19668d3517e42df173b38b3950ec54586a1162f5
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 "pack-objects.h"
18#include "progress.h"
19#include "refs.h"
20#include "streaming.h"
21#include "thread-utils.h"
22#include "pack-bitmap.h"
23#include "reachable.h"
24#include "sha1-array.h"
25#include "argv-array.h"
26static const char *pack_usage[] = {
28 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
29 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
30 NULL
31};
32/*
34 * Objects we are going to pack are collected in the `to_pack` structure.
35 * It contains an array (dynamically expanded) of the object data, and a map
36 * that can resolve SHA1s to their position in the array.
37 */
38static struct packing_data to_pack;
39static struct pack_idx_entry **written_list;
41static uint32_t nr_result, nr_written;
42static int non_empty;
44static int reuse_delta = 1, reuse_object = 1;
45static int keep_unreachable, unpack_unreachable, include_tag;
46static unsigned long unpack_unreachable_expiration;
47static int pack_loose_unreachable;
48static int local;
49static int have_non_local_packs;
50static int incremental;
51static int ignore_packed_keep;
52static int allow_ofs_delta;
53static struct pack_idx_option pack_idx_opts;
54static const char *base_name;
55static int progress = 1;
56static int window = 10;
57static unsigned long pack_size_limit;
58static int depth = 50;
59static int delta_search_threads;
60static int pack_to_stdout;
61static int num_preferred_base;
62static struct progress *progress_state;
63static int pack_compression_level = Z_DEFAULT_COMPRESSION;
64static int pack_compression_seen;
65static struct packed_git *reuse_packfile;
67static uint32_t reuse_packfile_objects;
68static off_t reuse_packfile_offset;
69static int use_bitmap_index = 1;
71static int write_bitmap_index;
72static uint16_t write_bitmap_options;
73static unsigned long delta_cache_size = 0;
75static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
76static unsigned long cache_max_small_delta_size = 1000;
77static unsigned long window_memory_limit = 0;
79/*
81 * stats
82 */
83static uint32_t written, written_delta;
84static uint32_t reused, reused_delta;
85/*
87 * Indexed commits
88 */
89static struct commit **indexed_commits;
90static unsigned int indexed_commits_nr;
91static unsigned int indexed_commits_alloc;
92static void index_commit_for_bitmap(struct commit *commit)
94{
95 if (indexed_commits_nr >= indexed_commits_alloc) {
96 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
97 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
98 }
99 indexed_commits[indexed_commits_nr++] = commit;
101}
102static void *get_delta(struct object_entry *entry)
104{
105 unsigned long size, base_size, delta_size;
106 void *buf, *base_buf, *delta_buf;
107 enum object_type type;
108 buf = read_sha1_file(entry->idx.sha1, &type, &size);
110 if (!buf)
111 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
112 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
113 if (!base_buf)
114 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
115 delta_buf = diff_delta(base_buf, base_size,
116 buf, size, &delta_size, 0);
117 if (!delta_buf || delta_size != entry->delta_size)
118 die("delta size changed");
119 free(buf);
120 free(base_buf);
121 return delta_buf;
122}
123static unsigned long do_compress(void **pptr, unsigned long size)
125{
126 git_zstream stream;
127 void *in, *out;
128 unsigned long maxsize;
129 git_deflate_init(&stream, pack_compression_level);
131 maxsize = git_deflate_bound(&stream, size);
132 in = *pptr;
134 out = xmalloc(maxsize);
135 *pptr = out;
136 stream.next_in = in;
138 stream.avail_in = size;
139 stream.next_out = out;
140 stream.avail_out = maxsize;
141 while (git_deflate(&stream, Z_FINISH) == Z_OK)
142 ; /* nothing */
143 git_deflate_end(&stream);
144 free(in);
146 return stream.total_out;
147}
148static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f,
150 const unsigned char *sha1)
151{
152 git_zstream stream;
153 unsigned char ibuf[1024 * 16];
154 unsigned char obuf[1024 * 16];
155 unsigned long olen = 0;
156 git_deflate_init(&stream, pack_compression_level);
158 for (;;) {
160 ssize_t readlen;
161 int zret = Z_OK;
162 readlen = read_istream(st, ibuf, sizeof(ibuf));
163 if (readlen == -1)
164 die(_("unable to read %s"), sha1_to_hex(sha1));
165 stream.next_in = ibuf;
167 stream.avail_in = readlen;
168 while ((stream.avail_in || readlen == 0) &&
169 (zret == Z_OK || zret == Z_BUF_ERROR)) {
170 stream.next_out = obuf;
171 stream.avail_out = sizeof(obuf);
172 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
173 sha1write(f, obuf, stream.next_out - obuf);
174 olen += stream.next_out - obuf;
175 }
176 if (stream.avail_in)
177 die(_("deflate error (%d)"), zret);
178 if (readlen == 0) {
179 if (zret != Z_STREAM_END)
180 die(_("deflate error (%d)"), zret);
181 break;
182 }
183 }
184 git_deflate_end(&stream);
185 return olen;
186}
187/*
189 * we are going to reuse the existing object data as is. make
190 * sure it is not corrupt.
191 */
192static int check_pack_inflate(struct packed_git *p,
193 struct pack_window **w_curs,
194 off_t offset,
195 off_t len,
196 unsigned long expect)
197{
198 git_zstream stream;
199 unsigned char fakebuf[4096], *in;
200 int st;
201 memset(&stream, 0, sizeof(stream));
203 git_inflate_init(&stream);
204 do {
205 in = use_pack(p, w_curs, offset, &stream.avail_in);
206 stream.next_in = in;
207 stream.next_out = fakebuf;
208 stream.avail_out = sizeof(fakebuf);
209 st = git_inflate(&stream, Z_FINISH);
210 offset += stream.next_in - in;
211 } while (st == Z_OK || st == Z_BUF_ERROR);
212 git_inflate_end(&stream);
213 return (st == Z_STREAM_END &&
214 stream.total_out == expect &&
215 stream.total_in == len) ? 0 : -1;
216}
217static void copy_pack_data(struct sha1file *f,
219 struct packed_git *p,
220 struct pack_window **w_curs,
221 off_t offset,
222 off_t len)
223{
224 unsigned char *in;
225 unsigned long avail;
226 while (len) {
228 in = use_pack(p, w_curs, offset, &avail);
229 if (avail > len)
230 avail = (unsigned long)len;
231 sha1write(f, in, avail);
232 offset += avail;
233 len -= avail;
234 }
235}
236/* Return 0 if we will bust the pack-size limit */
238static unsigned long write_no_reuse_object(struct sha1file *f, struct object_entry *entry,
239 unsigned long limit, int usable_delta)
240{
241 unsigned long size, datalen;
242 unsigned char header[10], dheader[10];
243 unsigned hdrlen;
244 enum object_type type;
245 void *buf;
246 struct git_istream *st = NULL;
247 if (!usable_delta) {
249 if (entry->type == OBJ_BLOB &&
250 entry->size > big_file_threshold &&
251 (st = open_istream(entry->idx.sha1, &type, &size, NULL)) != NULL)
252 buf = NULL;
253 else {
254 buf = read_sha1_file(entry->idx.sha1, &type, &size);
255 if (!buf)
256 die(_("unable to read %s"), sha1_to_hex(entry->idx.sha1));
257 }
258 /*
259 * make sure no cached delta data remains from a
260 * previous attempt before a pack split occurred.
261 */
262 free(entry->delta_data);
263 entry->delta_data = NULL;
264 entry->z_delta_size = 0;
265 } else if (entry->delta_data) {
266 size = entry->delta_size;
267 buf = entry->delta_data;
268 entry->delta_data = NULL;
269 type = (allow_ofs_delta && entry->delta->idx.offset) ?
270 OBJ_OFS_DELTA : OBJ_REF_DELTA;
271 } else {
272 buf = get_delta(entry);
273 size = entry->delta_size;
274 type = (allow_ofs_delta && entry->delta->idx.offset) ?
275 OBJ_OFS_DELTA : OBJ_REF_DELTA;
276 }
277 if (st) /* large blob case, just assume we don't compress well */
279 datalen = size;
280 else if (entry->z_delta_size)
281 datalen = entry->z_delta_size;
282 else
283 datalen = do_compress(&buf, size);
284 /*
286 * The object header is a byte of 'type' followed by zero or
287 * more bytes of length.
288 */
289 hdrlen = encode_in_pack_object_header(type, size, header);
290 if (type == OBJ_OFS_DELTA) {
292 /*
293 * Deltas with relative base contain an additional
294 * encoding of the relative offset for the delta
295 * base from this object's position in the pack.
296 */
297 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
298 unsigned pos = sizeof(dheader) - 1;
299 dheader[pos] = ofs & 127;
300 while (ofs >>= 7)
301 dheader[--pos] = 128 | (--ofs & 127);
302 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
303 if (st)
304 close_istream(st);
305 free(buf);
306 return 0;
307 }
308 sha1write(f, header, hdrlen);
309 sha1write(f, dheader + pos, sizeof(dheader) - pos);
310 hdrlen += sizeof(dheader) - pos;
311 } else if (type == OBJ_REF_DELTA) {
312 /*
313 * Deltas with a base reference contain
314 * an additional 20 bytes for the base sha1.
315 */
316 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
317 if (st)
318 close_istream(st);
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 if (st)
328 close_istream(st);
329 free(buf);
330 return 0;
331 }
332 sha1write(f, header, hdrlen);
333 }
334 if (st) {
335 datalen = write_large_blob_data(st, f, entry->idx.sha1);
336 close_istream(st);
337 } else {
338 sha1write(f, buf, datalen);
339 free(buf);
340 }
341 return hdrlen + datalen;
343}
344/* Return 0 if we will bust the pack-size limit */
346static unsigned long write_reuse_object(struct sha1file *f, struct object_entry *entry,
347 unsigned long limit, int usable_delta)
348{
349 struct packed_git *p = entry->in_pack;
350 struct pack_window *w_curs = NULL;
351 struct revindex_entry *revidx;
352 off_t offset;
353 enum object_type type = entry->type;
354 unsigned long datalen;
355 unsigned char header[10], dheader[10];
356 unsigned hdrlen;
357 if (entry->delta)
359 type = (allow_ofs_delta && entry->delta->idx.offset) ?
