still be reused.
1057 */
1058 used = unpack_object_header_buffer(buf, avail,
1059 &entry->in_pack_type,
1060 &entry->size);
1061 if (used == 0)
1062 goto give_up;
1064 /*
1065 * Determine if this is a delta and if so whether we can
1066 * reuse it or not. Otherwise let's find out as cheaply as
1067 * possible what the actual type and size for this object is.
1068 */
1069 switch (entry->in_pack_type) {
1070 default:
1071 /* Not a delta hence we've already got all we need. */
1072 entry->type = entry->in_pack_type;
1073 entry->in_pack_header_size = used;
1074 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1075 goto give_up;
1076 unuse_pack(&w_curs);
1077 return;
1078 case OBJ_REF_DELTA:
1079 if (reuse_delta && !entry->preferred_base)
1080 base_ref = use_pack(p, &w_curs,
1081 entry->in_pack_offset + used, NULL);
1082 entry->in_pack_header_size = used + 20;
1083 break;
1084 case OBJ_OFS_DELTA:
1085 buf = use_pack(p, &w_curs,
1086 entry->in_pack_offset + used, NULL);
1087 used_0 = 0;
1088 c = buf[used_0++];
1089 ofs = c & 127;
1090 while (c & 128) {
1091 ofs += 1;
1092 if (!ofs || MSB(ofs, 7)) {
1093 error("delta base offset overflow in pack for %s",
1094 sha1_to_hex(entry->idx.sha1));
1095 goto give_up;
1096 }
1097 c = buf[used_0++];
1098 ofs = (ofs << 7) + (c & 127);
1099 }
1100 ofs = entry->in_pack_offset - ofs;
1101 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1102 error("delta base offset out of bound for %s",
1103 sha1_to_hex(entry->idx.sha1));
1104 goto give_up;
1105 }
1106 if (reuse_delta && !entry->preferred_base) {
1107 struct revindex_entry *revidx;
1108 revidx = find_pack_revindex(p, ofs);
1109 if (!revidx)
1110 goto give_up;
1111 base_ref = nth_packed_object_sha1(p, revidx->nr);
1112 }
1113 entry->in_pack_header_size = used + used_0;
1114 break;
1115 }
1117 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1118 /*
1119 * If base_ref was set above that means we wish to
1120 * reuse delta data, and we even found that base
1121 * in the list of objects we want to pack. Goodie!
1122 *
1123 * Depth value does not matter - find_deltas() will
1124 * never consider reused delta as the base object to
1125 * deltify other objects against, in order to avoid
1126 * circular deltas.
1127 */
1128 entry->type = entry->in_pack_type;
1129 entry->delta = base_entry;
1130 entry->delta_size = entry->size;
1131 entry->delta_sibling = base_entry->delta_child;
1132 base_entry->delta_child = entry;
1133 unuse_pack(&w_curs);
1134 return;
1135 }
1137 if (entry->type) {
1138 /*
1139 * This must be a delta and we already know what the
1140 * final object type is. Let's extract the actual
1141 * object size from the delta header.
1142 */
1143 entry->size = get_size_from_delta(p, &w_curs,
1144 entry->in_pack_offset + entry->in_pack_header_size);
1145 if (entry->size == 0)
1146 goto give_up;
1147 unuse_pack(&w_curs);
1148 return;
1149 }
1151 /*
1152 * No choice but to fall back to the recursive delta walk
1153 * with sha1_object_info() to find about the object type
1154 * at this point...
1155 */
1156 give_up:
1157 unuse_pack(&w_curs);
1158 }
1160 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1161 /*
1162 * The error condition is checked in prepare_pack(). This is
1163 * to permit a missing preferred base object to be ignored
1164 * as a preferred base. Doing so can result in a larger
1165 * pack file, but the transfer will still take place.
1166 */
1167 }
1169 static int pack_offset_sort(const void *_a, const void *_b)
1170 {
1171 const struct object_entry *a = *(struct object_entry **)_a;
1172 const struct object_entry *b = *(struct object_entry **)_b;
1174 /* avoid filesystem trashing with loose objects */
1175 if (!a->in_pack && !b->in_pack)
1176 return hashcmp(a->idx.sha1, b->idx.sha1);
1178 if (a->in_pack < b->in_pack)
1179 return -1;
1180 if (a->in_pack > b->in_pack)
1181 return 1;
1182 return a->in_pack_offset < b->in_pack_offset ? -1 :
1183 (a->in_pack_offset > b->in_pack_offset);
1184 }
1186 static void get_object_details(void)
1187 {
1188 uint32_t i;
1189 struct object_entry **sorted_by_offset;
1191 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1192 for (i = 0; i < nr_objects; i++)
1193 sorted_by_offset[i] = objects + i;
1194 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1196 for (i = 0; i < nr_objects; i++)
1197 check_object(sorted_by_offset[i]);
1199 free(sorted_by_offset);
1200 }
1202 /*
1203 * We search for deltas in a list sorted by type, by filename hash, and then
1204 * by size, so that we see progressively smaller and smaller files.
