worktree: handle broken symrefs in find_shared_symref()
[git.git] / builtin / pack-objects.c
blob6e77dfd44439f4c5f928a73dce3233ff6e63c3b4
1 #include "builtin.h"
2 #include "cache.h"
3 #include "config.h"
4 #include "attr.h"
5 #include "object.h"
6 #include "blob.h"
7 #include "commit.h"
8 #include "tag.h"
9 #include "tree.h"
10 #include "delta.h"
11 #include "pack.h"
12 #include "pack-revindex.h"
13 #include "csum-file.h"
14 #include "tree-walk.h"
15 #include "diff.h"
16 #include "revision.h"
17 #include "list-objects.h"
18 #include "pack-objects.h"
19 #include "progress.h"
20 #include "refs.h"
21 #include "streaming.h"
22 #include "thread-utils.h"
23 #include "pack-bitmap.h"
24 #include "reachable.h"
25 #include "sha1-array.h"
26 #include "argv-array.h"
27 #include "mru.h"
28 #include "packfile.h"
30 static const char *pack_usage[] = {
31 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
32 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
33 NULL
37 * Objects we are going to pack are collected in the `to_pack` structure.
38 * It contains an array (dynamically expanded) of the object data, and a map
39 * that can resolve SHA1s to their position in the array.
41 static struct packing_data to_pack;
43 static struct pack_idx_entry **written_list;
44 static uint32_t nr_result, nr_written;
46 static int non_empty;
47 static int reuse_delta = 1, reuse_object = 1;
48 static int keep_unreachable, unpack_unreachable, include_tag;
49 static timestamp_t unpack_unreachable_expiration;
50 static int pack_loose_unreachable;
51 static int local;
52 static int have_non_local_packs;
53 static int incremental;
54 static int ignore_packed_keep;
55 static int allow_ofs_delta;
56 static struct pack_idx_option pack_idx_opts;
57 static const char *base_name;
58 static int progress = 1;
59 static int window = 10;
60 static unsigned long pack_size_limit;
61 static int depth = 50;
62 static int delta_search_threads;
63 static int pack_to_stdout;
64 static int num_preferred_base;
65 static struct progress *progress_state;
67 static struct packed_git *reuse_packfile;
68 static uint32_t reuse_packfile_objects;
69 static off_t reuse_packfile_offset;
71 static int use_bitmap_index_default = 1;
72 static int use_bitmap_index = -1;
73 static int write_bitmap_index;
74 static uint16_t write_bitmap_options;
76 static unsigned long delta_cache_size = 0;
77 static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
78 static unsigned long cache_max_small_delta_size = 1000;
80 static unsigned long window_memory_limit = 0;
83 * stats
85 static uint32_t written, written_delta;
86 static uint32_t reused, reused_delta;
89 * Indexed commits
91 static struct commit **indexed_commits;
92 static unsigned int indexed_commits_nr;
93 static unsigned int indexed_commits_alloc;
95 static void index_commit_for_bitmap(struct commit *commit)
97 if (indexed_commits_nr >= indexed_commits_alloc) {
98 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
99 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
102 indexed_commits[indexed_commits_nr++] = commit;
105 static void *get_delta(struct object_entry *entry)
107 unsigned long size, base_size, delta_size;
108 void *buf, *base_buf, *delta_buf;
109 enum object_type type;
111 buf = read_sha1_file(entry->idx.oid.hash, &type, &size);
112 if (!buf)
113 die("unable to read %s", oid_to_hex(&entry->idx.oid));
114 base_buf = read_sha1_file(entry->delta->idx.oid.hash, &type,
115 &base_size);
116 if (!base_buf)
117 die("unable to read %s",
118 oid_to_hex(&entry->delta->idx.oid));
119 delta_buf = diff_delta(base_buf, base_size,
120 buf, size, &delta_size, 0);
121 if (!delta_buf || delta_size != entry->delta_size)
122 die("delta size changed");
123 free(buf);
124 free(base_buf);
125 return delta_buf;
128 static unsigned long do_compress(void **pptr, unsigned long size)
130 git_zstream stream;
131 void *in, *out;
132 unsigned long maxsize;
134 git_deflate_init(&stream, pack_compression_level);
135 maxsize = git_deflate_bound(&stream, size);
137 in = *pptr;
138 out = xmalloc(maxsize);
139 *pptr = out;
141 stream.next_in = in;
142 stream.avail_in = size;
143 stream.next_out = out;
144 stream.avail_out = maxsize;
145 while (git_deflate(&stream, Z_FINISH) == Z_OK)
146 ; /* nothing */
147 git_deflate_end(&stream);
149 free(in);
150 return stream.total_out;
153 static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f,
154 const unsigned char *sha1)
156 git_zstream stream;
157 unsigned char ibuf[1024 * 16];
158 unsigned char obuf[1024 * 16];
159 unsigned long olen = 0;
161 git_deflate_init(&stream, pack_compression_level);
163 for (;;) {
164 ssize_t readlen;
165 int zret = Z_OK;
166 readlen = read_istream(st, ibuf, sizeof(ibuf));
167 if (readlen == -1)
168 die(_("unable to read %s"), sha1_to_hex(sha1));
170 stream.next_in = ibuf;
171 stream.avail_in = readlen;
172 while ((stream.avail_in || readlen == 0) &&
173 (zret == Z_OK || zret == Z_BUF_ERROR)) {
174 stream.next_out = obuf;
175 stream.avail_out = sizeof(obuf);
176 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
177 sha1write(f, obuf, stream.next_out - obuf);
178 olen += stream.next_out - obuf;
180 if (stream.avail_in)
181 die(_("deflate error (%d)"), zret);
182 if (readlen == 0) {
183 if (zret != Z_STREAM_END)
184 die(_("deflate error (%d)"), zret);
185 break;
188 git_deflate_end(&stream);
189 return olen;
193 * we are going to reuse the existing object data as is. make
194 * sure it is not corrupt.