360 OBJ_OFS_DELTA : OBJ_REF_DELTA;
361 hdrlen = encode_in_pack_object_header(type, entry->size, header);
362 offset = entry->in_pack_offset;
364 revidx = find_pack_revindex(p, offset);
365 datalen = revidx[1].offset - offset;
366 if (!pack_to_stdout && p->index_version > 1 &&
367 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
368 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
369 unuse_pack(&w_curs);
370 return write_no_reuse_object(f, entry, limit, usable_delta);
371 }
372 offset += entry->in_pack_header_size;
374 datalen -= entry->in_pack_header_size;
375 if (!pack_to_stdout && p->index_version == 1 &&
377 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
378 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
379 unuse_pack(&w_curs);
380 return write_no_reuse_object(f, entry, limit, usable_delta);
381 }
382 if (type == OBJ_OFS_DELTA) {
384 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
385 unsigned pos = sizeof(dheader) - 1;
386 dheader[pos] = ofs & 127;
387 while (ofs >>= 7)
388 dheader[--pos] = 128 | (--ofs & 127);
389 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
390 unuse_pack(&w_curs);
391 return 0;
392 }
393 sha1write(f, header, hdrlen);
394 sha1write(f, dheader + pos, sizeof(dheader) - pos);
395 hdrlen += sizeof(dheader) - pos;
396 reused_delta++;
397 } else if (type == OBJ_REF_DELTA) {
398 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
399 unuse_pack(&w_curs);
400 return 0;
401 }
402 sha1write(f, header, hdrlen);
403 sha1write(f, entry->delta->idx.sha1, 20);
404 hdrlen += 20;
405 reused_delta++;
406 } else {
407 if (limit && hdrlen + datalen + 20 >= limit) {
408 unuse_pack(&w_curs);
409 return 0;
410 }
411 sha1write(f, header, hdrlen);
412 }
413 copy_pack_data(f, p, &w_curs, offset, datalen);
414 unuse_pack(&w_curs);
415 reused++;
416 return hdrlen + datalen;
417}
418/* Return 0 if we will bust the pack-size limit */
420static unsigned long write_object(struct sha1file *f,
421 struct object_entry *entry,
422 off_t write_offset)
423{
424 unsigned long limit, len;
425 int usable_delta, to_reuse;
426 if (!pack_to_stdout)
428 crc32_begin(f);
429 /* apply size limit if limited packsize and not first object */
431 if (!pack_size_limit || !nr_written)
432 limit = 0;
433 else if (pack_size_limit <= write_offset)
434 /*
435 * the earlier object did not fit the limit; avoid
436 * mistaking this with unlimited (i.e. limit = 0).
437 */
438 limit = 1;
439 else
440 limit = pack_size_limit - write_offset;
441 if (!entry->delta)
443 usable_delta = 0; /* no delta */
444 else if (!pack_size_limit)
445 usable_delta = 1; /* unlimited packfile */
446 else if (entry->delta->idx.offset == (off_t)-1)
447 usable_delta = 0; /* base was written to another pack */
448 else if (entry->delta->idx.offset)
449 usable_delta = 1; /* base already exists in this pack */
450 else
451 usable_delta = 0; /* base could end up in another pack */
452 if (!reuse_object)
454 to_reuse = 0; /* explicit */
455 else if (!entry->in_pack)
456 to_reuse = 0; /* can't reuse what we don't have */
457 else if (entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA)
458 /* check_object() decided it for us ... */
459 to_reuse = usable_delta;
460 /* ... but pack split may override that */
461 else if (entry->type != entry->in_pack_type)
462 to_reuse = 0; /* pack has delta which is unusable */
463 else if (entry->delta)
464 to_reuse = 0; /* we want to pack afresh */
465 else
466 to_reuse = 1; /* we have it in-pack undeltified,
467 * and we do not need to deltify it.
468 */
469 if (!to_reuse)
471 len = write_no_reuse_object(f, entry, limit, usable_delta);
472 else
473 len = write_reuse_object(f, entry, limit, usable_delta);
474 if (!len)
475 return 0;
476 if (usable_delta)
478 written_delta++;
479 written++;
480 if (!pack_to_stdout)
481 entry->idx.crc32 = crc32_end(f);
482 return len;
483}
484enum write_one_status {
486 WRITE_ONE_SKIP = -1, /* already written */
487 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
488 WRITE_ONE_WRITTEN = 1, /* normal */
489 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
490};
491static enum write_one_status write_one(struct sha1file *f,
493 struct object_entry *e,
494 off_t *offset)
495{
496 unsigned long size;
497 int recursing;
498 /*
500 * we set offset to 1 (which is an impossible value) to mark
501 * the fact that this object is involved in "write its base
502 * first before writing a deltified object" recursion.
503 */
504 recursing = (e->idx.offset == 1);
505 if (recursing) {
506 warning("recursive delta detected for object %s",
507 sha1_to_hex(e->idx.sha1));
508 return WRITE_ONE_RECURSIVE;
509 } else if (e->idx.offset || e->preferred_base) {
510 /* offset is non zero if object is written already. */
511 return WRITE_ONE_SKIP;
512 }
513 /* if we are deltified, write out base object first. */
515 if (e->delta) {
516 e->idx.offset = 1; /* now recurse */
517 switch (write_one(f, e->delta, offset)) {
518 case WRITE_ONE_RECURSIVE:
519 /* we cannot depend on this one */
520 e->delta = NULL;
521 break;
522 default:
523 break;
524 case WRITE_ONE_BREAK:
525 e->idx.offset = recursing;
526 return WRITE_ONE_BREAK;
527 }
528 }
529 e->idx.offset = *offset;
531 size = write_object(f, e, *offset);
532 if (!size) {
533 e->idx.offset = recursing;
534 return WRITE_ONE_BREAK;
535 }
536 written_list[nr_written++] = &e->idx;
537 /* make sure off_t is sufficiently large not to wrap */
539 if (signed_add_overflows(*offset, size))
540 die("pack too large for current definition of off_t");
541 *offset += size;
542 return WRITE_ONE_WRITTEN;
543}
544static int mark_tagged(const char *path, const struct object_id *oid, int flag,
546 void *cb_data)
547{
548 unsigned char peeled[20];
549 struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
550 if (entry)
552 entry->tagged = 1;
553 if (!peel_ref(path, peeled)) {
554 entry = packlist_find(&to_pack, peeled, NULL);
555 if (entry)
556 entry->tagged = 1;
557 }
558 return 0;
559}
560static inline void add_to_write_order(struct object_entry **wo,
562 unsigned int *endp,
563 struct object_entry *e)
564{
565 if (e->filled)
566 return;
567 wo[(*endp)++] = e;
568 e->filled = 1;
569}
570static void add_descendants_to_write_order(struct object_entry **wo,
572 unsigned int *endp,
573 struct object_entry *e)
574{
575 int add_to_order = 1;
576 while (e) {
577 if (add_to_order) {
578 struct object_entry *s;
579 /* add this node... */
580 add_to_write_order(wo, endp, e);
581 /* all its siblings... */
582 for (s = e->delta_sibling; s; s = s->delta_sibling) {
583 add_to_write_order(wo, endp, s);
584 }
585 }
586 /* drop down a level to add left subtree nodes if possible */
587 if (e->delta_child) {
588 add_to_order = 1;
589 e = e->delta_child;
590 } else {
591 add_to_order = 0;
592 /* our sibling might have some children, it is next */
593 if (e->delta_sibling) {
594 e = e->delta_sibling;
595 continue;
596 }
597 /* go back to our parent node */
598 e = e->delta;
599 while (e && !e->delta_sibling) {
600 /* we're on the right side of a subtree, keep
601 * going up until we can go right again */
602 e = e->delta;
603 }
604 if (!e) {
605 /* done- we hit our original root node */
606 return;
607 }
608 /* pass it off to sibling at this level */
609 e = e->delta_sibling;
610 }
611 };
612}
613static void add_family_to_write_order(struct object_entry **wo,
615 unsigned int *endp,
616 struct object_entry *e)
617{
618 struct object_entry *root;
619 for (root = e; root->delta; root = root->delta)
621 ; /* nothing */
622 add_descendants_to_write_order(wo, endp, root);
623}
624static struct object_entry **compute_write_order(void)
626{
627 unsigned int i, wo_end, last_untagged;
628 struct object_entry **wo;
630 struct object_entry *objects = to_pack.objects;
631 for (i = 0; i < to_pack.nr_objects; i++) {
633 objects[i].tagged = 0;
634 objects[i].filled = 0;
635 objects[i].delta_child = NULL;
636 objects[i].delta_sibling = NULL;
637 }
638 /*
640 * Fully connect delta_child/delta_sibling network.
641 * Make sure delta_sibling is sorted in the original
642 * recency order.
643 */
644 for (i = to_pack.nr_objects; i > 0;) {
645 struct object_entry *e = &objects[--i];
646 if (!e->delta)
647 continue;
648 /* Mark me as the first child */
649 e->delta_sibling = e->delta->delta_child;
650 e->delta->delta_child = e;
651 }
652 /*
654 * Mark objects that are at the tip of tags.
655 */
656 for_each_tag_ref(mark_tagged, NULL);
657 /*
659 * Give the objects in the original recency order until
660 * we see a tagged tip.
661 */
662 ALLOC_ARRAY(wo, to_pack.nr_objects);
663 for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
664 if (objects[i].tagged)
665 break;
666 add_to_write_order(wo, &wo_end, &objects[i]);
667 }
668 last_untagged = i;
669 /*
671 * Then fill all the tagged tips.
672 */
673 for (; i < to_pack.nr_objects; i++) {
674 if (objects[i].tagged)
675 add_to_write_order(wo, &wo_end, &objects[i]);
676 }
677 /*
679 * And then all remaining commits and tags.
680 */
681 for (i = last_untagged; i < to_pack.nr_objects; i++) {
682 if (objects[i].type != OBJ_COMMIT &&
683 objects[i].type != OBJ_TAG)
684 continue;
685 add_to_write_order(wo, &wo_end, &objects[i]);
686 }
687 /*
689 * And then all the trees.
690 */
691 for (i = last_untagged; i < to_pack.nr_objects; i++) {
692 if (objects[i].type != OBJ_TREE)
693 continue;
694 add_to_write_order(wo, &wo_end, &objects[i]);
695 }
696 /*
698 * Finally all the rest in really tight order
699 */
700 for (i = last_untagged; i < to_pack.nr_objects; i++) {
701 if (!objects[i].filled)
702 add_family_to_write_order(wo, &wo_end, &objects[i]);
703 }
704 if (wo_end != to_pack.nr_objects)
706 die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
707 return wo;
709}
710static off_t write_reused_pack(struct sha1file *f)
712{
713 unsigned char buffer[8192];
714 off_t to_write, total;
715 int fd;
716 if (!is_pack_valid(reuse_packfile))
718 die("packfile is invalid: %s", reuse_packfile->pack_name);
719 fd = git_open_noatime(reuse_packfile->pack_name);
721 if (fd < 0)
722 die_errno("unable to open packfile for reuse: %s",
723 reuse_packfile->pack_name);
724 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
726 die_errno("unable to seek in reused packfile");
727 if (reuse_packfile_offset < 0)
729 reuse_packfile_offset = reuse_packfile->pack_size - 20;
730 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
732 while (to_write) {
734 int read_pack = xread(fd, buffer, sizeof(buffer));
735 if (read_pack <= 0)
737 die_errno("unable to read from reused packfile");
738 if (read_pack > to_write)
740 read_pack = to_write;
741 sha1write(f, buffer, read_pack);
743 to_write -= read_pack;
744 /*
746 * We don't know the actual number of objects written,
747 * only how many bytes written, how many bytes total, and
748 * how many objects total. So we can fake it by pretending all
749 * objects we are writing are the same size. This gives us a
750 * smooth progress meter, and at the end it matches the true
751 * answer.