1205 * That's because we prefer deltas to be from the bigger file
1206 * to the smaller -- deletes are potentially cheaper, but perhaps
1207 * more importantly, the bigger file is likely the more recent
1208 * one. The deepest deltas are therefore the oldest objects which are
1209 * less susceptible to be accessed often.
1210 */
1211 static int type_size_sort(const void *_a, const void *_b)
1212 {
1213 const struct object_entry *a = *(struct object_entry **)_a;
1214 const struct object_entry *b = *(struct object_entry **)_b;
1216 if (a->type > b->type)
1217 return -1;
1218 if (a->type < b->type)
1219 return 1;
1220 if (a->hash > b->hash)
1221 return -1;
1222 if (a->hash < b->hash)
1223 return 1;
1224 if (a->preferred_base > b->preferred_base)
1225 return -1;
1226 if (a->preferred_base < b->preferred_base)
1227 return 1;
1228 if (a->size > b->size)
1229 return -1;
1230 if (a->size < b->size)
1231 return 1;
1232 return a < b ? -1 : (a > b); /* newest first */
1233 }
1235 struct unpacked {
1236 struct object_entry *entry;
1237 void *data;
1238 struct delta_index *index;
1239 unsigned depth;
1240 };
1242 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1243 unsigned long delta_size)
1244 {
1245 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1246 return 0;
1248 if (delta_size < cache_max_small_delta_size)
1249 return 1;
1251 /* cache delta, if objects are large enough compared to delta size */
1252 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1253 return 1;
1255 return 0;
1256 }
1258 #ifdef THREADED_DELTA_SEARCH
1260 static pthread_mutex_t read_mutex;
1261 #define read_lock() pthread_mutex_lock(&read_mutex)
1262 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1264 static pthread_mutex_t cache_mutex;
1265 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1266 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1268 static pthread_mutex_t progress_mutex;
1269 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1270 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1272 #else
1274 #define read_lock() (void)0
1275 #define read_unlock() (void)0
1276 #define cache_lock() (void)0
1277 #define cache_unlock() (void)0
1278 #define progress_lock() (void)0
1279 #define progress_unlock() (void)0
1281 #endif
1283 static int try_delta(struct unpacked *trg, struct unpacked *src,
1284 unsigned max_depth, unsigned long *mem_usage)
1285 {
1286 struct object_entry *trg_entry = trg->entry;
1287 struct object_entry *src_entry = src->entry;
1288 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1289 unsigned ref_depth;
1290 enum object_type type;
1291 void *delta_buf;
1293 /* Don't bother doing diffs between different types */
1294 if (trg_entry->type != src_entry->type)
1295 return -1;
1297 /*
1298 * We do not bother to try a delta that we discarded
1299 * on an earlier try, but only when reusing delta data.
1300 */
1301 if (reuse_delta && trg_entry->in_pack &&
1302 trg_entry->in_pack == src_entry->in_pack &&
1303 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1304 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1305 return 0;
1307 /* Let's not bust the allowed depth. */
1308 if (src->depth >= max_depth)
1309 return 0;
1311 /* Now some size filtering heuristics. */
1312 trg_size = trg_entry->size;
1313 if (!trg_entry->delta) {
1314 max_size = trg_size/2 - 20;
1315 ref_depth = 1;
1316 } else {
1317 max_size = trg_entry->delta_size;
1318 ref_depth = trg->depth;
1319 }
1320 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1321 (max_depth - ref_depth + 1);
1322 if (max_size == 0)
1323 return 0;
1324 src_size = src_entry->size;
1325 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1326 if (sizediff >= max_size)
1327 return 0;
1328 if (trg_size < src_size / 32)
1329 return 0;
1331 /* Load data if not already done */
1332 if (!trg->data) {
1333 read_lock();
1334 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1335 read_unlock();
1336 if (!trg->data)
1337 die("object %s cannot be read",
1338 sha1_to_hex(trg_entry->idx.sha1));
1339 if (sz != trg_size)
1340 die("object %s inconsistent object length (%lu vs %lu)",
1341 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1342 *mem_usage += sz;
1343 }
1344 if (!src->data) {
1345 read_lock();
1346 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1347 read_unlock();
1348 if (!src->data)
1349 die("object %s cannot be read",
1350 sha1_to_hex(src_entry->idx.sha1));
1351 if (sz != src_size)
1352 die("object %s inconsistent object length (%lu vs %lu)",
1353 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1354 *mem_usage += sz;
1355 }
1356 if (!src->index) {
1357 src->index = create_delta_index(src->data, src_size);
1358 if (!src->index) {
1359 static int warned = 0;
1360 if (!warned++)
1361 warning("suboptimal pack - out of memory");
1362 return 0;
1363 }
1364 *mem_usage += sizeof_delta_index(src->index);
1365 }
1367 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1368 if (!delta_buf)
1369 return 0;
1371 if (trg_entry->delta) {
1372 /* Prefer only shallower same-sized deltas. */
1373 if (delta_size == trg_entry->delta_size &&
1374 src->depth + 1 >= trg->depth) {
1375 free(delta_buf);
1376 return 0;
1377 }
1378 }
1380 /*
1381 * Handle memory allocation outside of the cache
1382 * accounting lock. Compiler will optimize the strangeness
1383 * away when THREADED_DELTA_SEARCH is not defined.