196 static int check_pack_inflate(struct packed_git *p,
197 struct pack_window **w_curs,
198 off_t offset,
199 off_t len,
200 unsigned long expect)
202 git_zstream stream;
203 unsigned char fakebuf[4096], *in;
204 int st;
206 memset(&stream, 0, sizeof(stream));
207 git_inflate_init(&stream);
208 do {
209 in = use_pack(p, w_curs, offset, &stream.avail_in);
210 stream.next_in = in;
211 stream.next_out = fakebuf;
212 stream.avail_out = sizeof(fakebuf);
213 st = git_inflate(&stream, Z_FINISH);
214 offset += stream.next_in - in;
215 } while (st == Z_OK || st == Z_BUF_ERROR);
216 git_inflate_end(&stream);
217 return (st == Z_STREAM_END &&
218 stream.total_out == expect &&
219 stream.total_in == len) ? 0 : -1;
222 static void copy_pack_data(struct sha1file *f,
223 struct packed_git *p,
224 struct pack_window **w_curs,
225 off_t offset,
226 off_t len)
228 unsigned char *in;
229 unsigned long avail;
231 while (len) {
232 in = use_pack(p, w_curs, offset, &avail);
233 if (avail > len)
234 avail = (unsigned long)len;
235 sha1write(f, in, avail);
236 offset += avail;
237 len -= avail;
241 /* Return 0 if we will bust the pack-size limit */
242 static unsigned long write_no_reuse_object(struct sha1file *f, struct object_entry *entry,
243 unsigned long limit, int usable_delta)
245 unsigned long size, datalen;
246 unsigned char header[MAX_PACK_OBJECT_HEADER],
247 dheader[MAX_PACK_OBJECT_HEADER];
248 unsigned hdrlen;
249 enum object_type type;
250 void *buf;
251 struct git_istream *st = NULL;
253 if (!usable_delta) {
254 if (entry->type == OBJ_BLOB &&
255 entry->size > big_file_threshold &&
256 (st = open_istream(entry->idx.oid.hash, &type, &size, NULL)) != NULL)
257 buf = NULL;
258 else {
259 buf = read_sha1_file(entry->idx.oid.hash, &type,
260 &size);
261 if (!buf)
262 die(_("unable to read %s"),
263 oid_to_hex(&entry->idx.oid));
266 * make sure no cached delta data remains from a
267 * previous attempt before a pack split occurred.
269 FREE_AND_NULL(entry->delta_data);
270 entry->z_delta_size = 0;
271 } else if (entry->delta_data) {
272 size = entry->delta_size;
273 buf = entry->delta_data;
274 entry->delta_data = NULL;
275 type = (allow_ofs_delta && entry->delta->idx.offset) ?
276 OBJ_OFS_DELTA : OBJ_REF_DELTA;
277 } else {
278 buf = get_delta(entry);
279 size = entry->delta_size;
280 type = (allow_ofs_delta && entry->delta->idx.offset) ?
281 OBJ_OFS_DELTA : OBJ_REF_DELTA;
284 if (st) /* large blob case, just assume we don't compress well */
285 datalen = size;
286 else if (entry->z_delta_size)
287 datalen = entry->z_delta_size;
288 else
289 datalen = do_compress(&buf, size);
292 * The object header is a byte of 'type' followed by zero or
293 * more bytes of length.
295 hdrlen = encode_in_pack_object_header(header, sizeof(header),
296 type, size);
298 if (type == OBJ_OFS_DELTA) {
300 * Deltas with relative base contain an additional
301 * encoding of the relative offset for the delta
302 * base from this object's position in the pack.
304 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
305 unsigned pos = sizeof(dheader) - 1;
306 dheader[pos] = ofs & 127;
307 while (ofs >>= 7)
308 dheader[--pos] = 128 | (--ofs & 127);
309 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
310 if (st)
311 close_istream(st);
312 free(buf);
313 return 0;
315 sha1write(f, header, hdrlen);
316 sha1write(f, dheader + pos, sizeof(dheader) - pos);
317 hdrlen += sizeof(dheader) - pos;
318 } else if (type == OBJ_REF_DELTA) {
320 * Deltas with a base reference contain
321 * an additional 20 bytes for the base sha1.
323 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
324 if (st)
325 close_istream(st);
326 free(buf);
327 return 0;
329 sha1write(f, header, hdrlen);
330 sha1write(f, entry->delta->idx.oid.hash, 20);
331 hdrlen += 20;
332 } else {
333 if (limit && hdrlen + datalen + 20 >= limit) {
334 if (st)
335 close_istream(st);
336 free(buf);
337 return 0;
339 sha1write(f, header, hdrlen);
341 if (st) {
342 datalen = write_large_blob_data(st, f, entry->idx.oid.hash);
343 close_istream(st);
344 } else {
345 sha1write(f, buf, datalen);
346 free(buf);
349 return hdrlen + datalen;
352 /* Return 0 if we will bust the pack-size limit */
353 static off_t write_reuse_object(struct sha1file *f, struct object_entry *entry,
354 unsigned long limit, int usable_delta)
356 struct packed_git *p = entry->in_pack;
357 struct pack_window *w_curs = NULL;
358 struct revindex_entry *revidx;
359 off_t offset;
360 enum object_type type = entry->type;
361 off_t datalen;
362 unsigned char header[MAX_PACK_OBJECT_HEADER],
363 dheader[MAX_PACK_OBJECT_HEADER];
364 unsigned hdrlen;
366 if (entry->delta)
367 type = (allow_ofs_delta && entry->delta->idx.offset) ?
368 OBJ_OFS_DELTA : OBJ_REF_DELTA;
369 hdrlen = encode_in_pack_object_header(header, sizeof(header),
370 type, entry->size);
372 offset = entry->in_pack_offset;
373 revidx = find_pack_revindex(p, offset);
374 datalen = revidx[1].offset - offset;
375 if (!pack_to_stdout && p->index_version > 1 &&
376 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
377 error("bad packed object CRC for %s",
378 oid_to_hex(&entry->idx.oid));
379 unuse_pack(&w_curs);
380 return write_no_reuse_object(f, entry, limit, usable_delta);
383 offset += entry->in_pack_header_size;
384 datalen -= entry->in_pack_header_size;
386 if (!pack_to_stdout && p->index_version == 1 &&
387 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
388 error("corrupt packed object for %s",
389 oid_to_hex(&entry->idx.oid));
390 unuse_pack(&w_curs);
391 return write_no_reuse_object(f, entry, limit, usable_delta);
394 if (type == OBJ_OFS_DELTA) {
395 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
396 unsigned pos = sizeof(dheader) - 1;
397 dheader[pos] = ofs & 127;
398 while (ofs >>= 7)
399 dheader[--pos] = 128 | (--ofs & 127);
400 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
401 unuse_pack(&w_curs);
402 return 0;
404 sha1write(f, header, hdrlen);
405 sha1write(f, dheader + pos, sizeof(dheader) - pos);
406 hdrlen += sizeof(dheader) - pos;
407 reused_delta++;
408 } else if (type == OBJ_REF_DELTA) {
409 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
410 unuse_pack(&w_curs);
411 return 0;
413 sha1write(f, header, hdrlen);
414 sha1write(f, entry->delta->idx.oid.hash, 20);
415 hdrlen += 20;
416 reused_delta++;
417 } else {
418 if (limit && hdrlen + datalen + 20 >= limit) {
419 unuse_pack(&w_curs);
420 return 0;
422 sha1write(f, header, hdrlen);
424 copy_pack_data(f, p, &w_curs, offset, datalen);
425 unuse_pack(&w_curs);
426 reused++;
427 return hdrlen + datalen;
430 /* Return 0 if we will bust the pack-size limit */
431 static off_t write_object(struct sha1file *f,
432 struct object_entry *entry,
433 off_t write_offset)
435 unsigned long limit;
436 off_t len;
437 int usable_delta, to_reuse;
439 if (!pack_to_stdout)
440 crc32_begin(f);
442 /* apply size limit if limited packsize and not first object */
443 if (!pack_size_limit || !nr_written)
444 limit = 0;
445 else if (pack_size_limit <= write_offset)
447 * the earlier object did not fit the limit; avoid
448 * mistaking this with unlimited (i.e. limit = 0).