752 */
753 written = reuse_packfile_objects *
754 (((double)(total - to_write)) / total);
755 display_progress(progress_state, written);
756 }
757 close(fd);
759 written = reuse_packfile_objects;
760 display_progress(progress_state, written);
761 return reuse_packfile_offset - sizeof(struct pack_header);
762}
763static const char no_split_warning[] = N_(
765"disabling bitmap writing, packs are split due to pack.packSizeLimit"
766);
767static void write_pack_file(void)
769{
770 uint32_t i = 0, j;
771 struct sha1file *f;
772 off_t offset;
773 uint32_t nr_remaining = nr_result;
774 time_t last_mtime = 0;
775 struct object_entry **write_order;
776 if (progress > pack_to_stdout)
778 progress_state = start_progress(_("Writing objects"), nr_result);
779 ALLOC_ARRAY(written_list, to_pack.nr_objects);
780 write_order = compute_write_order();
781 do {
783 unsigned char sha1[20];
784 char *pack_tmp_name = NULL;
785 if (pack_to_stdout)
787 f = sha1fd_throughput(1, "<stdout>", progress_state);
788 else
789 f = create_tmp_packfile(&pack_tmp_name);
790 offset = write_pack_header(f, nr_remaining);
792 if (reuse_packfile) {
794 off_t packfile_size;
795 assert(pack_to_stdout);
796 packfile_size = write_reused_pack(f);
798 offset += packfile_size;
799 }
800 nr_written = 0;
802 for (; i < to_pack.nr_objects; i++) {
803 struct object_entry *e = write_order[i];
804 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
805 break;
806 display_progress(progress_state, written);
807 }
808 /*
810 * Did we write the wrong # entries in the header?
811 * If so, rewrite it like in fast-import
812 */
813 if (pack_to_stdout) {
814 sha1close(f, sha1, CSUM_CLOSE);
815 } else if (nr_written == nr_remaining) {
816 sha1close(f, sha1, CSUM_FSYNC);
817 } else {
818 int fd = sha1close(f, sha1, 0);
819 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
820 nr_written, sha1, offset);
821 close(fd);
822 if (write_bitmap_index) {
823 warning(_(no_split_warning));
824 write_bitmap_index = 0;
825 }
826 }
827 if (!pack_to_stdout) {
829 struct stat st;
830 struct strbuf tmpname = STRBUF_INIT;
831 /*
833 * Packs are runtime accessed in their mtime
834 * order since newer packs are more likely to contain
835 * younger objects. So if we are creating multiple
836 * packs then we should modify the mtime of later ones
837 * to preserve this property.
838 */
839 if (stat(pack_tmp_name, &st) < 0) {
840 warning_errno("failed to stat %s", pack_tmp_name);
841 } else if (!last_mtime) {
842 last_mtime = st.st_mtime;
843 } else {
844 struct utimbuf utb;
845 utb.actime = st.st_atime;
846 utb.modtime = --last_mtime;
847 if (utime(pack_tmp_name, &utb) < 0)
848 warning_errno("failed utime() on %s", pack_tmp_name);
849 }
850 strbuf_addf(&tmpname, "%s-", base_name);
852 if (write_bitmap_index) {
854 bitmap_writer_set_checksum(sha1);
855 bitmap_writer_build_type_index(written_list, nr_written);
856 }
857 finish_tmp_packfile(&tmpname, pack_tmp_name,
859 written_list, nr_written,
860 &pack_idx_opts, sha1);
861 if (write_bitmap_index) {
863 strbuf_addf(&tmpname, "%s.bitmap", sha1_to_hex(sha1));
864 stop_progress(&progress_state);
866 bitmap_writer_show_progress(progress);
868 bitmap_writer_reuse_bitmaps(&to_pack);
869 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
870 bitmap_writer_build(&to_pack);
871 bitmap_writer_finish(written_list, nr_written,
872 tmpname.buf, write_bitmap_options);
873 write_bitmap_index = 0;
874 }
875 strbuf_release(&tmpname);
877 free(pack_tmp_name);
878 puts(sha1_to_hex(sha1));
879 }
880 /* mark written objects as written to previous pack */
882 for (j = 0; j < nr_written; j++) {
883 written_list[j]->offset = (off_t)-1;
884 }
885 nr_remaining -= nr_written;
886 } while (nr_remaining && i < to_pack.nr_objects);
887 free(written_list);
889 free(write_order);
890 stop_progress(&progress_state);
891 if (written != nr_result)
892 die("wrote %"PRIu32" objects while expecting %"PRIu32,
893 written, nr_result);
894}
895static void setup_delta_attr_check(struct git_attr_check *check)
897{
898 static struct git_attr *attr_delta;
899 if (!attr_delta)
901 attr_delta = git_attr("delta");
902 check[0].attr = attr_delta;
904}
905static int no_try_delta(const char *path)
907{
908 struct git_attr_check check[1];
909 setup_delta_attr_check(check);
911 if (git_check_attr(path, ARRAY_SIZE(check), check))
912 return 0;
913 if (ATTR_FALSE(check->value))
914 return 1;
915 return 0;
916}
917/*
919 * When adding an object, check whether we have already added it
920 * to our packing list. If so, we can skip. However, if we are
921 * being asked to excludei t, but the previous mention was to include
922 * it, make sure to adjust its flags and tweak our numbers accordingly.
923 *
924 * As an optimization, we pass out the index position where we would have
925 * found the item, since that saves us from having to look it up again a
926 * few lines later when we want to add the new entry.
927 */
928static int have_duplicate_entry(const unsigned char *sha1,
929 int exclude,
930 uint32_t *index_pos)
931{
932 struct object_entry *entry;
933 entry = packlist_find(&to_pack, sha1, index_pos);
935 if (!entry)
936 return 0;
937 if (exclude) {
939 if (!entry->preferred_base)
940 nr_result--;
941 entry->preferred_base = 1;
942 }
943 return 1;
945}
946static int want_found_object(int exclude, struct packed_git *p)
948{
949 if (exclude)
950 return 1;
951 if (incremental)
952 return 0;
953 /*
955 * When asked to do --local (do not include an object that appears in a
956 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
957 * an object that appears in a pack marked with .keep), finding a pack
958 * that matches the criteria is sufficient for us to decide to omit it.
959 * However, even if this pack does not satisfy the criteria, we need to
960 * make sure no copy of this object appears in _any_ pack that makes us
961 * to omit the object, so we need to check all the packs.
962 *
963 * We can however first check whether these options can possible matter;
964 * if they do not matter we know we want the object in generated pack.
965 * Otherwise, we signal "-1" at the end to tell the caller that we do
966 * not know either way, and it needs to check more packs.
967 */
968 if (!ignore_packed_keep &&
969 (!local || !have_non_local_packs))
970 return 1;
971 if (local && !p->pack_local)
973 return 0;
974 if (ignore_packed_keep && p->pack_local && p->pack_keep)
975 return 0;
976 /* we don't know yet; keep looking for more packs */
978 return -1;
979}
980/*
982 * Check whether we want the object in the pack (e.g., we do not want
983 * objects found in non-local stores if the "--local" option was used).
984 *
985 * If the caller already knows an existing pack it wants to take the object
986 * from, that is passed in *found_pack and *found_offset; otherwise this
987 * function finds if there is any pack that has the object and returns the pack
988 * and its offset in these variables.
989 */
990static int want_object_in_pack(const unsigned char *sha1,
991 int exclude,
992 struct packed_git **found_pack,
993 off_t *found_offset)
994{
995 struct packed_git *p;
996 int want;
997 if (!exclude && local && has_loose_object_nonlocal(sha1))
999 return 0;
1000 /*
1002 * If we already know the pack object lives in, start checks from that
1003 * pack - in the usual case when neither --local was given nor .keep files
1004 * are present we will determine the answer right now.
1005 */
1006 if (*found_pack) {
1007 want = want_found_object(exclude, *found_pack);
1008 if (want != -1)
1009 return want;
1010 }
1011 for (p = packed_git; p; p = p->next) {
1013 off_t offset;
1014 if (p == *found_pack)
1016 offset = *found_offset;
1017 else
1018 offset = find_pack_entry_one(sha1, p);
1019 if (offset) {
1021 if (!*found_pack) {
1022 if (!is_pack_valid(p))
1023 continue;
1024 *found_offset = offset;
1025 *found_pack = p;
1026 }
1027 want = want_found_object(exclude, p);
1028 if (want != -1)
1029 return want;
1030 }
1031 }
1032 return 1;
1034}
1035static void create_object_entry(const unsigned char *sha1,
1037 enum object_type type,
1038 uint32_t hash,
1039 int exclude,
1040 int no_try_delta,
1041 uint32_t index_pos,
1042 struct packed_git *found_pack,
1043 off_t found_offset)
1044{
1045 struct object_entry *entry;
1046 entry = packlist_alloc(&to_pack, sha1, index_pos);
1048 entry->hash = hash;
1049 if (type)
1050 entry->type = type;
1051 if (exclude)
1052 entry->preferred_base = 1;
1053 else
1054 nr_result++;
1055 if (found_pack) {
1056 entry->in_pack = found_pack;
1057 entry->in_pack_offset = found_offset;
1058 }
1059 entry->no_try_delta = no_try_delta;
1061}
1062static const char no_closure_warning[] = N_(
1064"disabling bitmap writing, as some objects are not being packed"
1065);
1066static int add_object_entry(const unsigned char *sha1, enum object_type type,
1068 const char *name, int exclude)
1069{
1070 struct packed_git *found_pack = NULL;
1071 off_t found_offset = 0;
1072 uint32_t index_pos;
1073 if (have_duplicate_entry(sha1, exclude, &index_pos))
1075 return 0;
1076 if (!want_object_in_pack(sha1, exclude, &found_pack, &found_offset)) {
1078 /* The pack is missing an object, so it will not have closure */
1079 if (write_bitmap_index) {
1080 warning(_(no_closure_warning));
1081 write_bitmap_index = 0;
1082 }
1083 return 0;
1084 }
1085 create_object_entry(sha1, type, pack_name_hash(name),
1087 exclude, name && no_try_delta(name),
1088 index_pos, found_pack, found_offset);
1089 display_progress(progress_state, nr_result);
1091 return 1;
1092}
1093static int add_object_entry_from_bitmap(const unsigned char *sha1,
1095 enum object_type type,
1096 int flags, uint32_t name_hash,
1097 struct packed_git *pack, off_t offset)
1098{
1099 uint32_t index_pos;
1100 if (have_duplicate_entry(sha1, 0, &index_pos))
1102 return 0;
1103 if (!want_object_in_pack(sha1, 0, &pack, &offset))
1105 return 0;
1106 create_object_entry(sha1, type, name_hash, 0, 0, index_pos, pack, offset);
1108 display_progress(progress_state, nr_result);
1110 return 1;
1111}
1112struct pbase_tree_cache {
1114 unsigned char sha1[20];
1115 int ref;
1116 int temporary;
1117 void *tree_data;
1118 unsigned long tree_size;
1119};
1120static struct pbase_tree_cache *(pbase_tree_cache[256]);
1122static int pbase_tree_cache_ix(const unsigned char *sha1)
1123{
1124 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
1125}
1126static int pbase_tree_cache_ix_incr(int ix)
1127{
1128 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1129}
1130static struct pbase_tree {
1132 struct pbase_tree *next;
1133 /* This is a phony "cache" entry; we are not
1134 * going to evict it or find it through _get()
1135 * mechanism -- this is for the toplevel node that
1136 * would almost always change with any commit.