1384 */
1385 free(trg_entry->delta_data);
1386 cache_lock();
1387 if (trg_entry->delta_data) {
1388 delta_cache_size -= trg_entry->delta_size;
1389 trg_entry->delta_data = NULL;
1390 }
1391 if (delta_cacheable(src_size, trg_size, delta_size)) {
1392 delta_cache_size += delta_size;
1393 cache_unlock();
1394 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1395 } else {
1396 cache_unlock();
1397 free(delta_buf);
1398 }
1400 trg_entry->delta = src_entry;
1401 trg_entry->delta_size = delta_size;
1402 trg->depth = src->depth + 1;
1404 return 1;
1405 }
1407 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1408 {
1409 struct object_entry *child = me->delta_child;
1410 unsigned int m = n;
1411 while (child) {
1412 unsigned int c = check_delta_limit(child, n + 1);
1413 if (m < c)
1414 m = c;
1415 child = child->delta_sibling;
1416 }
1417 return m;
1418 }
1420 static unsigned long free_unpacked(struct unpacked *n)
1421 {
1422 unsigned long freed_mem = sizeof_delta_index(n->index);
1423 free_delta_index(n->index);
1424 n->index = NULL;
1425 if (n->data) {
1426 freed_mem += n->entry->size;
1427 free(n->data);
1428 n->data = NULL;
1429 }
1430 n->entry = NULL;
1431 n->depth = 0;
1432 return freed_mem;
1433 }
1435 static void find_deltas(struct object_entry **list, unsigned *list_size,
1436 int window, int depth, unsigned *processed)
1437 {
1438 uint32_t i, idx = 0, count = 0;
1439 struct unpacked *array;
1440 unsigned long mem_usage = 0;
1442 array = xcalloc(window, sizeof(struct unpacked));
1444 for (;;) {
1445 struct object_entry *entry;
1446 struct unpacked *n = array + idx;
1447 int j, max_depth, best_base = -1;
1449 progress_lock();
1450 if (!*list_size) {
1451 progress_unlock();
1452 break;
1453 }
1454 entry = *list++;
1455 (*list_size)--;
1456 if (!entry->preferred_base) {
1457 (*processed)++;
1458 display_progress(progress_state, *processed);
1459 }
1460 progress_unlock();
1462 mem_usage -= free_unpacked(n);
1463 n->entry = entry;
1465 while (window_memory_limit &&
1466 mem_usage > window_memory_limit &&
1467 count > 1) {
1468 uint32_t tail = (idx + window - count) % window;
1469 mem_usage -= free_unpacked(array + tail);
1470 count--;
1471 }
1473 /* We do not compute delta to *create* objects we are not
1474 * going to pack.
1475 */
1476 if (entry->preferred_base)
1477 goto next;
1479 /*
1480 * If the current object is at pack edge, take the depth the
1481 * objects that depend on the current object into account
1482 * otherwise they would become too deep.
1483 */
1484 max_depth = depth;
1485 if (entry->delta_child) {
1486 max_depth -= check_delta_limit(entry, 0);
1487 if (max_depth <= 0)
1488 goto next;
1489 }
1491 j = window;
1492 while (--j > 0) {
1493 int ret;
1494 uint32_t other_idx = idx + j;
1495 struct unpacked *m;
1496 if (other_idx >= window)
1497 other_idx -= window;
1498 m = array + other_idx;
1499 if (!m->entry)
1500 break;
1501 ret = try_delta(n, m, max_depth, &mem_usage);
1502 if (ret < 0)
1503 break;
1504 else if (ret > 0)
1505 best_base = other_idx;
1506 }
1508 /*
1509 * If we decided to cache the delta data, then it is best
1510 * to compress it right away. First because we have to do
1511 * it anyway, and doing it here while we're threaded will
1512 * save a lot of time in the non threaded write phase,
1513 * as well as allow for caching more deltas within
1514 * the same cache size limit.