450 limit = 1;
451 else
452 limit = pack_size_limit - write_offset;
454 if (!entry->delta)
455 usable_delta = 0; /* no delta */
456 else if (!pack_size_limit)
457 usable_delta = 1; /* unlimited packfile */
458 else if (entry->delta->idx.offset == (off_t)-1)
459 usable_delta = 0; /* base was written to another pack */
460 else if (entry->delta->idx.offset)
461 usable_delta = 1; /* base already exists in this pack */
462 else
463 usable_delta = 0; /* base could end up in another pack */
465 if (!reuse_object)
466 to_reuse = 0; /* explicit */
467 else if (!entry->in_pack)
468 to_reuse = 0; /* can't reuse what we don't have */
469 else if (entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA)
470 /* check_object() decided it for us ... */
471 to_reuse = usable_delta;
472 /* ... but pack split may override that */
473 else if (entry->type != entry->in_pack_type)
474 to_reuse = 0; /* pack has delta which is unusable */
475 else if (entry->delta)
476 to_reuse = 0; /* we want to pack afresh */
477 else
478 to_reuse = 1; /* we have it in-pack undeltified,
479 * and we do not need to deltify it.
482 if (!to_reuse)
483 len = write_no_reuse_object(f, entry, limit, usable_delta);
484 else
485 len = write_reuse_object(f, entry, limit, usable_delta);
486 if (!len)
487 return 0;
489 if (usable_delta)
490 written_delta++;
491 written++;
492 if (!pack_to_stdout)
493 entry->idx.crc32 = crc32_end(f);
494 return len;
497 enum write_one_status {
498 WRITE_ONE_SKIP = -1, /* already written */
499 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
500 WRITE_ONE_WRITTEN = 1, /* normal */
501 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
504 static enum write_one_status write_one(struct sha1file *f,
505 struct object_entry *e,
506 off_t *offset)
508 off_t size;
509 int recursing;
512 * we set offset to 1 (which is an impossible value) to mark
513 * the fact that this object is involved in "write its base
514 * first before writing a deltified object" recursion.
516 recursing = (e->idx.offset == 1);
517 if (recursing) {
518 warning("recursive delta detected for object %s",
519 oid_to_hex(&e->idx.oid));
520 return WRITE_ONE_RECURSIVE;
521 } else if (e->idx.offset || e->preferred_base) {
522 /* offset is non zero if object is written already. */
523 return WRITE_ONE_SKIP;
526 /* if we are deltified, write out base object first. */
527 if (e->delta) {
528 e->idx.offset = 1; /* now recurse */
529 switch (write_one(f, e->delta, offset)) {
530 case WRITE_ONE_RECURSIVE:
531 /* we cannot depend on this one */
532 e->delta = NULL;
533 break;
534 default:
535 break;
536 case WRITE_ONE_BREAK:
537 e->idx.offset = recursing;
538 return WRITE_ONE_BREAK;
542 e->idx.offset = *offset;
543 size = write_object(f, e, *offset);
544 if (!size) {
545 e->idx.offset = recursing;
546 return WRITE_ONE_BREAK;
548 written_list[nr_written++] = &e->idx;
550 /* make sure off_t is sufficiently large not to wrap */
551 if (signed_add_overflows(*offset, size))
552 die("pack too large for current definition of off_t");
553 *offset += size;
554 return WRITE_ONE_WRITTEN;
557 static int mark_tagged(const char *path, const struct object_id *oid, int flag,
558 void *cb_data)
560 unsigned char peeled[20];
561 struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
563 if (entry)
564 entry->tagged = 1;
565 if (!peel_ref(path, peeled)) {
566 entry = packlist_find(&to_pack, peeled, NULL);
567 if (entry)
568 entry->tagged = 1;
570 return 0;
573 static inline void add_to_write_order(struct object_entry **wo,
574 unsigned int *endp,
575 struct object_entry *e)
577 if (e->filled)
578 return;
579 wo[(*endp)++] = e;
580 e->filled = 1;
583 static void add_descendants_to_write_order(struct object_entry **wo,
584 unsigned int *endp,
585 struct object_entry *e)
587 int add_to_order = 1;
588 while (e) {
589 if (add_to_order) {
590 struct object_entry *s;
591 /* add this node... */
592 add_to_write_order(wo, endp, e);
593 /* all its siblings... */
594 for (s = e->delta_sibling; s; s = s->delta_sibling) {
595 add_to_write_order(wo, endp, s);
598 /* drop down a level to add left subtree nodes if possible */
599 if (e->delta_child) {
600 add_to_order = 1;
601 e = e->delta_child;
602 } else {
603 add_to_order = 0;
604 /* our sibling might have some children, it is next */
605 if (e->delta_sibling) {
606 e = e->delta_sibling;
607 continue;
609 /* go back to our parent node */
610 e = e->delta;
611 while (e && !e->delta_sibling) {
612 /* we're on the right side of a subtree, keep
613 * going up until we can go right again */
614 e = e->delta;
616 if (!e) {
617 /* done- we hit our original root node */
618 return;
620 /* pass it off to sibling at this level */
621 e = e->delta_sibling;
626 static void add_family_to_write_order(struct object_entry **wo,
627 unsigned int *endp,
628 struct object_entry *e)
630 struct object_entry *root;
632 for (root = e; root->delta; root = root->delta)
633 ; /* nothing */
634 add_descendants_to_write_order(wo, endp, root);
637 static struct object_entry **compute_write_order(void)
639 unsigned int i, wo_end, last_untagged;
641 struct object_entry **wo;
642 struct object_entry *objects = to_pack.objects;
644 for (i = 0; i < to_pack.nr_objects; i++) {
645 objects[i].tagged = 0;
646 objects[i].filled = 0;
647 objects[i].delta_child = NULL;
648 objects[i].delta_sibling = NULL;
652 * Fully connect delta_child/delta_sibling network.
653 * Make sure delta_sibling is sorted in the original
654 * recency order.
656 for (i = to_pack.nr_objects; i > 0;) {
657 struct object_entry *e = &objects[--i];
658 if (!e->delta)
659 continue;
660 /* Mark me as the first child */
661 e->delta_sibling = e->delta->delta_child;
662 e->delta->delta_child = e;
666 * Mark objects that are at the tip of tags.
668 for_each_tag_ref(mark_tagged, NULL);
671 * Give the objects in the original recency order until
672 * we see a tagged tip.
674 ALLOC_ARRAY(wo, to_pack.nr_objects);
675 for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
676 if (objects[i].tagged)
677 break;
678 add_to_write_order(wo, &wo_end, &objects[i]);
680 last_untagged = i;
683 * Then fill all the tagged tips.
685 for (; i < to_pack.nr_objects; i++) {
686 if (objects[i].tagged)
687 add_to_write_order(wo, &wo_end, &objects[i]);
691 * And then all remaining commits and tags.