1137 */
1138 struct pbase_tree_cache pcache;
1139} *pbase_tree;
1140static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1142{
1143 struct pbase_tree_cache *ent, *nent;
1144 void *data;
1145 unsigned long size;
1146 enum object_type type;
1147 int neigh;
1148 int my_ix = pbase_tree_cache_ix(sha1);
1149 int available_ix = -1;
1150 /* pbase-tree-cache acts as a limited hashtable.
1152 * your object will be found at your index or within a few
1153 * slots after that slot if it is cached.
1154 */
1155 for (neigh = 0; neigh < 8; neigh++) {
1156 ent = pbase_tree_cache[my_ix];
1157 if (ent && !hashcmp(ent->sha1, sha1)) {
1158 ent->ref++;
1159 return ent;
1160 }
1161 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1162 ((0 <= available_ix) &&
1163 (!ent && pbase_tree_cache[available_ix])))
1164 available_ix = my_ix;
1165 if (!ent)
1166 break;
1167 my_ix = pbase_tree_cache_ix_incr(my_ix);
1168 }
1169 /* Did not find one. Either we got a bogus request or
1171 * we need to read and perhaps cache.
1172 */
1173 data = read_sha1_file(sha1, &type, &size);
1174 if (!data)
1175 return NULL;
1176 if (type != OBJ_TREE) {
1177 free(data);
1178 return NULL;
1179 }
1180 /* We need to either cache or return a throwaway copy */
1182 if (available_ix < 0)
1184 ent = NULL;
1185 else {
1186 ent = pbase_tree_cache[available_ix];
1187 my_ix = available_ix;
1188 }
1189 if (!ent) {
1191 nent = xmalloc(sizeof(*nent));
1192 nent->temporary = (available_ix < 0);
1193 }
1194 else {
1195 /* evict and reuse */
1196 free(ent->tree_data);
1197 nent = ent;
1198 }
1199 hashcpy(nent->sha1, sha1);
1200 nent->tree_data = data;
1201 nent->tree_size = size;
1202 nent->ref = 1;
1203 if (!nent->temporary)
1204 pbase_tree_cache[my_ix] = nent;
1205 return nent;
1206}
1207static void pbase_tree_put(struct pbase_tree_cache *cache)
1209{
1210 if (!cache->temporary) {
1211 cache->ref--;
1212 return;
1213 }
1214 free(cache->tree_data);
1215 free(cache);
1216}
1217static int name_cmp_len(const char *name)
1219{
1220 int i;
1221 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1222 ;
1223 return i;
1224}
1225static void add_pbase_object(struct tree_desc *tree,
1227 const char *name,
1228 int cmplen,
1229 const char *fullname)
1230{
1231 struct name_entry entry;
1232 int cmp;
1233 while (tree_entry(tree,&entry)) {
1235 if (S_ISGITLINK(entry.mode))
1236 continue;
1237 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1238 memcmp(name, entry.path, cmplen);
1239 if (cmp > 0)
1240 continue;
1241 if (cmp < 0)
1242 return;
1243 if (name[cmplen] != '/') {
1244 add_object_entry(entry.oid->hash,
1245 object_type(entry.mode),
1246 fullname, 1);
1247 return;
1248 }
1249 if (S_ISDIR(entry.mode)) {
1250 struct tree_desc sub;
1251 struct pbase_tree_cache *tree;
1252 const char *down = name+cmplen+1;
1253 int downlen = name_cmp_len(down);
1254 tree = pbase_tree_get(entry.oid->hash);
1256 if (!tree)
1257 return;
1258 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1259 add_pbase_object(&sub, down, downlen, fullname);
1261 pbase_tree_put(tree);
1262 }
1263 }
1264}
1265static unsigned *done_pbase_paths;
1267static int done_pbase_paths_num;
1268static int done_pbase_paths_alloc;
1269static int done_pbase_path_pos(unsigned hash)
1270{
1271 int lo = 0;
1272 int hi = done_pbase_paths_num;
1273 while (lo < hi) {
1274 int mi = (hi + lo) / 2;
1275 if (done_pbase_paths[mi] == hash)
1276 return mi;
1277 if (done_pbase_paths[mi] < hash)
1278 hi = mi;
1279 else
1280 lo = mi + 1;
1281 }
1282 return -lo-1;
1283}
1284static int check_pbase_path(unsigned hash)
1286{
1287 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1288 if (0 <= pos)
1289 return 1;
1290 pos = -pos - 1;
1291 ALLOC_GROW(done_pbase_paths,
1292 done_pbase_paths_num + 1,
1293 done_pbase_paths_alloc);
1294 done_pbase_paths_num++;
1295 if (pos < done_pbase_paths_num)
1296 memmove(done_pbase_paths + pos + 1,
1297 done_pbase_paths + pos,
1298 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1299 done_pbase_paths[pos] = hash;
1300 return 0;
1301}
1302static void add_preferred_base_object(const char *name)
1304{
1305 struct pbase_tree *it;
1306 int cmplen;
1307 unsigned hash = pack_name_hash(name);
1308 if (!num_preferred_base || check_pbase_path(hash))
1310 return;
1311 cmplen = name_cmp_len(name);
1313 for (it = pbase_tree; it; it = it->next) {
1314 if (cmplen == 0) {
1315 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1316 }
1317 else {
1318 struct tree_desc tree;
1319 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1320 add_pbase_object(&tree, name, cmplen, name);
1321 }
1322 }
1323}
1324static void add_preferred_base(unsigned char *sha1)
1326{
1327 struct pbase_tree *it;
1328 void *data;
1329 unsigned long size;
1330 unsigned char tree_sha1[20];
1331 if (window <= num_preferred_base++)
1333 return;
1334 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1336 if (!data)
1337 return;
1338 for (it = pbase_tree; it; it = it->next) {
1340 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1341 free(data);
1342 return;
1343 }
1344 }
1345 it = xcalloc(1, sizeof(*it));
1347 it->next = pbase_tree;
1348 pbase_tree = it;
1349 hashcpy(it->pcache.sha1, tree_sha1);
1351 it->pcache.tree_data = data;
1352 it->pcache.tree_size = size;
1353}
1354static void cleanup_preferred_base(void)
1356{
1357 struct pbase_tree *it;
1358 unsigned i;
1359 it = pbase_tree;
1361 pbase_tree = NULL;
1362 while (it) {
1363 struct pbase_tree *this = it;
1364 it = this->next;
1365 free(this->pcache.tree_data);
1366 free(this);
1367 }
1368 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1370 if (!pbase_tree_cache[i])
1371 continue;
1372 free(pbase_tree_cache[i]->tree_data);
1373 free(pbase_tree_cache[i]);
1374 pbase_tree_cache[i] = NULL;
1375 }
1376 free(done_pbase_paths);
1378 done_pbase_paths = NULL;
1379 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1380}
1381static void check_object(struct object_entry *entry)
1383{
1384 if (entry->in_pack) {
1385 struct packed_git *p = entry->in_pack;
1386 struct pack_window *w_curs = NULL;
1387 const unsigned char *base_ref = NULL;
1388 struct object_entry *base_entry;
1389 unsigned long used, used_0;
1390 unsigned long avail;
1391 off_t ofs;
1392 unsigned char *buf, c;
1393 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1395 /*
1397 * We want in_pack_type even if we do not reuse delta
1398 * since non-delta representations could still be reused.
1399 */
1400 used = unpack_object_header_buffer(buf, avail,
1401 &entry->in_pack_type,
1402 &entry->size);
1403 if (used == 0)
1404 goto give_up;
1405 /*
1407 * Determine if this is a delta and if so whether we can
1408 * reuse it or not. Otherwise let's find out as cheaply as
1409 * possible what the actual type and size for this object is.
1410 */
1411 switch (entry->in_pack_type) {
1412 default:
1413 /* Not a delta hence we've already got all we need. */
1414 entry->type = entry->in_pack_type;
1415 entry->in_pack_header_size = used;
1416 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1417 goto give_up;
1418 unuse_pack(&w_curs);
1419 return;
1420 case OBJ_REF_DELTA:
1421 if (reuse_delta && !entry->preferred_base)
1422 base_ref = use_pack(p, &w_curs,
1423 entry->in_pack_offset + used, NULL);
1424 entry->in_pack_header_size = used + 20;
1425 break;
1426 case OBJ_OFS_DELTA:
1427 buf = use_pack(p, &w_curs,
1428 entry->in_pack_offset + used, NULL);
1429 used_0 = 0;
1430 c = buf[used_0++];
1431 ofs = c & 127;
1432 while (c & 128) {
1433 ofs += 1;
1434 if (!ofs || MSB(ofs, 7)) {
1435 error("delta base offset overflow in pack for %s",
1436 sha1_to_hex(entry->idx.sha1));
1437 goto give_up;
1438 }
1439 c = buf[used_0++];
1440 ofs = (ofs << 7) + (c & 127);
1441 }
1442 ofs = entry->in_pack_offset - ofs;
1443 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1444 error("delta base offset out of bound for %s",
1445 sha1_to_hex(entry->idx.sha1));
1446 goto give_up;
1447 }
1448 if (reuse_delta && !entry->preferred_base) {
1449 struct revindex_entry *revidx;
1450 revidx = find_pack_revindex(p, ofs);
1451 if (!revidx)
1452 goto give_up;
1453 base_ref = nth_packed_object_sha1(p, revidx->nr);
1454 }
1455 entry->in_pack_header_size = used + used_0;
1456 break;
1457 }
1458 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1460 /*
1461 * If base_ref was set above that means we wish to
1462 * reuse delta data, and we even found that base
1463 * in the list of objects we want to pack. Goodie!
1464 *
1465 * Depth value does not matter - find_deltas() will
1466 * never consider reused delta as the base object to
1467 * deltify other objects against, in order to avoid
1468 * circular deltas.
1469 */
1470 entry->type = entry->in_pack_type;
1471 entry->delta = base_entry;
1472 entry->delta_size = entry->size;
1473 entry->delta_sibling = base_entry->delta_child;
1474 base_entry->delta_child = entry;
1475 unuse_pack(&w_curs);
1476 return;
1477 }
1478 if (entry->type) {
1480 /*
1481 * This must be a delta and we already know what the
1482 * final object type is. Let's extract the actual
1483 * object size from the delta header.
1484 */
1485 entry->size = get_size_from_delta(p, &w_curs,
1486 entry->in_pack_offset + entry->in_pack_header_size);
1487 if (entry->size == 0)
1488 goto give_up;
1489 unuse_pack(&w_curs);
1490 return;
1491 }
1492 /*
1494 * No choice but to fall back to the recursive delta walk
1495 * with sha1_object_info() to find about the object type
1496 * at this point...