1515 * ...
1516 * But only if not writing to stdout, since in that case
1517 * the network is most likely throttling writes anyway,
1518 * and therefore it is best to go to the write phase ASAP
1519 * instead, as we can afford spending more time compressing
1520 * between writes at that moment.
1521 */
1522 if (entry->delta_data && !pack_to_stdout) {
1523 entry->z_delta_size = do_compress(&entry->delta_data,
1524 entry->delta_size);
1525 cache_lock();
1526 delta_cache_size -= entry->delta_size;
1527 delta_cache_size += entry->z_delta_size;
1528 cache_unlock();
1529 }
1531 /* if we made n a delta, and if n is already at max
1532 * depth, leaving it in the window is pointless. we
1533 * should evict it first.
1534 */
1535 if (entry->delta && max_depth <= n->depth)
1536 continue;
1538 /*
1539 * Move the best delta base up in the window, after the
1540 * currently deltified object, to keep it longer. It will
1541 * be the first base object to be attempted next.
1542 */
1543 if (entry->delta) {
1544 struct unpacked swap = array[best_base];
1545 int dist = (window + idx - best_base) % window;
1546 int dst = best_base;
1547 while (dist--) {
1548 int src = (dst + 1) % window;
1549 array[dst] = array[src];
1550 dst = src;
1551 }
1552 array[dst] = swap;
1553 }
1555 next:
1556 idx++;
1557 if (count + 1 < window)
1558 count++;
1559 if (idx >= window)
1560 idx = 0;
1561 }
1563 for (i = 0; i < window; ++i) {
1564 free_delta_index(array[i].index);
1565 free(array[i].data);
1566 }
1567 free(array);
1568 }
1570 #ifdef THREADED_DELTA_SEARCH
1572 /*
1573 * The main thread waits on the condition that (at least) one of the workers
1574 * has stopped working (which is indicated in the .working member of
1575 * struct thread_params).
1576 * When a work thread has completed its work, it sets .working to 0 and
1577 * signals the main thread and waits on the condition that .data_ready
1578 * becomes 1.
1579 */
1581 struct thread_params {
1582 pthread_t thread;
1583 struct object_entry **list;
1584 unsigned list_size;
1585 unsigned remaining;
1586 int window;
1587 int depth;
1588 int working;
1589 int data_ready;
1590 pthread_mutex_t mutex;
1591 pthread_cond_t cond;
1592 unsigned *processed;
1593 };
1595 static pthread_cond_t progress_cond;
1597 /*
1598 * Mutex and conditional variable can't be statically-initialized on Windows.
1599 */
1600 static void init_threaded_search(void)
1601 {
1602 pthread_mutex_init(&read_mutex, NULL);
1603 pthread_mutex_init(&cache_mutex, NULL);
1604 pthread_mutex_init(&progress_mutex, NULL);
1605 pthread_cond_init(&progress_cond, NULL);
1606 }
1608 static void cleanup_threaded_search(void)
1609 {
1610 pthread_cond_destroy(&progress_cond);
1611 pthread_mutex_destroy(&read_mutex);
1612 pthread_mutex_destroy(&cache_mutex);
1613 pthread_mutex_destroy(&progress_mutex);
1614 }
1616 static void *threaded_find_deltas(void *arg)
1617 {
1618 struct thread_params *me = arg;
1620 while (me->remaining) {
1621 find_deltas(me->list, &me->remaining,
1622 me->window, me->depth, me->processed);
1624 progress_lock();
1625 me->working = 0;
1626 pthread_cond_signal(&progress_cond);
1627 progress_unlock();
1629 /*
1630 * We must not set ->data_ready before we wait on the
1631 * condition because the main thread may have set it to 1
1632 * before we get here. In order to be sure that new
1633 * work is available if we see 1 in ->data_ready, it
1634 * was initialized to 0 before this thread was spawned
1635 * and we reset it to 0 right away.