693 for (i = last_untagged; i < to_pack.nr_objects; i++) {
694 if (objects[i].type != OBJ_COMMIT &&
695 objects[i].type != OBJ_TAG)
696 continue;
697 add_to_write_order(wo, &wo_end, &objects[i]);
701 * And then all the trees.
703 for (i = last_untagged; i < to_pack.nr_objects; i++) {
704 if (objects[i].type != OBJ_TREE)
705 continue;
706 add_to_write_order(wo, &wo_end, &objects[i]);
710 * Finally all the rest in really tight order
712 for (i = last_untagged; i < to_pack.nr_objects; i++) {
713 if (!objects[i].filled)
714 add_family_to_write_order(wo, &wo_end, &objects[i]);
717 if (wo_end != to_pack.nr_objects)
718 die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
720 return wo;
723 static off_t write_reused_pack(struct sha1file *f)
725 unsigned char buffer[8192];
726 off_t to_write, total;
727 int fd;
729 if (!is_pack_valid(reuse_packfile))
730 die("packfile is invalid: %s", reuse_packfile->pack_name);
732 fd = git_open(reuse_packfile->pack_name);
733 if (fd < 0)
734 die_errno("unable to open packfile for reuse: %s",
735 reuse_packfile->pack_name);
737 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
738 die_errno("unable to seek in reused packfile");
740 if (reuse_packfile_offset < 0)
741 reuse_packfile_offset = reuse_packfile->pack_size - 20;
743 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
745 while (to_write) {
746 int read_pack = xread(fd, buffer, sizeof(buffer));
748 if (read_pack <= 0)
749 die_errno("unable to read from reused packfile");
751 if (read_pack > to_write)
752 read_pack = to_write;
754 sha1write(f, buffer, read_pack);
755 to_write -= read_pack;
758 * We don't know the actual number of objects written,
759 * only how many bytes written, how many bytes total, and
760 * how many objects total. So we can fake it by pretending all
761 * objects we are writing are the same size. This gives us a
762 * smooth progress meter, and at the end it matches the true
763 * answer.
765 written = reuse_packfile_objects *
766 (((double)(total - to_write)) / total);
767 display_progress(progress_state, written);
770 close(fd);
771 written = reuse_packfile_objects;
772 display_progress(progress_state, written);
773 return reuse_packfile_offset - sizeof(struct pack_header);
776 static const char no_split_warning[] = N_(
777 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
780 static void write_pack_file(void)
782 uint32_t i = 0, j;
783 struct sha1file *f;
784 off_t offset;
785 uint32_t nr_remaining = nr_result;
786 time_t last_mtime = 0;
787 struct object_entry **write_order;
789 if (progress > pack_to_stdout)
790 progress_state = start_progress(_("Writing objects"), nr_result);
791 ALLOC_ARRAY(written_list, to_pack.nr_objects);
792 write_order = compute_write_order();
794 do {
795 unsigned char sha1[20];
796 char *pack_tmp_name = NULL;
798 if (pack_to_stdout)
799 f = sha1fd_throughput(1, "<stdout>", progress_state);
800 else
801 f = create_tmp_packfile(&pack_tmp_name);
803 offset = write_pack_header(f, nr_remaining);
805 if (reuse_packfile) {
806 off_t packfile_size;
807 assert(pack_to_stdout);
809 packfile_size = write_reused_pack(f);
810 offset += packfile_size;
813 nr_written = 0;
814 for (; i < to_pack.nr_objects; i++) {
815 struct object_entry *e = write_order[i];
816 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
817 break;
818 display_progress(progress_state, written);
822 * Did we write the wrong # entries in the header?
823 * If so, rewrite it like in fast-import
825 if (pack_to_stdout) {
826 sha1close(f, sha1, CSUM_CLOSE);
827 } else if (nr_written == nr_remaining) {
828 sha1close(f, sha1, CSUM_FSYNC);
829 } else {
830 int fd = sha1close(f, sha1, 0);
831 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
832 nr_written, sha1, offset);
833 close(fd);
834 if (write_bitmap_index) {
835 warning(_(no_split_warning));
836 write_bitmap_index = 0;
840 if (!pack_to_stdout) {
841 struct stat st;
842 struct strbuf tmpname = STRBUF_INIT;
845 * Packs are runtime accessed in their mtime
846 * order since newer packs are more likely to contain
847 * younger objects. So if we are creating multiple
848 * packs then we should modify the mtime of later ones
849 * to preserve this property.
851 if (stat(pack_tmp_name, &st) < 0) {
852 warning_errno("failed to stat %s", pack_tmp_name);
853 } else if (!last_mtime) {
854 last_mtime = st.st_mtime;
855 } else {
856 struct utimbuf utb;
857 utb.actime = st.st_atime;
858 utb.modtime = --last_mtime;
859 if (utime(pack_tmp_name, &utb) < 0)
860 warning_errno("failed utime() on %s", pack_tmp_name);
863 strbuf_addf(&tmpname, "%s-", base_name);
865 if (write_bitmap_index) {
866 bitmap_writer_set_checksum(sha1);
867 bitmap_writer_build_type_index(written_list, nr_written);
870 finish_tmp_packfile(&tmpname, pack_tmp_name,
871 written_list, nr_written,
872 &pack_idx_opts, sha1);
874 if (write_bitmap_index) {
875 strbuf_addf(&tmpname, "%s.bitmap", sha1_to_hex(sha1));
877 stop_progress(&progress_state);
879 bitmap_writer_show_progress(progress);
880 bitmap_writer_reuse_bitmaps(&to_pack);
881 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
882 bitmap_writer_build(&to_pack);
883 bitmap_writer_finish(written_list, nr_written,
884 tmpname.buf, write_bitmap_options);
885 write_bitmap_index = 0;
888 strbuf_release(&tmpname);
889 free(pack_tmp_name);
890 puts(sha1_to_hex(sha1));
893 /* mark written objects as written to previous pack */
894 for (j = 0; j < nr_written; j++) {
895 written_list[j]->offset = (off_t)-1;
897 nr_remaining -= nr_written;
898 } while (nr_remaining && i < to_pack.nr_objects);
900 free(written_list);
901 free(write_order);
902 stop_progress(&progress_state);
903 if (written != nr_result)
904 die("wrote %"PRIu32" objects while expecting %"PRIu32,
905 written, nr_result);
908 static int no_try_delta(const char *path)
910 static struct attr_check *check;
912 if (!check)
913 check = attr_check_initl("delta", NULL);
914 if (git_check_attr(path, check))
915 return 0;
916 if (ATTR_FALSE(check->items[0].value))
917 return 1;
918 return 0;
922 * When adding an object, check whether we have already added it
923 * to our packing list. If so, we can skip. However, if we are
924 * being asked to excludei t, but the previous mention was to include
925 * it, make sure to adjust its flags and tweak our numbers accordingly.
927 * As an optimization, we pass out the index position where we would have
928 * found the item, since that saves us from having to look it up again a
929 * few lines later when we want to add the new entry.