1497 */
1498 give_up:
1499 unuse_pack(&w_curs);
1500 }
1501 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1503 /*
1504 * The error condition is checked in prepare_pack(). This is
1505 * to permit a missing preferred base object to be ignored
1506 * as a preferred base. Doing so can result in a larger
1507 * pack file, but the transfer will still take place.
1508 */
1509}
1510static int pack_offset_sort(const void *_a, const void *_b)
1512{
1513 const struct object_entry *a = *(struct object_entry **)_a;
1514 const struct object_entry *b = *(struct object_entry **)_b;
1515 /* avoid filesystem trashing with loose objects */
1517 if (!a->in_pack && !b->in_pack)
1518 return hashcmp(a->idx.sha1, b->idx.sha1);
1519 if (a->in_pack < b->in_pack)
1521 return -1;
1522 if (a->in_pack > b->in_pack)
1523 return 1;
1524 return a->in_pack_offset < b->in_pack_offset ? -1 :
1525 (a->in_pack_offset > b->in_pack_offset);
1526}
1527static void get_object_details(void)
1529{
1530 uint32_t i;
1531 struct object_entry **sorted_by_offset;
1532 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1534 for (i = 0; i < to_pack.nr_objects; i++)
1535 sorted_by_offset[i] = to_pack.objects + i;
1536 qsort(sorted_by_offset, to_pack.nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1537 for (i = 0; i < to_pack.nr_objects; i++) {
1539 struct object_entry *entry = sorted_by_offset[i];
1540 check_object(entry);
1541 if (big_file_threshold < entry->size)
1542 entry->no_try_delta = 1;
1543 }
1544 free(sorted_by_offset);
1546}
1547/*
1549 * We search for deltas in a list sorted by type, by filename hash, and then
1550 * by size, so that we see progressively smaller and smaller files.
1551 * That's because we prefer deltas to be from the bigger file
1552 * to the smaller -- deletes are potentially cheaper, but perhaps
1553 * more importantly, the bigger file is likely the more recent
1554 * one. The deepest deltas are therefore the oldest objects which are
1555 * less susceptible to be accessed often.
1556 */
1557static int type_size_sort(const void *_a, const void *_b)
1558{
1559 const struct object_entry *a = *(struct object_entry **)_a;
1560 const struct object_entry *b = *(struct object_entry **)_b;
1561 if (a->type > b->type)
1563 return -1;
1564 if (a->type < b->type)
1565 return 1;
1566 if (a->hash > b->hash)
1567 return -1;
1568 if (a->hash < b->hash)
1569 return 1;
1570 if (a->preferred_base > b->preferred_base)
1571 return -1;
1572 if (a->preferred_base < b->preferred_base)
1573 return 1;
1574 if (a->size > b->size)
1575 return -1;
1576 if (a->size < b->size)
1577 return 1;
1578 return a < b ? -1 : (a > b); /* newest first */
1579}
1580struct unpacked {
1582 struct object_entry *entry;
1583 void *data;
1584 struct delta_index *index;
1585 unsigned depth;
1586};
1587static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1589 unsigned long delta_size)
1590{
1591 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1592 return 0;
1593 if (delta_size < cache_max_small_delta_size)
1595 return 1;
1596 /* cache delta, if objects are large enough compared to delta size */
1598 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1599 return 1;
1600 return 0;
1602}
1603#ifndef NO_PTHREADS
1605static pthread_mutex_t read_mutex;
1607#define read_lock() pthread_mutex_lock(&read_mutex)
1608#define read_unlock() pthread_mutex_unlock(&read_mutex)
1609static pthread_mutex_t cache_mutex;
1611#define cache_lock() pthread_mutex_lock(&cache_mutex)
1612#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1613static pthread_mutex_t progress_mutex;
1615#define progress_lock() pthread_mutex_lock(&progress_mutex)
1616#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1617#else
1619#define read_lock() (void)0
1621#define read_unlock() (void)0
1622#define cache_lock() (void)0
1623#define cache_unlock() (void)0
1624#define progress_lock() (void)0
1625#define progress_unlock() (void)0
1626#endif
1628static int try_delta(struct unpacked *trg, struct unpacked *src,
1630 unsigned max_depth, unsigned long *mem_usage)
1631{
1632 struct object_entry *trg_entry = trg->entry;
1633 struct object_entry *src_entry = src->entry;
1634 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1635 unsigned ref_depth;
1636 enum object_type type;
1637 void *delta_buf;
1638 /* Don't bother doing diffs between different types */
1640 if (trg_entry->type != src_entry->type)
1641 return -1;
1642 /*
1644 * We do not bother to try a delta that we discarded on an
1645 * earlier try, but only when reusing delta data. Note that
1646 * src_entry that is marked as the preferred_base should always
1647 * be considered, as even if we produce a suboptimal delta against
1648 * it, we will still save the transfer cost, as we already know
1649 * the other side has it and we won't send src_entry at all.
1650 */
1651 if (reuse_delta && trg_entry->in_pack &&
1652 trg_entry->in_pack == src_entry->in_pack &&
1653 !src_entry->preferred_base &&
1654 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1655 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1656 return 0;
1657 /* Let's not bust the allowed depth. */
1659 if (src->depth >= max_depth)
1660 return 0;
1661 /* Now some size filtering heuristics. */
1663 trg_size = trg_entry->size;
1664 if (!trg_entry->delta) {
1665 max_size = trg_size/2 - 20;
1666 ref_depth = 1;
1667 } else {
1668 max_size = trg_entry->delta_size;
1669 ref_depth = trg->depth;
1670 }
1671 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1672 (max_depth - ref_depth + 1);
1673 if (max_size == 0)
1674 return 0;
1675 src_size = src_entry->size;
1676 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1677 if (sizediff >= max_size)
1678 return 0;
1679 if (trg_size < src_size / 32)
1680 return 0;
1681 /* Load data if not already done */
1683 if (!trg->data) {
1684 read_lock();
1685 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1686 read_unlock();
1687 if (!trg->data)
1688 die("object %s cannot be read",
1689 sha1_to_hex(trg_entry->idx.sha1));
1690 if (sz != trg_size)
1691 die("object %s inconsistent object length (%lu vs %lu)",
1692 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1693 *mem_usage += sz;
1694 }
1695 if (!src->data) {
1696 read_lock();
1697 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1698 read_unlock();
1699 if (!src->data) {
1700 if (src_entry->preferred_base) {
1701 static int warned = 0;
1702 if (!warned++)
1703 warning("object %s cannot be read",
1704 sha1_to_hex(src_entry->idx.sha1));
1705 /*
1706 * Those objects are not included in the
1707 * resulting pack. Be resilient and ignore
1708 * them if they can't be read, in case the
1709 * pack could be created nevertheless.
1710 */
1711 return 0;
1712 }
1713 die("object %s cannot be read",
1714 sha1_to_hex(src_entry->idx.sha1));
1715 }
1716 if (sz != src_size)
1717 die("object %s inconsistent object length (%lu vs %lu)",
1718 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1719 *mem_usage += sz;
1720 }
1721 if (!src->index) {
1722 src->index = create_delta_index(src->data, src_size);
1723 if (!src->index) {
1724 static int warned = 0;
1725 if (!warned++)
1726 warning("suboptimal pack - out of memory");
1727 return 0;
1728 }
1729 *mem_usage += sizeof_delta_index(src->index);
1730 }
1731 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1733 if (!delta_buf)
1734 return 0;
1735 if (trg_entry->delta) {
1737 /* Prefer only shallower same-sized deltas. */
1738 if (delta_size == trg_entry->delta_size &&
1739 src->depth + 1 >= trg->depth) {
1740 free(delta_buf);
1741 return 0;
1742 }
1743 }
1744 /*
1746 * Handle memory allocation outside of the cache
1747 * accounting lock. Compiler will optimize the strangeness
1748 * away when NO_PTHREADS is defined.
1749 */
1750 free(trg_entry->delta_data);
1751 cache_lock();
1752 if (trg_entry->delta_data) {
1753 delta_cache_size -= trg_entry->delta_size;
1754 trg_entry->delta_data = NULL;
1755 }
1756 if (delta_cacheable(src_size, trg_size, delta_size)) {
1757 delta_cache_size += delta_size;
1758 cache_unlock();
1759 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1760 } else {
1761 cache_unlock();
1762 free(delta_buf);
1763 }
1764 trg_entry->delta = src_entry;
1766 trg_entry->delta_size = delta_size;
1767 trg->depth = src->depth + 1;
1768 return 1;
1770}
1771static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1773{
1774 struct object_entry *child = me->delta_child;
1775 unsigned int m = n;
1776 while (child) {
1777 unsigned int c = check_delta_limit(child, n + 1);
1778 if (m < c)
1779 m = c;
1780 child = child->delta_sibling;
1781 }
1782 return m;
1783}
1784static unsigned long free_unpacked(struct unpacked *n)
1786{
1787 unsigned long freed_mem = sizeof_delta_index(n->index);
1788 free_delta_index(n->index);
1789 n->index = NULL;
1790 if (n->data) {
1791 freed_mem += n->entry->size;
1792 free(n->data);
1793 n->data = NULL;
1794 }
1795 n->entry = NULL;
1796 n->depth = 0;
1797 return freed_mem;
1798}
1799static void find_deltas(struct object_entry **list, unsigned *list_size,
1801 int window, int depth, unsigned *processed)
1802{
1803 uint32_t i, idx = 0, count = 0;
1804 struct unpacked *array;
1805 unsigned long mem_usage = 0;
1806 array = xcalloc(window, sizeof(struct unpacked));
1808 for (;;) {
1810 struct object_entry *entry;
1811 struct unpacked *n = array + idx;
1812 int j, max_depth, best_base = -1;
1813 progress_lock();
1815 if (!*list_size) {
1816 progress_unlock();
1817 break;
1818 }
1819 entry = *list++;
1820 (*list_size)--;
1821 if (!entry->preferred_base) {
1822 (*processed)++;
1823 display_progress(progress_state, *processed);
1824 }
1825 progress_unlock();
1826 mem_usage -= free_unpacked(n);
1828 n->entry = entry;
1829 while (window_memory_limit &&
1831 mem_usage > window_memory_limit &&
1832 count > 1) {
1833 uint32_t tail = (idx + window - count) % window;
1834 mem_usage -= free_unpacked(array + tail);
1835 count--;
1836 }
1837 /* We do not compute delta to *create* objects we are not
1839 * going to pack.
1840 */
1841 if (entry->preferred_base)
1842 goto next;
1843 /*
1845 * If the current object is at pack edge, take the depth the
1846 * objects that depend on the current object into account
1847 * otherwise they would become too deep.