1636 */
1637 pthread_mutex_lock(&me->mutex);
1638 while (!me->data_ready)
1639 pthread_cond_wait(&me->cond, &me->mutex);
1640 me->data_ready = 0;
1641 pthread_mutex_unlock(&me->mutex);
1642 }
1643 /* leave ->working 1 so that this doesn't get more work assigned */
1644 return NULL;
1645 }
1647 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1648 int window, int depth, unsigned *processed)
1649 {
1650 struct thread_params *p;
1651 int i, ret, active_threads = 0;
1653 init_threaded_search();
1655 if (!delta_search_threads) /* --threads=0 means autodetect */
1656 delta_search_threads = online_cpus();
1657 if (delta_search_threads <= 1) {
1658 find_deltas(list, &list_size, window, depth, processed);
1659 cleanup_threaded_search();
1660 return;
1661 }
1662 if (progress > pack_to_stdout)
1663 fprintf(stderr, "Delta compression using up to %d threads.\n",
1664 delta_search_threads);
1665 p = xcalloc(delta_search_threads, sizeof(*p));
1667 /* Partition the work amongst work threads. */
1668 for (i = 0; i < delta_search_threads; i++) {
1669 unsigned sub_size = list_size / (delta_search_threads - i);
1671 /* don't use too small segments or no deltas will be found */
1672 if (sub_size < 2*window && i+1 < delta_search_threads)
1673 sub_size = 0;
1675 p[i].window = window;
1676 p[i].depth = depth;
1677 p[i].processed = processed;
1678 p[i].working = 1;
1679 p[i].data_ready = 0;
1681 /* try to split chunks on "path" boundaries */
1682 while (sub_size && sub_size < list_size &&
1683 list[sub_size]->hash &&
1684 list[sub_size]->hash == list[sub_size-1]->hash)
1685 sub_size++;
1687 p[i].list = list;
1688 p[i].list_size = sub_size;
1689 p[i].remaining = sub_size;
1691 list += sub_size;
1692 list_size -= sub_size;
1693 }
1695 /* Start work threads. */
1696 for (i = 0; i < delta_search_threads; i++) {
1697 if (!p[i].list_size)
1698 continue;
1699 pthread_mutex_init(&p[i].mutex, NULL);
1700 pthread_cond_init(&p[i].cond, NULL);
1701 ret = pthread_create(&p[i].thread, NULL,
1702 threaded_find_deltas, &p[i]);
1703 if (ret)
1704 die("unable to create thread: %s", strerror(ret));
1705 active_threads++;
1706 }
1708 /*
1709 * Now let's wait for work completion. Each time a thread is done
1710 * with its work, we steal half of the remaining work from the
1711 * thread with the largest number of unprocessed objects and give
1712 * it to that newly idle thread. This ensure good load balancing
1713 * until the remaining object list segments are simply too short
1714 * to be worth splitting anymore.
1715 */
1716 while (active_threads) {
1717 struct thread_params *target = NULL;
1718 struct thread_params *victim = NULL;
1719 unsigned sub_size = 0;
1721 progress_lock();
1722 for (;;) {
1723 for (i = 0; !target && i < delta_search_threads; i++)
1724 if (!p[i].working)
1725 target = &p[i];
1726 if (target)
1727 break;
1728 pthread_cond_wait(&progress_cond, &progress_mutex);
1729 }
1731 for (i = 0; i < delta_search_threads; i++)
1732 if (p[i].remaining > 2*window &&
1733 (!victim || victim->remaining < p[i].remaining))
1734 victim = &p[i];
1735 if (victim) {
1736 sub_size = victim->remaining / 2;
1737 list = victim->list + victim->list_size - sub_size;
1738 while (sub_size && list[0]->hash &&
1739 list[0]->hash == list[-1]->hash) {
1740 list++;
1741 sub_size--;
1742 }
1743 if (!sub_size) {
1744 /*
1745 * It is possible for some "paths" to have
1746 * so many objects that no hash boundary
1747 * might be found. Let's just steal the
1748 * exact half in that case.
1749 */
1750 sub_size = victim->remaining / 2;
1751 list -= sub_size;
1752 }
1753 target->list = list;
1754 victim->list_size -= sub_size;
1755 victim->remaining -= sub_size;
1756 }
1757 target->list_size = sub_size;
1758 target->remaining = sub_size;
1759 target->working = 1;
1760 progress_unlock();
1762 pthread_mutex_lock(&target->mutex);
1763 target->data_ready = 1;
1764 pthread_cond_signal(&target->cond);
1765 pthread_mutex_unlock(&target->mutex);
1767 if (!sub_size) {
1768 pthread_join(target->thread, NULL);
1769 pthread_cond_destroy(&target->cond);
1770 pthread_mutex_destroy(&target->mutex);
1771 active_threads--;
1772 }
1773 }
1774 cleanup_threaded_search();
1775 free(p);
1776 }
1778 #else
1779 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1780 #endif
1782 static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1783 {
1784 unsigned char peeled[20];
1786 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1787 !peel_ref(path, peeled) && /* peelable? */
1788 !is_null_sha1(peeled) && /* annotated tag? */
1789 locate_object_entry(peeled)) /* object packed? */
1790 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1791 return 0;
1792 }
1794 static void prepare_pack(int window, int depth)
1795 {
1796 struct object_entry **delta_list;
1797 uint32_t i, nr_deltas;
1798 unsigned n;
1800 get_object_details();
1802 /*
1803 * If we're locally repacking then we need to be doubly careful
1804 * from now on in order to make sure no stealth corruption gets
1805 * propagated to the new pack. Clients receiving streamed packs
1806 * should validate everything they get anyway so no need to incur
1807 * the additional cost here in that case.