931 static int have_duplicate_entry(const unsigned char *sha1,
932 int exclude,
933 uint32_t *index_pos)
935 struct object_entry *entry;
937 entry = packlist_find(&to_pack, sha1, index_pos);
938 if (!entry)
939 return 0;
941 if (exclude) {
942 if (!entry->preferred_base)
943 nr_result--;
944 entry->preferred_base = 1;
947 return 1;
950 static int want_found_object(int exclude, struct packed_git *p)
952 if (exclude)
953 return 1;
954 if (incremental)
955 return 0;
958 * When asked to do --local (do not include an object that appears in a
959 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
960 * an object that appears in a pack marked with .keep), finding a pack
961 * that matches the criteria is sufficient for us to decide to omit it.
962 * However, even if this pack does not satisfy the criteria, we need to
963 * make sure no copy of this object appears in _any_ pack that makes us
964 * to omit the object, so we need to check all the packs.
966 * We can however first check whether these options can possible matter;
967 * if they do not matter we know we want the object in generated pack.
968 * Otherwise, we signal "-1" at the end to tell the caller that we do
969 * not know either way, and it needs to check more packs.
971 if (!ignore_packed_keep &&
972 (!local || !have_non_local_packs))
973 return 1;
975 if (local && !p->pack_local)
976 return 0;
977 if (ignore_packed_keep && p->pack_local && p->pack_keep)
978 return 0;
980 /* we don't know yet; keep looking for more packs */
981 return -1;
985 * Check whether we want the object in the pack (e.g., we do not want
986 * objects found in non-local stores if the "--local" option was used).
988 * If the caller already knows an existing pack it wants to take the object
989 * from, that is passed in *found_pack and *found_offset; otherwise this
990 * function finds if there is any pack that has the object and returns the pack
991 * and its offset in these variables.
993 static int want_object_in_pack(const unsigned char *sha1,
994 int exclude,
995 struct packed_git **found_pack,
996 off_t *found_offset)
998 struct mru_entry *entry;
999 int want;
1001 if (!exclude && local && has_loose_object_nonlocal(sha1))
1002 return 0;
1005 * If we already know the pack object lives in, start checks from that
1006 * pack - in the usual case when neither --local was given nor .keep files
1007 * are present we will determine the answer right now.
1009 if (*found_pack) {
1010 want = want_found_object(exclude, *found_pack);
1011 if (want != -1)
1012 return want;
1015 for (entry = packed_git_mru.head; entry; entry = entry->next) {
1016 struct packed_git *p = entry->item;
1017 off_t offset;
1019 if (p == *found_pack)
1020 offset = *found_offset;
1021 else
1022 offset = find_pack_entry_one(sha1, p);
1024 if (offset) {
1025 if (!*found_pack) {
1026 if (!is_pack_valid(p))
1027 continue;
1028 *found_offset = offset;
1029 *found_pack = p;
1031 want = want_found_object(exclude, p);
1032 if (!exclude && want > 0)
1033 mru_mark(&packed_git_mru, entry);
1034 if (want != -1)
1035 return want;
1039 return 1;
1042 static void create_object_entry(const unsigned char *sha1,
1043 enum object_type type,
1044 uint32_t hash,
1045 int exclude,
1046 int no_try_delta,
1047 uint32_t index_pos,
1048 struct packed_git *found_pack,
1049 off_t found_offset)
1051 struct object_entry *entry;
1053 entry = packlist_alloc(&to_pack, sha1, index_pos);
1054 entry->hash = hash;
1055 if (type)
1056 entry->type = type;
1057 if (exclude)
1058 entry->preferred_base = 1;
1059 else
1060 nr_result++;
1061 if (found_pack) {
1062 entry->in_pack = found_pack;
1063 entry->in_pack_offset = found_offset;
1066 entry->no_try_delta = no_try_delta;
1069 static const char no_closure_warning[] = N_(
1070 "disabling bitmap writing, as some objects are not being packed"
1073 static int add_object_entry(const unsigned char *sha1, enum object_type type,
1074 const char *name, int exclude)
1076 struct packed_git *found_pack = NULL;
1077 off_t found_offset = 0;
1078 uint32_t index_pos;
1080 if (have_duplicate_entry(sha1, exclude, &index_pos))
1081 return 0;
1083 if (!want_object_in_pack(sha1, exclude, &found_pack, &found_offset)) {
1084 /* The pack is missing an object, so it will not have closure */
1085 if (write_bitmap_index) {
1086 warning(_(no_closure_warning));
1087 write_bitmap_index = 0;
1089 return 0;
1092 create_object_entry(sha1, type, pack_name_hash(name),
1093 exclude, name && no_try_delta(name),
1094 index_pos, found_pack, found_offset);
1096 display_progress(progress_state, nr_result);
1097 return 1;
1100 static int add_object_entry_from_bitmap(const unsigned char *sha1,
1101 enum object_type type,
1102 int flags, uint32_t name_hash,
1103 struct packed_git *pack, off_t offset)
1105 uint32_t index_pos;
1107 if (have_duplicate_entry(sha1, 0, &index_pos))
1108 return 0;
1110 if (!want_object_in_pack(sha1, 0, &pack, &offset))
1111 return 0;
1113 create_object_entry(sha1, type, name_hash, 0, 0, index_pos, pack, offset);
1115 display_progress(progress_state, nr_result);
1116 return 1;
1119 struct pbase_tree_cache {
1120 unsigned char sha1[20];
1121 int ref;
1122 int temporary;
1123 void *tree_data;
1124 unsigned long tree_size;
1127 static struct pbase_tree_cache *(pbase_tree_cache[256]);
1128 static int pbase_tree_cache_ix(const unsigned char *sha1)
1130 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
1132 static int pbase_tree_cache_ix_incr(int ix)
1134 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1137 static struct pbase_tree {
1138 struct pbase_tree *next;
1139 /* This is a phony "cache" entry; we are not
1140 * going to evict it or find it through _get()
1141 * mechanism -- this is for the toplevel node that
1142 * would almost always change with any commit.
1144 struct pbase_tree_cache pcache;
1145 } *pbase_tree;
1147 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1149 struct pbase_tree_cache *ent, *nent;
1150 void *data;
1151 unsigned long size;
1152 enum object_type type;
1153 int neigh;
1154 int my_ix = pbase_tree_cache_ix(sha1);
1155 int available_ix = -1;
1157 /* pbase-tree-cache acts as a limited hashtable.
1158 * your object will be found at your index or within a few
1159 * slots after that slot if it is cached.
1161 for (neigh = 0; neigh < 8; neigh++) {
1162 ent = pbase_tree_cache[my_ix];
1163 if (ent && !hashcmp(ent->sha1, sha1)) {
1164 ent->ref++;
1165 return ent;
1167 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1168 ((0 <= available_ix) &&
1169 (!ent && pbase_tree_cache[available_ix])))
1170 available_ix = my_ix;
1171 if (!ent)
1172 break;
1173 my_ix = pbase_tree_cache_ix_incr(my_ix);
1176 /* Did not find one. Either we got a bogus request or
1177 * we need to read and perhaps cache.