1848 */
1849 max_depth = depth;
1850 if (entry->delta_child) {
1851 max_depth -= check_delta_limit(entry, 0);
1852 if (max_depth <= 0)
1853 goto next;
1854 }
1855 j = window;
1857 while (--j > 0) {
1858 int ret;
1859 uint32_t other_idx = idx + j;
1860 struct unpacked *m;
1861 if (other_idx >= window)
1862 other_idx -= window;
1863 m = array + other_idx;
1864 if (!m->entry)
1865 break;
1866 ret = try_delta(n, m, max_depth, &mem_usage);
1867 if (ret < 0)
1868 break;
1869 else if (ret > 0)
1870 best_base = other_idx;
1871 }
1872 /*
1874 * If we decided to cache the delta data, then it is best
1875 * to compress it right away. First because we have to do
1876 * it anyway, and doing it here while we're threaded will
1877 * save a lot of time in the non threaded write phase,
1878 * as well as allow for caching more deltas within
1879 * the same cache size limit.
1880 * ...
1881 * But only if not writing to stdout, since in that case
1882 * the network is most likely throttling writes anyway,
1883 * and therefore it is best to go to the write phase ASAP
1884 * instead, as we can afford spending more time compressing
1885 * between writes at that moment.
1886 */
1887 if (entry->delta_data && !pack_to_stdout) {
1888 entry->z_delta_size = do_compress(&entry->delta_data,
1889 entry->delta_size);
1890 cache_lock();
1891 delta_cache_size -= entry->delta_size;
1892 delta_cache_size += entry->z_delta_size;
1893 cache_unlock();
1894 }
1895 /* if we made n a delta, and if n is already at max
1897 * depth, leaving it in the window is pointless. we
1898 * should evict it first.
1899 */
1900 if (entry->delta && max_depth <= n->depth)
1901 continue;
1902 /*
1904 * Move the best delta base up in the window, after the
1905 * currently deltified object, to keep it longer. It will
1906 * be the first base object to be attempted next.
1907 */
1908 if (entry->delta) {
1909 struct unpacked swap = array[best_base];
1910 int dist = (window + idx - best_base) % window;
1911 int dst = best_base;
1912 while (dist--) {
1913 int src = (dst + 1) % window;
1914 array[dst] = array[src];
1915 dst = src;
1916 }
1917 array[dst] = swap;
1918 }
1919 next:
1921 idx++;
1922 if (count + 1 < window)
1923 count++;
1924 if (idx >= window)
1925 idx = 0;
1926 }
1927 for (i = 0; i < window; ++i) {
1929 free_delta_index(array[i].index);
1930 free(array[i].data);
1931 }
1932 free(array);
1933}
1934#ifndef NO_PTHREADS
1936static void try_to_free_from_threads(size_t size)
1938{
1939 read_lock();
1940 release_pack_memory(size);
1941 read_unlock();
1942}
1943static try_to_free_t old_try_to_free_routine;
1945/*
1947 * The main thread waits on the condition that (at least) one of the workers
1948 * has stopped working (which is indicated in the .working member of
1949 * struct thread_params).
1950 * When a work thread has completed its work, it sets .working to 0 and
1951 * signals the main thread and waits on the condition that .data_ready
1952 * becomes 1.
1953 */
1954struct thread_params {
1956 pthread_t thread;
1957 struct object_entry **list;
1958 unsigned list_size;
1959 unsigned remaining;
1960 int window;
1961 int depth;
1962 int working;
1963 int data_ready;
1964 pthread_mutex_t mutex;
1965 pthread_cond_t cond;
1966 unsigned *processed;
1967};
1968static pthread_cond_t progress_cond;
1970/*
1972 * Mutex and conditional variable can't be statically-initialized on Windows.
1973 */
1974static void init_threaded_search(void)
1975{
1976 init_recursive_mutex(&read_mutex);
1977 pthread_mutex_init(&cache_mutex, NULL);
1978 pthread_mutex_init(&progress_mutex, NULL);
1979 pthread_cond_init(&progress_cond, NULL);
1980 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
1981}
1982static void cleanup_threaded_search(void)
1984{
1985 set_try_to_free_routine(old_try_to_free_routine);
1986 pthread_cond_destroy(&progress_cond);
1987 pthread_mutex_destroy(&read_mutex);
1988 pthread_mutex_destroy(&cache_mutex);
1989 pthread_mutex_destroy(&progress_mutex);
1990}
1991static void *threaded_find_deltas(void *arg)
1993{
1994 struct thread_params *me = arg;
1995 while (me->remaining) {
1997 find_deltas(me->list, &me->remaining,
1998 me->window, me->depth, me->processed);
1999 progress_lock();
2001 me->working = 0;
2002 pthread_cond_signal(&progress_cond);
2003 progress_unlock();
2004 /*
2006 * We must not set ->data_ready before we wait on the
2007 * condition because the main thread may have set it to 1
2008 * before we get here. In order to be sure that new
2009 * work is available if we see 1 in ->data_ready, it
2010 * was initialized to 0 before this thread was spawned
2011 * and we reset it to 0 right away.
2012 */
2013 pthread_mutex_lock(&me->mutex);
2014 while (!me->data_ready)
2015 pthread_cond_wait(&me->cond, &me->mutex);
2016 me->data_ready = 0;
2017 pthread_mutex_unlock(&me->mutex);
2018 }
2019 /* leave ->working 1 so that this doesn't get more work assigned */
2020 return NULL;
2021}
2022static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2024 int window, int depth, unsigned *processed)
2025{
2026 struct thread_params *p;
2027 int i, ret, active_threads = 0;
2028 init_threaded_search();
2030 if (delta_search_threads <= 1) {
2032 find_deltas(list, &list_size, window, depth, processed);
2033 cleanup_threaded_search();
2034 return;
2035 }
2036 if (progress > pack_to_stdout)
2037 fprintf(stderr, "Delta compression using up to %d threads.\n",
2038 delta_search_threads);
2039 p = xcalloc(delta_search_threads, sizeof(*p));
2040 /* Partition the work amongst work threads. */
2042 for (i = 0; i < delta_search_threads; i++) {
2043 unsigned sub_size = list_size / (delta_search_threads - i);
2044 /* don't use too small segments or no deltas will be found */
2046 if (sub_size < 2*window && i+1 < delta_search_threads)
2047 sub_size = 0;
2048 p[i].window = window;
2050 p[i].depth = depth;
2051 p[i].processed = processed;
2052 p[i].working = 1;
2053 p[i].data_ready = 0;
2054 /* try to split chunks on "path" boundaries */
2056 while (sub_size && sub_size < list_size &&
2057 list[sub_size]->hash &&
2058 list[sub_size]->hash == list[sub_size-1]->hash)
2059 sub_size++;
2060 p[i].list = list;
2062 p[i].list_size = sub_size;
2063 p[i].remaining = sub_size;
2064 list += sub_size;
2066 list_size -= sub_size;
2067 }
2068 /* Start work threads. */
2070 for (i = 0; i < delta_search_threads; i++) {
2071 if (!p[i].list_size)
2072 continue;
2073 pthread_mutex_init(&p[i].mutex, NULL);
2074 pthread_cond_init(&p[i].cond, NULL);
2075 ret = pthread_create(&p[i].thread, NULL,
2076 threaded_find_deltas, &p[i]);
2077 if (ret)
2078 die("unable to create thread: %s", strerror(ret));
2079 active_threads++;
2080 }
2081 /*
2083 * Now let's wait for work completion. Each time a thread is done
2084 * with its work, we steal half of the remaining work from the
2085 * thread with the largest number of unprocessed objects and give
2086 * it to that newly idle thread. This ensure good load balancing
2087 * until the remaining object list segments are simply too short
2088 * to be worth splitting anymore.
2089 */
2090 while (active_threads) {
2091 struct thread_params *target = NULL;
2092 struct thread_params *victim = NULL;
2093 unsigned sub_size = 0;
2094 progress_lock();
2096 for (;;) {
2097 for (i = 0; !target && i < delta_search_threads; i++)
2098 if (!p[i].working)
2099 target = &p[i];
2100 if (target)
2101 break;
2102 pthread_cond_wait(&progress_cond, &progress_mutex);
2103 }
2104 for (i = 0; i < delta_search_threads; i++)
2106 if (p[i].remaining > 2*window &&
2107 (!victim || victim->remaining < p[i].remaining))
2108 victim = &p[i];
2109 if (victim) {
2110 sub_size = victim->remaining / 2;
2111 list = victim->list + victim->list_size - sub_size;
2112 while (sub_size && list[0]->hash &&
2113 list[0]->hash == list[-1]->hash) {
2114 list++;
2115 sub_size--;
2116 }
2117 if (!sub_size) {
2118 /*
2119 * It is possible for some "paths" to have
2120 * so many objects that no hash boundary
2121 * might be found. Let's just steal the
2122 * exact half in that case.
2123 */
2124 sub_size = victim->remaining / 2;
2125 list -= sub_size;
2126 }
2127 target->list = list;
2128 victim->list_size -= sub_size;
2129 victim->remaining -= sub_size;
2130 }
2131 target->list_size = sub_size;
2132 target->remaining = sub_size;
2133 target->working = 1;
2134 progress_unlock();
2135 pthread_mutex_lock(&target->mutex);
2137 target->data_ready = 1;
2138 pthread_cond_signal(&target->cond);
2139 pthread_mutex_unlock(&target->mutex);
2140 if (!sub_size) {
2142 pthread_join(target->thread, NULL);
2143 pthread_cond_destroy(&target->cond);
2144 pthread_mutex_destroy(&target->mutex);
2145 active_threads--;
2146 }
2147 }
2148 cleanup_threaded_search();
2149 free(p);
2150}
2151#else
2153#define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2154#endif
2155static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2157{
2158 struct object_id peeled;
2159 if (starts_with(path, "refs/tags/") && /* is a tag? */
2161 !peel_ref(path, peeled.hash) && /* peelable? */
2162 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */
2163 add_object_entry(oid->hash, OBJ_TAG, NULL, 0);
2164 return 0;
2165}
2166static void prepare_pack(int window, int depth)
2168{
2169 struct object_entry **delta_list;
2170 uint32_t i, nr_deltas;
2171 unsigned n;
2172 get_object_details();
2174 /*
2176 * If we're locally repacking then we need to be doubly careful
2177 * from now on in order to make sure no stealth corruption gets
2178 * propagated to the new pack. Clients receiving streamed packs
2179 * should validate everything they get anyway so no need to incur
2180 * the additional cost here in that case.
2181 */
2182 if (!pack_to_stdout)
2183 do_check_packed_object_crc = 1;
2184 if (!to_pack.nr_objects || !window || !depth)
2186 return;
2187 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2189 nr_deltas = n = 0;
2190 for (i = 0; i < to_pack.nr_objects; i++) {
2192 struct object_entry *entry = to_pack.objects + i;
2193 if (entry->delta)
2195 /* This happens if we decided to reuse existing
2196 * delta from a pack. "reuse_delta &&" is implied.