1808 */
1809 if (!pack_to_stdout)
1810 do_check_packed_object_crc = 1;
1812 if (!nr_objects || !window || !depth)
1813 return;
1815 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1816 nr_deltas = n = 0;
1818 for (i = 0; i < nr_objects; i++) {
1819 struct object_entry *entry = objects + i;
1821 if (entry->delta)
1822 /* This happens if we decided to reuse existing
1823 * delta from a pack. "reuse_delta &&" is implied.
1824 */
1825 continue;
1827 if (entry->size < 50)
1828 continue;
1830 if (entry->no_try_delta)
1831 continue;
1833 if (!entry->preferred_base) {
1834 nr_deltas++;
1835 if (entry->type < 0)
1836 die("unable to get type of object %s",
1837 sha1_to_hex(entry->idx.sha1));
1838 } else {
1839 if (entry->type < 0) {
1840 /*
1841 * This object is not found, but we
1842 * don't have to include it anyway.
1843 */
1844 continue;
1845 }
1846 }
1848 delta_list[n++] = entry;
1849 }
1851 if (nr_deltas && n > 1) {
1852 unsigned nr_done = 0;
1853 if (progress)
1854 progress_state = start_progress("Compressing objects",
1855 nr_deltas);
1856 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
1857 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
1858 stop_progress(&progress_state);
1859 if (nr_done != nr_deltas)
1860 die("inconsistency with delta count");
1861 }
1862 free(delta_list);
1863 }
1865 static int git_pack_config(const char *k, const char *v, void *cb)
1866 {
1867 if (!strcmp(k, "pack.window")) {
1868 window = git_config_int(k, v);
1869 return 0;
1870 }
1871 if (!strcmp(k, "pack.windowmemory")) {
1872 window_memory_limit = git_config_ulong(k, v);
1873 return 0;
1874 }
1875 if (!strcmp(k, "pack.depth")) {
1876 depth = git_config_int(k, v);
1877 return 0;
1878 }
1879 if (!strcmp(k, "pack.compression")) {
1880 int level = git_config_int(k, v);
1881 if (level == -1)
1882 level = Z_DEFAULT_COMPRESSION;
1883 else if (level < 0 || level > Z_BEST_COMPRESSION)
1884 die("bad pack compression level %d", level);
1885 pack_compression_level = level;
1886 pack_compression_seen = 1;
1887 return 0;
1888 }
1889 if (!strcmp(k, "pack.deltacachesize")) {
1890 max_delta_cache_size = git_config_int(k, v);
1891 return 0;
1892 }
1893 if (!strcmp(k, "pack.deltacachelimit")) {
1894 cache_max_small_delta_size = git_config_int(k, v);
1895 return 0;
1896 }
1897 if (!strcmp(k, "pack.threads")) {
1898 delta_search_threads = git_config_int(k, v);
1899 if (delta_search_threads < 0)
1900 die("invalid number of threads specified (%d)",
1901 delta_search_threads);
1902 #ifndef THREADED_DELTA_SEARCH
1903 if (delta_search_threads != 1)
1904 warning("no threads support, ignoring %s", k);
1905 #endif
1906 return 0;
1907 }
1908 if (!strcmp(k, "pack.indexversion")) {
1909 pack_idx_default_version = git_config_int(k, v);
1910 if (pack_idx_default_version > 2)
1911 die("bad pack.indexversion=%"PRIu32,
1912 pack_idx_default_version);
1913 return 0;
1914 }
1915 if (!strcmp(k, "pack.packsizelimit")) {
1916 pack_size_limit_cfg = git_config_ulong(k, v);
1917 return 0;
1918 }
1919 return git_default_config(k, v, cb);
1920 }
1922 static void read_object_list_from_stdin(void)
1923 {
1924 char line[40 + 1 + PATH_MAX + 2];
1925 unsigned char sha1[20];
1927 for (;;) {
1928 if (!fgets(line, sizeof(line), stdin)) {
1929 if (feof(stdin))
1930 break;
1931 if (!ferror(stdin))
1932 die("fgets returned NULL, not EOF, not error!");
1933 if (errno != EINTR)
1934 die_errno("fgets");
1935 clearerr(stdin);
1936 continue;
1937 }
1938 if (line[0] == '-') {
1939 if (get_sha1_hex(line+1, sha1))
1940 die("expected edge sha1, got garbage:\n %s",
1941 line);
1942 add_preferred_base(sha1);
1943 continue;
1944 }
1945 if (get_sha1_hex(line, sha1))
1946 die("expected sha1, got garbage:\n %s", line);
1948 add_preferred_base_object(line+41);
1949 add_object_entry(sha1, 0, line+41, 0);
1950 }
1951 }
1953 #define OBJECT_ADDED (1u<<20)