1179 data = read_sha1_file(sha1, &type, &size);
1180 if (!data)
1181 return NULL;
1182 if (type != OBJ_TREE) {
1183 free(data);
1184 return NULL;
1187 /* We need to either cache or return a throwaway copy */
1189 if (available_ix < 0)
1190 ent = NULL;
1191 else {
1192 ent = pbase_tree_cache[available_ix];
1193 my_ix = available_ix;
1196 if (!ent) {
1197 nent = xmalloc(sizeof(*nent));
1198 nent->temporary = (available_ix < 0);
1200 else {
1201 /* evict and reuse */
1202 free(ent->tree_data);
1203 nent = ent;
1205 hashcpy(nent->sha1, sha1);
1206 nent->tree_data = data;
1207 nent->tree_size = size;
1208 nent->ref = 1;
1209 if (!nent->temporary)
1210 pbase_tree_cache[my_ix] = nent;
1211 return nent;
1214 static void pbase_tree_put(struct pbase_tree_cache *cache)
1216 if (!cache->temporary) {
1217 cache->ref--;
1218 return;
1220 free(cache->tree_data);
1221 free(cache);
1224 static int name_cmp_len(const char *name)
1226 int i;
1227 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1229 return i;
1232 static void add_pbase_object(struct tree_desc *tree,
1233 const char *name,
1234 int cmplen,
1235 const char *fullname)
1237 struct name_entry entry;
1238 int cmp;
1240 while (tree_entry(tree,&entry)) {
1241 if (S_ISGITLINK(entry.mode))
1242 continue;
1243 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1244 memcmp(name, entry.path, cmplen);
1245 if (cmp > 0)
1246 continue;
1247 if (cmp < 0)
1248 return;
1249 if (name[cmplen] != '/') {
1250 add_object_entry(entry.oid->hash,
1251 object_type(entry.mode),
1252 fullname, 1);
1253 return;
1255 if (S_ISDIR(entry.mode)) {
1256 struct tree_desc sub;
1257 struct pbase_tree_cache *tree;
1258 const char *down = name+cmplen+1;
1259 int downlen = name_cmp_len(down);
1261 tree = pbase_tree_get(entry.oid->hash);
1262 if (!tree)
1263 return;
1264 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1266 add_pbase_object(&sub, down, downlen, fullname);
1267 pbase_tree_put(tree);
1272 static unsigned *done_pbase_paths;
1273 static int done_pbase_paths_num;
1274 static int done_pbase_paths_alloc;
1275 static int done_pbase_path_pos(unsigned hash)
1277 int lo = 0;
1278 int hi = done_pbase_paths_num;
1279 while (lo < hi) {
1280 int mi = lo + (hi - lo) / 2;
1281 if (done_pbase_paths[mi] == hash)
1282 return mi;
1283 if (done_pbase_paths[mi] < hash)
1284 hi = mi;
1285 else
1286 lo = mi + 1;
1288 return -lo-1;
1291 static int check_pbase_path(unsigned hash)
1293 int pos = done_pbase_path_pos(hash);
1294 if (0 <= pos)
1295 return 1;
1296 pos = -pos - 1;
1297 ALLOC_GROW(done_pbase_paths,
1298 done_pbase_paths_num + 1,
1299 done_pbase_paths_alloc);
1300 done_pbase_paths_num++;
1301 if (pos < done_pbase_paths_num)
1302 MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,
1303 done_pbase_paths_num - pos - 1);
1304 done_pbase_paths[pos] = hash;
1305 return 0;
1308 static void add_preferred_base_object(const char *name)
1310 struct pbase_tree *it;
1311 int cmplen;
1312 unsigned hash = pack_name_hash(name);
1314 if (!num_preferred_base || check_pbase_path(hash))
1315 return;
1317 cmplen = name_cmp_len(name);
1318 for (it = pbase_tree; it; it = it->next) {
1319 if (cmplen == 0) {
1320 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1322 else {
1323 struct tree_desc tree;
1324 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1325 add_pbase_object(&tree, name, cmplen, name);
1330 static void add_preferred_base(unsigned char *sha1)
1332 struct pbase_tree *it;
1333 void *data;
1334 unsigned long size;
1335 unsigned char tree_sha1[20];
1337 if (window <= num_preferred_base++)
1338 return;
1340 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1341 if (!data)
1342 return;
1344 for (it = pbase_tree; it; it = it->next) {
1345 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1346 free(data);
1347 return;
1351 it = xcalloc(1, sizeof(*it));
1352 it->next = pbase_tree;
1353 pbase_tree = it;
1355 hashcpy(it->pcache.sha1, tree_sha1);
1356 it->pcache.tree_data = data;
1357 it->pcache.tree_size = size;
1360 static void cleanup_preferred_base(void)
1362 struct pbase_tree *it;
1363 unsigned i;
1365 it = pbase_tree;
1366 pbase_tree = NULL;
1367 while (it) {
1368 struct pbase_tree *this = it;
1369 it = this->next;
1370 free(this->pcache.tree_data);
1371 free(this);
1374 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1375 if (!pbase_tree_cache[i])
1376 continue;
1377 free(pbase_tree_cache[i]->tree_data);
1378 FREE_AND_NULL(pbase_tree_cache[i]);
1381 FREE_AND_NULL(done_pbase_paths);
1382 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1385 static void check_object(struct object_entry *entry)
1387 if (entry->in_pack) {
1388 struct packed_git *p = entry->in_pack;
1389 struct pack_window *w_curs = NULL;
1390 const unsigned char *base_ref = NULL;
1391 struct object_entry *base_entry;
1392 unsigned long used, used_0;
1393 unsigned long avail;
1394 off_t ofs;
1395 unsigned char *buf, c;
1397 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1400 * We want in_pack_type even if we do not reuse delta
1401 * since non-delta representations could still be reused.
1403 used = unpack_object_header_buffer(buf, avail,
1404 &entry->in_pack_type,
1405 &entry->size);
1406 if (used == 0)
1407 goto give_up;
1410 * Determine if this is a delta and if so whether we can
1411 * reuse it or not. Otherwise let's find out as cheaply as
1412 * possible what the actual type and size for this object is.
1414 switch (entry->in_pack_type) {
1415 default:
1416 /* Not a delta hence we've already got all we need. */
1417 entry->type = entry->in_pack_type;
1418 entry->in_pack_header_size = used;
1419 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1420 goto give_up;
1421 unuse_pack(&w_curs);
1422 return;
1423 case OBJ_REF_DELTA:
1424 if (reuse_delta && !entry->preferred_base)
1425 base_ref = use_pack(p, &w_curs,
1426 entry->in_pack_offset + used, NULL);
1427 entry->in_pack_header_size = used + 20;
1428 break;
1429 case OBJ_OFS_DELTA:
1430 buf = use_pack(p, &w_curs,
1431 entry->in_pack_offset + used, NULL);
1432 used_0 = 0;
1433 c = buf[used_0++];
1434 ofs = c & 127;
1435 while (c & 128) {
1436 ofs += 1;
1437 if (!ofs || MSB(ofs, 7)) {
1438 error("delta base offset overflow in pack for %s",
1439 oid_to_hex(&entry->idx.oid));
1440 goto give_up;
1442 c = buf[used_0++];
1443 ofs = (ofs << 7) + (c & 127);
1445 ofs = entry->in_pack_offset - ofs;
1446 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1447 error("delta base offset out of bound for %s",
1448 oid_to_hex(&entry->idx.oid));
1449 goto give_up;
1451 if (reuse_delta && !entry->preferred_base) {
1452 struct revindex_entry *revidx;
1453 revidx = find_pack_revindex(p, ofs);
1454 if (!revidx)
1455 goto give_up;
1456 base_ref = nth_packed_object_sha1(p, revidx->nr);
1458 entry->in_pack_header_size = used + used_0;
1459 break;
1462 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1464 * If base_ref was set above that means we wish to
1465 * reuse delta data, and we even found that base
1466 * in the list of objects we want to pack. Goodie!