2197 */
2198 continue;
2199 if (entry->size < 50)
2201 continue;
2202 if (entry->no_try_delta)
2204 continue;
2205 if (!entry->preferred_base) {
2207 nr_deltas++;
2208 if (entry->type < 0)
2209 die("unable to get type of object %s",
2210 sha1_to_hex(entry->idx.sha1));
2211 } else {
2212 if (entry->type < 0) {
2213 /*
2214 * This object is not found, but we
2215 * don't have to include it anyway.
2216 */
2217 continue;
2218 }
2219 }
2220 delta_list[n++] = entry;
2222 }
2223 if (nr_deltas && n > 1) {
2225 unsigned nr_done = 0;
2226 if (progress)
2227 progress_state = start_progress(_("Compressing objects"),
2228 nr_deltas);
2229 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
2230 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2231 stop_progress(&progress_state);
2232 if (nr_done != nr_deltas)
2233 die("inconsistency with delta count");
2234 }
2235 free(delta_list);
2236}
2237static int git_pack_config(const char *k, const char *v, void *cb)
2239{
2240 if (!strcmp(k, "pack.window")) {
2241 window = git_config_int(k, v);
2242 return 0;
2243 }
2244 if (!strcmp(k, "pack.windowmemory")) {
2245 window_memory_limit = git_config_ulong(k, v);
2246 return 0;
2247 }
2248 if (!strcmp(k, "pack.depth")) {
2249 depth = git_config_int(k, v);
2250 return 0;
2251 }
2252 if (!strcmp(k, "pack.compression")) {
2253 int level = git_config_int(k, v);
2254 if (level == -1)
2255 level = Z_DEFAULT_COMPRESSION;
2256 else if (level < 0 || level > Z_BEST_COMPRESSION)
2257 die("bad pack compression level %d", level);
2258 pack_compression_level = level;
2259 pack_compression_seen = 1;
2260 return 0;
2261 }
2262 if (!strcmp(k, "pack.deltacachesize")) {
2263 max_delta_cache_size = git_config_int(k, v);
2264 return 0;
2265 }
2266 if (!strcmp(k, "pack.deltacachelimit")) {
2267 cache_max_small_delta_size = git_config_int(k, v);
2268 return 0;
2269 }
2270 if (!strcmp(k, "pack.writebitmaphashcache")) {
2271 if (git_config_bool(k, v))
2272 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2273 else
2274 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2275 }
2276 if (!strcmp(k, "pack.usebitmaps")) {
2277 use_bitmap_index = git_config_bool(k, v);
2278 return 0;
2279 }
2280 if (!strcmp(k, "pack.threads")) {
2281 delta_search_threads = git_config_int(k, v);
2282 if (delta_search_threads < 0)
2283 die("invalid number of threads specified (%d)",
2284 delta_search_threads);
2285#ifdef NO_PTHREADS
2286 if (delta_search_threads != 1)
2287 warning("no threads support, ignoring %s", k);
2288#endif
2289 return 0;
2290 }
2291 if (!strcmp(k, "pack.indexversion")) {
2292 pack_idx_opts.version = git_config_int(k, v);
2293 if (pack_idx_opts.version > 2)
2294 die("bad pack.indexversion=%"PRIu32,
2295 pack_idx_opts.version);
2296 return 0;
2297 }
2298 return git_default_config(k, v, cb);
2299}
2300static void read_object_list_from_stdin(void)
2302{
2303 char line[40 + 1 + PATH_MAX + 2];
2304 unsigned char sha1[20];
2305 for (;;) {
2307 if (!fgets(line, sizeof(line), stdin)) {
2308 if (feof(stdin))
2309 break;
2310 if (!ferror(stdin))
2311 die("fgets returned NULL, not EOF, not error!");
2312 if (errno != EINTR)
2313 die_errno("fgets");
2314 clearerr(stdin);
2315 continue;
2316 }
2317 if (line[0] == '-') {
2318 if (get_sha1_hex(line+1, sha1))
2319 die("expected edge sha1, got garbage:\n %s",
2320 line);
2321 add_preferred_base(sha1);
2322 continue;
2323 }
2324 if (get_sha1_hex(line, sha1))
2325 die("expected sha1, got garbage:\n %s", line);
2326 add_preferred_base_object(line+41);
2328 add_object_entry(sha1, 0, line+41, 0);
2329 }
2330}
2331#define OBJECT_ADDED (1u<<20)
2333static void show_commit(struct commit *commit, void *data)
2335{
2336 add_object_entry(commit->object.oid.hash, OBJ_COMMIT, NULL, 0);
2337 commit->object.flags |= OBJECT_ADDED;
2338 if (write_bitmap_index)
2340 index_commit_for_bitmap(commit);
2341}
2342static void show_object(struct object *obj, const char *name, void *data)
2344{
2345 add_preferred_base_object(name);
2346 add_object_entry(obj->oid.hash, obj->type, name, 0);
2347 obj->flags |= OBJECT_ADDED;
2348}
2349static void show_edge(struct commit *commit)
2351{
2352 add_preferred_base(commit->object.oid.hash);
2353}
2354struct in_pack_object {
2356 off_t offset;
2357 struct object *object;
2358};
2359struct in_pack {
2361 int alloc;
2362 int nr;
2363 struct in_pack_object *array;
2364};
2365static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2367{
2368 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2369 in_pack->array[in_pack->nr].object = object;
2370 in_pack->nr++;
2371}
2372/*
2374 * Compare the objects in the offset order, in order to emulate the
2375 * "git rev-list --objects" output that produced the pack originally.
2376 */
2377static int ofscmp(const void *a_, const void *b_)
2378{
2379 struct in_pack_object *a = (struct in_pack_object *)a_;
2380 struct in_pack_object *b = (struct in_pack_object *)b_;
2381 if (a->offset < b->offset)
2383 return -1;
2384 else if (a->offset > b->offset)
2385 return 1;
2386 else
2387 return oidcmp(&a->object->oid, &b->object->oid);
2388}
2389static void add_objects_in_unpacked_packs(struct rev_info *revs)
2391{
2392 struct packed_git *p;
2393 struct in_pack in_pack;
2394 uint32_t i;
2395 memset(&in_pack, 0, sizeof(in_pack));
2397 for (p = packed_git; p; p = p->next) {
2399 const unsigned char *sha1;
2400 struct object *o;
2401 if (!p->pack_local || p->pack_keep)
2403 continue;
2404 if (open_pack_index(p))
2405 die("cannot open pack index");
2406 ALLOC_GROW(in_pack.array,
2408 in_pack.nr + p->num_objects,
2409 in_pack.alloc);
2410 for (i = 0; i < p->num_objects; i++) {
2412 sha1 = nth_packed_object_sha1(p, i);
2413 o = lookup_unknown_object(sha1);
2414 if (!(o->flags & OBJECT_ADDED))
2415 mark_in_pack_object(o, p, &in_pack);
2416 o->flags |= OBJECT_ADDED;
2417 }
2418 }
2419 if (in_pack.nr) {
2421 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2422 ofscmp);
2423 for (i = 0; i < in_pack.nr; i++) {
2424 struct object *o = in_pack.array[i].object;
2425 add_object_entry(o->oid.hash, o->type, "", 0);
2426 }
2427 }
2428 free(in_pack.array);
2429}
2430static int add_loose_object(const unsigned char *sha1, const char *path,
2432 void *data)
2433{
2434 enum object_type type = sha1_object_info(sha1, NULL);
2435 if (type < 0) {
2437 warning("loose object at %s could not be examined", path);
2438 return 0;
2439 }
2440 add_object_entry(sha1, type, "", 0);
2442 return 0;
2443}
2444/*
2446 * We actually don't even have to worry about reachability here.
2447 * add_object_entry will weed out duplicates, so we just add every
2448 * loose object we find.
2449 */
2450static void add_unreachable_loose_objects(void)
2451{
2452 for_each_loose_file_in_objdir(get_object_directory(),
2453 add_loose_object,
2454 NULL, NULL, NULL);
2455}
2456static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2458{
2459 static struct packed_git *last_found = (void *)1;
2460 struct packed_git *p;
2461 p = (last_found != (void *)1) ? last_found : packed_git;
2463 while (p) {
2465 if ((!p->pack_local || p->pack_keep) &&
2466 find_pack_entry_one(sha1, p)) {
2467 last_found = p;
2468 return 1;
2469 }
2470 if (p == last_found)
2471 p = packed_git;
2472 else
2473 p = p->next;
2474 if (p == last_found)
2475 p = p->next;
2476 }
2477 return 0;
2478}
2479/*
2481 * Store a list of sha1s that are should not be discarded
2482 * because they are either written too recently, or are
2483 * reachable from another object that was.
2484 *
2485 * This is filled by get_object_list.
2486 */
2487static struct sha1_array recent_objects;
2488static int loosened_object_can_be_discarded(const unsigned char *sha1,
2490 unsigned long mtime)
2491{
2492 if (!unpack_unreachable_expiration)
2493 return 0;
2494 if (mtime > unpack_unreachable_expiration)
2495 return 0;
2496 if (sha1_array_lookup(&recent_objects, sha1) >= 0)
2497 return 0;
2498 return 1;
2499}
2500static void loosen_unused_packed_objects(struct rev_info *revs)
2502{
2503 struct packed_git *p;
2504 uint32_t i;
2505 const unsigned char *sha1;
2506 for (p = packed_git; p; p = p->next) {
2508 if (!p->pack_local || p->pack_keep)
2509 continue;
2510 if (open_pack_index(p))
2512 die("cannot open pack index");
2513 for (i = 0; i < p->num_objects; i++) {
2515 sha1 = nth_packed_object_sha1(p, i);
2516 if (!packlist_find(&to_pack, sha1, NULL) &&
2517 !has_sha1_pack_kept_or_nonlocal(sha1) &&
2518 !loosened_object_can_be_discarded(sha1, p->mtime))
2519 if (force_object_loose(sha1, p->mtime))
2520 die("unable to force loose object");
2521 }
2522 }
2523}
2524/*
2526 * This tracks any options which a reader of the pack might
2527 * not understand, and which would therefore prevent blind reuse
2528 * of what we have on disk.