1955 static void show_commit(struct commit *commit, void *data)
1956 {
1957 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
1958 commit->object.flags |= OBJECT_ADDED;
1959 }
1961 static void show_object(struct object *obj, const struct name_path *path, const char *last)
1962 {
1963 char *name = path_name(path, last);
1965 add_preferred_base_object(name);
1966 add_object_entry(obj->sha1, obj->type, name, 0);
1967 obj->flags |= OBJECT_ADDED;
1969 /*
1970 * We will have generated the hash from the name,
1971 * but not saved a pointer to it - we can free it
1972 */
1973 free((char *)name);
1974 }
1976 static void show_edge(struct commit *commit)
1977 {
1978 add_preferred_base(commit->object.sha1);
1979 }
1981 struct in_pack_object {
1982 off_t offset;
1983 struct object *object;
1984 };
1986 struct in_pack {
1987 int alloc;
1988 int nr;
1989 struct in_pack_object *array;
1990 };
1992 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
1993 {
1994 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
1995 in_pack->array[in_pack->nr].object = object;
1996 in_pack->nr++;
1997 }
1999 /*
2000 * Compare the objects in the offset order, in order to emulate the
2001 * "git rev-list --objects" output that produced the pack originally.
2002 */
2003 static int ofscmp(const void *a_, const void *b_)
2004 {
2005 struct in_pack_object *a = (struct in_pack_object *)a_;
2006 struct in_pack_object *b = (struct in_pack_object *)b_;
2008 if (a->offset < b->offset)
2009 return -1;
2010 else if (a->offset > b->offset)
2011 return 1;
2012 else
2013 return hashcmp(a->object->sha1, b->object->sha1);
2014 }
2016 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2017 {
2018 struct packed_git *p;
2019 struct in_pack in_pack;
2020 uint32_t i;
2022 memset(&in_pack, 0, sizeof(in_pack));
2024 for (p = packed_git; p; p = p->next) {
2025 const unsigned char *sha1;
2026 struct object *o;
2028 if (!p->pack_local || p->pack_keep)
2029 continue;
2030 if (open_pack_index(p))
2031 die("cannot open pack index");
2033 ALLOC_GROW(in_pack.array,
2034 in_pack.nr + p->num_objects,
2035 in_pack.alloc);
2037 for (i = 0; i < p->num_objects; i++) {
2038 sha1 = nth_packed_object_sha1(p, i);
2039 o = lookup_unknown_object(sha1);
2040 if (!(o->flags & OBJECT_ADDED))
2041 mark_in_pack_object(o, p, &in_pack);
2042 o->flags |= OBJECT_ADDED;
2043 }
2044 }
2046 if (in_pack.nr) {
2047 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2048 ofscmp);
2049 for (i = 0; i < in_pack.nr; i++) {
2050 struct object *o = in_pack.array[i].object;
2051 add_object_entry(o->sha1, o->type, "", 0);
2052 }
2053 }
2054 free(in_pack.array);
2055 }
2057 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2058 {
2059 static struct packed_git *last_found = (void *)1;
2060 struct packed_git *p;
2062 p = (last_found != (void *)1) ? last_found : packed_git;
2064 while (p) {
2065 if ((!p->pack_local || p->pack_keep) &&
2066 find_pack_entry_one(sha1, p)) {
2067 last_found = p;
2068 return 1;
2069 }
2070 if (p == last_found)
2071 p = packed_git;
2072 else
2073 p = p->next;
2074 if (p == last_found)
2075 p = p->next;
2076 }
2077 return 0;
2078 }
2080 static void loosen_unused_packed_objects(struct rev_info *revs)
2081 {
2082 struct packed_git *p;
2083 uint32_t i;
2084 const unsigned char *sha1;
2086 for (p = packed_git; p; p = p->next) {
2087 if (!p->pack_local || p->pack_keep)
2088 continue;
2090 if (open_pack_index(p))
2091 die("cannot open pack index");
2093 for (i = 0; i < p->num_objects; i++) {
2094 sha1 = nth_packed_object_sha1(p, i);
2095 if (!locate_object_entry(sha1) &&
2096 !has_sha1_pack_kept_or_nonlocal(sha1))
2097 if (force_object_loose(sha1, p->mtime))
2098 die("unable to force loose object");
2099 }
2100 }
2101 }
2103 static void get_object_list(int ac, const char **av)
2104 {
2105 struct rev_info revs;
2106 char line[1000];
2107 int flags = 0;
2109 init_revisions(&revs, NULL);