1468 * Depth value does not matter - find_deltas() will
1469 * never consider reused delta as the base object to
1470 * deltify other objects against, in order to avoid
1471 * circular deltas.
1473 entry->type = entry->in_pack_type;
1474 entry->delta = base_entry;
1475 entry->delta_size = entry->size;
1476 entry->delta_sibling = base_entry->delta_child;
1477 base_entry->delta_child = entry;
1478 unuse_pack(&w_curs);
1479 return;
1482 if (entry->type) {
1484 * This must be a delta and we already know what the
1485 * final object type is. Let's extract the actual
1486 * object size from the delta header.
1488 entry->size = get_size_from_delta(p, &w_curs,
1489 entry->in_pack_offset + entry->in_pack_header_size);
1490 if (entry->size == 0)
1491 goto give_up;
1492 unuse_pack(&w_curs);
1493 return;
1497 * No choice but to fall back to the recursive delta walk
1498 * with sha1_object_info() to find about the object type
1499 * at this point...
1501 give_up:
1502 unuse_pack(&w_curs);
1505 entry->type = sha1_object_info(entry->idx.oid.hash, &entry->size);
1507 * The error condition is checked in prepare_pack(). This is
1508 * to permit a missing preferred base object to be ignored
1509 * as a preferred base. Doing so can result in a larger
1510 * pack file, but the transfer will still take place.
1514 static int pack_offset_sort(const void *_a, const void *_b)
1516 const struct object_entry *a = *(struct object_entry **)_a;
1517 const struct object_entry *b = *(struct object_entry **)_b;
1519 /* avoid filesystem trashing with loose objects */
1520 if (!a->in_pack && !b->in_pack)
1521 return oidcmp(&a->idx.oid, &b->idx.oid);
1523 if (a->in_pack < b->in_pack)
1524 return -1;
1525 if (a->in_pack > b->in_pack)
1526 return 1;
1527 return a->in_pack_offset < b->in_pack_offset ? -1 :
1528 (a->in_pack_offset > b->in_pack_offset);
1532 * Drop an on-disk delta we were planning to reuse. Naively, this would
1533 * just involve blanking out the "delta" field, but we have to deal
1534 * with some extra book-keeping:
1536 * 1. Removing ourselves from the delta_sibling linked list.
1538 * 2. Updating our size/type to the non-delta representation. These were
1539 * either not recorded initially (size) or overwritten with the delta type
1540 * (type) when check_object() decided to reuse the delta.
1542 * 3. Resetting our delta depth, as we are now a base object.
1544 static void drop_reused_delta(struct object_entry *entry)
1546 struct object_entry **p = &entry->delta->delta_child;
1547 struct object_info oi = OBJECT_INFO_INIT;
1549 while (*p) {
1550 if (*p == entry)
1551 *p = (*p)->delta_sibling;
1552 else
1553 p = &(*p)->delta_sibling;
1555 entry->delta = NULL;
1556 entry->depth = 0;
1558 oi.sizep = &entry->size;
1559 oi.typep = &entry->type;
1560 if (packed_object_info(entry->in_pack, entry->in_pack_offset, &oi) < 0) {
1562 * We failed to get the info from this pack for some reason;
1563 * fall back to sha1_object_info, which may find another copy.
1564 * And if that fails, the error will be recorded in entry->type
1565 * and dealt with in prepare_pack().
1567 entry->type = sha1_object_info(entry->idx.oid.hash,
1568 &entry->size);
1573 * Follow the chain of deltas from this entry onward, throwing away any links
1574 * that cause us to hit a cycle (as determined by the DFS state flags in
1575 * the entries).
1577 * We also detect too-long reused chains that would violate our --depth
1578 * limit.
1580 static void break_delta_chains(struct object_entry *entry)
1583 * The actual depth of each object we will write is stored as an int,
1584 * as it cannot exceed our int "depth" limit. But before we break
1585 * changes based no that limit, we may potentially go as deep as the
1586 * number of objects, which is elsewhere bounded to a uint32_t.
1588 uint32_t total_depth;
1589 struct object_entry *cur, *next;
1591 for (cur = entry, total_depth = 0;
1592 cur;
1593 cur = cur->delta, total_depth++) {
1594 if (cur->dfs_state == DFS_DONE) {
1596 * We've already seen this object and know it isn't
1597 * part of a cycle. We do need to append its depth
1598 * to our count.
1600 total_depth += cur->depth;
1601 break;
1605 * We break cycles before looping, so an ACTIVE state (or any
1606 * other cruft which made its way into the state variable)
1607 * is a bug.
1609 if (cur->dfs_state != DFS_NONE)
1610 die("BUG: confusing delta dfs state in first pass: %d",
1611 cur->dfs_state);
1614 * Now we know this is the first time we've seen the object. If
1615 * it's not a delta, we're done traversing, but we'll mark it
1616 * done to save time on future traversals.
1618 if (!cur->delta) {
1619 cur->dfs_state = DFS_DONE;
1620 break;
1624 * Mark ourselves as active and see if the next step causes
1625 * us to cycle to another active object. It's important to do
1626 * this _before_ we loop, because it impacts where we make the
1627 * cut, and thus how our total_depth counter works.
1628 * E.g., We may see a partial loop like:
1630 * A -> B -> C -> D -> B
1632 * Cutting B->C breaks the cycle. But now the depth of A is
1633 * only 1, and our total_depth counter is at 3. The size of the
1634 * error is always one less than the size of the cycle we
1635 * broke. Commits C and D were "lost" from A's chain.
1637 * If we instead cut D->B, then the depth of A is correct at 3.
1638 * We keep all commits in the chain that we examined.
1640 cur->dfs_state = DFS_ACTIVE;
1641 if (cur->delta->dfs_state == DFS_ACTIVE) {
1642 drop_reused_delta(cur);
1643 cur->dfs_state = DFS_DONE;
1644 break;
1649 * And now that we've gone all the way to the bottom of the chain, we
1650 * need to clear the active flags and set the depth fields as
1651 * appropriate. Unlike the loop above, which can quit when it drops a
1652 * delta, we need to keep going to look for more depth cuts. So we need
1653 * an extra "next" pointer to keep going after we reset cur->delta.