2529 */
2530static int pack_options_allow_reuse(void)
2531{
2532 return allow_ofs_delta;
2533}
2534static int get_object_list_from_bitmap(struct rev_info *revs)
2536{
2537 if (prepare_bitmap_walk(revs) < 0)
2538 return -1;
2539 if (pack_options_allow_reuse() &&
2541 !reuse_partial_packfile_from_bitmap(
2542 &reuse_packfile,
2543 &reuse_packfile_objects,
2544 &reuse_packfile_offset)) {
2545 assert(reuse_packfile_objects);
2546 nr_result += reuse_packfile_objects;
2547 display_progress(progress_state, nr_result);
2548 }
2549 traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2551 return 0;
2552}
2553static void record_recent_object(struct object *obj,
2555 const char *name,
2556 void *data)
2557{
2558 sha1_array_append(&recent_objects, obj->oid.hash);
2559}
2560static void record_recent_commit(struct commit *commit, void *data)
2562{
2563 sha1_array_append(&recent_objects, commit->object.oid.hash);
2564}
2565static void get_object_list(int ac, const char **av)
2567{
2568 struct rev_info revs;
2569 char line[1000];
2570 int flags = 0;
2571 init_revisions(&revs, NULL);
2573 save_commit_buffer = 0;
2574 setup_revisions(ac, av, &revs, NULL);
2575 /* make sure shallows are read */
2577 is_repository_shallow();
2578 while (fgets(line, sizeof(line), stdin) != NULL) {
2580 int len = strlen(line);
2581 if (len && line[len - 1] == '\n')
2582 line[--len] = 0;
2583 if (!len)
2584 break;
2585 if (*line == '-') {
2586 if (!strcmp(line, "--not")) {
2587 flags ^= UNINTERESTING;
2588 write_bitmap_index = 0;
2589 continue;
2590 }
2591 if (starts_with(line, "--shallow ")) {
2592 unsigned char sha1[20];
2593 if (get_sha1_hex(line + 10, sha1))
2594 die("not an SHA-1 '%s'", line + 10);
2595 register_shallow(sha1);
2596 use_bitmap_index = 0;
2597 continue;
2598 }
2599 die("not a rev '%s'", line);
2600 }
2601 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2602 die("bad revision '%s'", line);
2603 }
2604 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2606 return;
2607 if (prepare_revision_walk(&revs))
2609 die("revision walk setup failed");
2610 mark_edges_uninteresting(&revs, show_edge);
2611 traverse_commit_list(&revs, show_commit, show_object, NULL);
2612 if (unpack_unreachable_expiration) {
2614 revs.ignore_missing_links = 1;
2615 if (add_unseen_recent_objects_to_traversal(&revs,
2616 unpack_unreachable_expiration))
2617 die("unable to add recent objects");
2618 if (prepare_revision_walk(&revs))
2619 die("revision walk setup failed");
2620 traverse_commit_list(&revs, record_recent_commit,
2621 record_recent_object, NULL);
2622 }
2623 if (keep_unreachable)
2625 add_objects_in_unpacked_packs(&revs);
2626 if (pack_loose_unreachable)
2627 add_unreachable_loose_objects();
2628 if (unpack_unreachable)
2629 loosen_unused_packed_objects(&revs);
2630 sha1_array_clear(&recent_objects);
2632}
2633static int option_parse_index_version(const struct option *opt,
2635 const char *arg, int unset)
2636{
2637 char *c;
2638 const char *val = arg;
2639 pack_idx_opts.version = strtoul(val, &c, 10);
2640 if (pack_idx_opts.version > 2)
2641 die(_("unsupported index version %s"), val);
2642 if (*c == ',' && c[1])
2643 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2644 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2645 die(_("bad index version '%s'"), val);
2646 return 0;
2647}
2648static int option_parse_unpack_unreachable(const struct option *opt,
2650 const char *arg, int unset)
2651{
2652 if (unset) {
2653 unpack_unreachable = 0;
2654 unpack_unreachable_expiration = 0;
2655 }
2656 else {
2657 unpack_unreachable = 1;
2658 if (arg)
2659 unpack_unreachable_expiration = approxidate(arg);
2660 }
2661 return 0;
2662}
2663int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2665{
2666 int use_internal_rev_list = 0;
2667 int thin = 0;
2668 int shallow = 0;
2669 int all_progress_implied = 0;
2670 struct argv_array rp = ARGV_ARRAY_INIT;
2671 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2672 int rev_list_index = 0;
2673 struct option pack_objects_options[] = {
2674 OPT_SET_INT('q', "quiet", &progress,
2675 N_("do not show progress meter"), 0),
2676 OPT_SET_INT(0, "progress", &progress,
2677 N_("show progress meter"), 1),
2678 OPT_SET_INT(0, "all-progress", &progress,
2679 N_("show progress meter during object writing phase"), 2),
2680 OPT_BOOL(0, "all-progress-implied",
2681 &all_progress_implied,
2682 N_("similar to --all-progress when progress meter is shown")),
2683 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2684 N_("write the pack index file in the specified idx format version"),
2685 0, option_parse_index_version },
2686 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
2687 N_("maximum size of each output pack file")),
2688 OPT_BOOL(0, "local", &local,
2689 N_("ignore borrowed objects from alternate object store")),
2690 OPT_BOOL(0, "incremental", &incremental,
2691 N_("ignore packed objects")),
2692 OPT_INTEGER(0, "window", &window,
2693 N_("limit pack window by objects")),
2694 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
2695 N_("limit pack window by memory in addition to object limit")),
2696 OPT_INTEGER(0, "depth", &depth,
2697 N_("maximum length of delta chain allowed in the resulting pack")),
2698 OPT_BOOL(0, "reuse-delta", &reuse_delta,
2699 N_("reuse existing deltas")),
2700 OPT_BOOL(0, "reuse-object", &reuse_object,
2701 N_("reuse existing objects")),
2702 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
2703 N_("use OFS_DELTA objects")),
2704 OPT_INTEGER(0, "threads", &delta_search_threads,
2705 N_("use threads when searching for best delta matches")),
2706 OPT_BOOL(0, "non-empty", &non_empty,
2707 N_("do not create an empty pack output")),
2708 OPT_BOOL(0, "revs", &use_internal_rev_list,
2709 N_("read revision arguments from standard input")),
2710 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
2711 N_("limit the objects to those that are not yet packed"),
2712 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2713 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
2714 N_("include objects reachable from any reference"),
2715 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2716 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
2717 N_("include objects referred by reflog entries"),
2718 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2719 { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL,
2720 N_("include objects referred to by the index"),
2721 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2722 OPT_BOOL(0, "stdout", &pack_to_stdout,
2723 N_("output pack to stdout")),
2724 OPT_BOOL(0, "include-tag", &include_tag,
2725 N_("include tag objects that refer to objects to be packed")),
2726 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
2727 N_("keep unreachable objects")),
2728 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
2729 N_("pack loose unreachable objects")),
2730 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
2731 N_("unpack unreachable objects newer than <time>"),
2732 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
2733 OPT_BOOL(0, "thin", &thin,
2734 N_("create thin packs")),
2735 OPT_BOOL(0, "shallow", &shallow,
2736 N_("create packs suitable for shallow fetches")),
2737 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
2738 N_("ignore packs that have companion .keep file")),
2739 OPT_INTEGER(0, "compression", &pack_compression_level,
2740 N_("pack compression level")),
2741 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
2742 N_("do not hide commits by grafts"), 0),
2743 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
2744 N_("use a bitmap index if available to speed up counting objects")),
2745 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
2746 N_("write a bitmap index together with the pack index")),
2747 OPT_END(),
2748 };
2749 check_replace_refs = 0;
2751 reset_pack_idx_option(&pack_idx_opts);
2753 git_config(git_pack_config, NULL);
2754 if (!pack_compression_seen && core_compression_seen)
2755 pack_compression_level = core_compression_level;
2756 progress = isatty(2);
2758 argc = parse_options(argc, argv, prefix, pack_objects_options,
2759 pack_usage, 0);
2760 if (argc) {
2762 base_name = argv[0];
2763 argc--;
2764 }
2765 if (pack_to_stdout != !base_name || argc)
2766 usage_with_options(pack_usage, pack_objects_options);
2767 argv_array_push(&rp, "pack-objects");
2769 if (thin) {
2770 use_internal_rev_list = 1;
2771 argv_array_push(&rp, shallow
2772 ? "--objects-edge-aggressive"
2773 : "--objects-edge");
2774 } else
2775 argv_array_push(&rp, "--objects");
2776 if (rev_list_all) {
2778 use_internal_rev_list = 1;
2779 argv_array_push(&rp, "--all");
2780 }
2781 if (rev_list_reflog) {
2782 use_internal_rev_list = 1;
2783 argv_array_push(&rp, "--reflog");
2784 }
2785 if (rev_list_index) {
2786 use_internal_rev_list = 1;
2787 argv_array_push(&rp, "--indexed-objects");
2788 }
2789 if (rev_list_unpacked) {
2790 use_internal_rev_list = 1;
2791 argv_array_push(&rp, "--unpacked");
2792 }
2793 if (!reuse_object)
2795 reuse_delta = 0;
2796 if (pack_compression_level == -1)
2797 pack_compression_level = Z_DEFAULT_COMPRESSION;
2798 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
2799 die("bad pack compression level %d", pack_compression_level);
2800 if (!delta_search_threads) /* --threads=0 means autodetect */
2802 delta_search_threads = online_cpus();
2803#ifdef NO_PTHREADS
2805 if (delta_search_threads != 1)
2806 warning("no threads support, ignoring --threads");
2807#endif
2808 if (!pack_to_stdout && !pack_size_limit)
2809 pack_size_limit = pack_size_limit_cfg;
2810 if (pack_to_stdout && pack_size_limit)
2811 die("--max-pack-size cannot be used to build a pack for transfer.");
2812 if (pack_size_limit && pack_size_limit < 1024*1024) {
2813 warning("minimum pack size limit is 1 MiB");
2814 pack_size_limit = 1024*1024;
2815 }
2816 if (!pack_to_stdout && thin)
2818 die("--thin cannot be used to build an indexable pack.");
2819 if (keep_unreachable && unpack_unreachable)
2821 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2822 if (!rev_list_all || !rev_list_reflog || !rev_list_index)
2823 unpack_unreachable_expiration = 0;
2824 if (!use_internal_rev_list || !pack_to_stdout || is_repository_shallow())
2826 use_bitmap_index = 0;
2827 if (pack_to_stdout || !rev_list_all)
2829 write_bitmap_index = 0;
2830 if (progress && all_progress_implied)
2832 progress = 2;
2833 prepare_packed_git();
2835 if (ignore_packed_keep) {
2836 struct packed_git *p;
2837 for (p = packed_git; p; p = p->next)
2838 if (p->pack_local && p->pack_keep)
2839 break;
2840 if (!p) /* no keep-able packs found */
2841 ignore_packed_keep = 0;
2842 }
2843 if (local) {
2844 /*
2845 * unlike ignore_packed_keep above, we do not want to
2846 * unset "local" based on looking at packs, as it
2847 * also covers non-local objects
2848 */
2849 struct packed_git *p;
2850 for (p = packed_git; p; p = p->next) {
2851 if (!p->pack_local) {
2852 have_non_local_packs = 1;
2853 break;
2854 }
2855 }
2856 }
2857 if (progress)
2859 progress_state = start_progress(_("Counting objects"), 0);
2860 if (!use_internal_rev_list)
2861 read_object_list_from_stdin();
2862 else {
2863 get_object_list(rp.argc, rp.argv);
2864 argv_array_clear(&rp);
2865 }
2866 cleanup_preferred_base();
2867 if (include_tag && nr_result)
2868 for_each_ref(add_ref_tag, NULL);
2869 stop_progress(&progress_state);
2870 if (non_empty && !nr_result)
2872 return 0;
2873 if (nr_result)
2874 prepare_pack(window, depth);
2875 write_pack_file();
2876 if (progress)
2877 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2878 " reused %"PRIu32" (delta %"PRIu32")\n",
2879 written, written_delta, reused, reused_delta);
2880 return 0;
2881}