2110 save_commit_buffer = 0;
2111 setup_revisions(ac, av, &revs, NULL);
2113 while (fgets(line, sizeof(line), stdin) != NULL) {
2114 int len = strlen(line);
2115 if (len && line[len - 1] == '\n')
2116 line[--len] = 0;
2117 if (!len)
2118 break;
2119 if (*line == '-') {
2120 if (!strcmp(line, "--not")) {
2121 flags ^= UNINTERESTING;
2122 continue;
2123 }
2124 die("not a rev '%s'", line);
2125 }
2126 if (handle_revision_arg(line, &revs, flags, 1))
2127 die("bad revision '%s'", line);
2128 }
2130 if (prepare_revision_walk(&revs))
2131 die("revision walk setup failed");
2132 mark_edges_uninteresting(revs.commits, &revs, show_edge);
2133 traverse_commit_list(&revs, show_commit, show_object, NULL);
2135 if (keep_unreachable)
2136 add_objects_in_unpacked_packs(&revs);
2137 if (unpack_unreachable)
2138 loosen_unused_packed_objects(&revs);
2139 }
2141 static int adjust_perm(const char *path, mode_t mode)
2142 {
2143 if (chmod(path, mode))
2144 return -1;
2145 return adjust_shared_perm(path);
2146 }
2148 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2149 {
2150 int use_internal_rev_list = 0;
2151 int thin = 0;
2152 int all_progress_implied = 0;
2153 uint32_t i;
2154 const char **rp_av;
2155 int rp_ac_alloc = 64;
2156 int rp_ac;
2158 read_replace_refs = 0;
2160 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2162 rp_av[0] = "pack-objects";
2163 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2164 rp_ac = 2;
2166 git_config(git_pack_config, NULL);
2167 if (!pack_compression_seen && core_compression_seen)
2168 pack_compression_level = core_compression_level;
2170 progress = isatty(2);
2171 for (i = 1; i < argc; i++) {
2172 const char *arg = argv[i];
2174 if (*arg != '-')
2175 break;
2177 if (!strcmp("--non-empty", arg)) {
2178 non_empty = 1;
2179 continue;
2180 }
2181 if (!strcmp("--local", arg)) {
2182 local = 1;
2183 continue;
2184 }
2185 if (!strcmp("--incremental", arg)) {
2186 incremental = 1;
2187 continue;
2188 }
2189 if (!strcmp("--honor-pack-keep", arg)) {
2190 ignore_packed_keep = 1;
2191 continue;
2192 }
2193 if (!prefixcmp(arg, "--compression=")) {
2194 char *end;
2195 int level = strtoul(arg+14, &end, 0);
2196 if (!arg[14] || *end)
2197 usage(pack_usage);
2198 if (level == -1)
2199 level = Z_DEFAULT_COMPRESSION;
2200 else if (level < 0 || level > Z_BEST_COMPRESSION)
2201 die("bad pack compression level %d", level);
2202 pack_compression_level = level;
2203 continue;
2204 }
2205 if (!prefixcmp(arg, "--max-pack-size=")) {
2206 char *end;
2207 pack_size_limit_cfg = 0;
2208 pack_size_limit = strtoul(arg+16, &end, 0) * 1024 * 1024;
2209 if (!arg[16] || *end)
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 #ifndef THREADED_DELTA_SEARCH
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;
2350 if (pack_to_stdout && pack_size_limit)
2351 die("--max-pack-size cannot be used to build a pack for transfer.");
2353 if (!pack_to_stdout && thin)
2354 die("--thin cannot be used to build an indexable pack.");
2356 if (keep_unreachable && unpack_unreachable)
2357 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2359 if (progress && all_progress_implied)
2360 progress = 2;
2362 prepare_packed_git();
2364 if (progress)
2365 progress_state = start_progress("Counting objects", 0);
2366 if (!use_internal_rev_list)
2367 read_object_list_from_stdin();
2368 else {
2369 rp_av[rp_ac] = NULL;
2370 get_object_list(rp_ac, rp_av);
2371 }
2372 cleanup_preferred_base();
2373 if (include_tag && nr_result)
2374 for_each_ref(add_ref_tag, NULL);
2375 stop_progress(&progress_state);
2377 if (non_empty && !nr_result)
2378 return 0;
2379 if (nr_result)
2380 prepare_pack(window, depth);
2381 write_pack_file();
2382 if (progress)
2383 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2384 " reused %"PRIu32" (delta %"PRIu32")\n",
2385 written, written_delta, reused, reused_delta);
2386 return 0;
2387 }