1655 for (cur = entry; cur; cur = next) {
1656 next = cur->delta;
1659 * We should have a chain of zero or more ACTIVE states down to
1660 * a final DONE. We can quit after the DONE, because either it
1661 * has no bases, or we've already handled them in a previous
1662 * call.
1664 if (cur->dfs_state == DFS_DONE)
1665 break;
1666 else if (cur->dfs_state != DFS_ACTIVE)
1667 die("BUG: confusing delta dfs state in second pass: %d",
1668 cur->dfs_state);
1671 * If the total_depth is more than depth, then we need to snip
1672 * the chain into two or more smaller chains that don't exceed
1673 * the maximum depth. Most of the resulting chains will contain
1674 * (depth + 1) entries (i.e., depth deltas plus one base), and
1675 * the last chain (i.e., the one containing entry) will contain
1676 * whatever entries are left over, namely
1677 * (total_depth % (depth + 1)) of them.
1679 * Since we are iterating towards decreasing depth, we need to
1680 * decrement total_depth as we go, and we need to write to the
1681 * entry what its final depth will be after all of the
1682 * snipping. Since we're snipping into chains of length (depth
1683 * + 1) entries, the final depth of an entry will be its
1684 * original depth modulo (depth + 1). Any time we encounter an
1685 * entry whose final depth is supposed to be zero, we snip it
1686 * from its delta base, thereby making it so.
1688 cur->depth = (total_depth--) % (depth + 1);
1689 if (!cur->depth)
1690 drop_reused_delta(cur);
1692 cur->dfs_state = DFS_DONE;
1696 static void get_object_details(void)
1698 uint32_t i;
1699 struct object_entry **sorted_by_offset;
1701 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1702 for (i = 0; i < to_pack.nr_objects; i++)
1703 sorted_by_offset[i] = to_pack.objects + i;
1704 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
1706 for (i = 0; i < to_pack.nr_objects; i++) {
1707 struct object_entry *entry = sorted_by_offset[i];
1708 check_object(entry);
1709 if (big_file_threshold < entry->size)
1710 entry->no_try_delta = 1;
1714 * This must happen in a second pass, since we rely on the delta
1715 * information for the whole list being completed.
1717 for (i = 0; i < to_pack.nr_objects; i++)
1718 break_delta_chains(&to_pack.objects[i]);
1720 free(sorted_by_offset);
1724 * We search for deltas in a list sorted by type, by filename hash, and then
1725 * by size, so that we see progressively smaller and smaller files.
1726 * That's because we prefer deltas to be from the bigger file
1727 * to the smaller -- deletes are potentially cheaper, but perhaps
1728 * more importantly, the bigger file is likely the more recent
1729 * one. The deepest deltas are therefore the oldest objects which are
1730 * less susceptible to be accessed often.
1732 static int type_size_sort(const void *_a, const void *_b)
1734 const struct object_entry *a = *(struct object_entry **)_a;
1735 const struct object_entry *b = *(struct object_entry **)_b;
1737 if (a->type > b->type)
1738 return -1;
1739 if (a->type < b->type)
1740 return 1;
1741 if (a->hash > b->hash)
1742 return -1;
1743 if (a->hash < b->hash)
1744 return 1;
1745 if (a->preferred_base > b->preferred_base)
1746 return -1;
1747 if (a->preferred_base < b->preferred_base)
1748 return 1;
1749 if (a->size > b->size)
1750 return -1;
1751 if (a->size < b->size)
1752 return 1;
1753 return a < b ? -1 : (a > b); /* newest first */
1756 struct unpacked {
1757 struct object_entry *entry;
1758 void *data;
1759 struct delta_index *index;
1760 unsigned depth;
1763 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1764 unsigned long delta_size)
1766 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1767 return 0;
1769 if (delta_size < cache_max_small_delta_size)
1770 return 1;
1772 /* cache delta, if objects are large enough compared to delta size */
1773 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1774 return 1;
1776 return 0;
1779 #ifndef NO_PTHREADS
1781 static pthread_mutex_t read_mutex;
1782 #define read_lock() pthread_mutex_lock(&read_mutex)
1783 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1785 static pthread_mutex_t cache_mutex;
1786 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1787 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1789 static pthread_mutex_t progress_mutex;
1790 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1791 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1793 #else
1795 #define read_lock() (void)0
1796 #define read_unlock() (void)0
1797 #define cache_lock() (void)0
1798 #define cache_unlock() (void)0
1799 #define progress_lock() (void)0
1800 #define progress_unlock() (void)0
1802 #endif
1804 static int try_delta(struct unpacked *trg, struct unpacked *src,
1805 unsigned max_depth, unsigned long *mem_usage)
1807 struct object_entry *trg_entry = trg->entry;
1808 struct object_entry *src_entry = src->entry;
1809 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1810 unsigned ref_depth;
1811 enum object_type type;
1812 void *delta_buf;
1814 /* Don't bother doing diffs between different types */
1815 if (trg_entry->type != src_entry->type)
1816 return -1;
1819 * We do not bother to try a delta that we discarded on an
1820 * earlier try, but only when reusing delta data. Note that
1821 * src_entry that is marked as the preferred_base should always
1822 * be considered, as even if we produce a suboptimal delta against
1823 * it, we will still save the transfer cost, as we already know
1824 * the other side has it and we won't send src_entry at all.
1826 if (reuse_delta && trg_entry->in_pack &&
1827 trg_entry->in_pack == src_entry->in_pack &&
1828 !src_entry->preferred_base &&
1829 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1830 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1831 return 0;
1833 /* Let's not bust the allowed depth. */
1834 if (src->depth >= max_depth)
1835 return 0;
1837 /* Now some size filtering heuristics. */
1838 trg_size = trg_entry->size;
1839 if (!trg_entry->delta) {
1840 max_size = trg_size/2 - 20;
1841 ref_depth = 1;
1842 } else {
1843 max_size = trg_entry->delta_size;
1844 ref_depth = trg->depth;
1846 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1847 (max_depth - ref_depth + 1);
1848 if (max_size == 0)
1849 return 0;
1850 src_size = src_entry->size;
1851 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1852 if (sizediff >= max_size)
1853 return 0;
1854 if (trg_size < src_size / 32)
1855 return 0;
1857 /* Load data if not already done */
1858 if (!trg->data) {
1859 read_lock();
1860 trg->data = read_sha1_file(trg_entry->idx.oid.hash, &type,
1861 &sz);
1862 read_unlock();
1863 if (!trg->data)
1864 die("object %s cannot be read",
1865 oid_to_hex(&trg_entry->idx.oid));
1866 if (sz != trg_size)
1867 die("object %s inconsistent object length (%lu vs %lu)",
1868 oid_to_hex(&trg_entry->idx.oid), sz,
1869 trg_size);
1870 *mem_usage += sz;
1872 if (!src->data) {
1873 read_lock();
1874 src->data = read_sha1_file(src_entry->idx.oid.hash, &type,
1875 &sz);
1876 read_unlock();
1877 if (!src->data) {
1878 if (src_entry->preferred_base) {
1879 static int warned = 0;
1880 if (!warned++)
1881 warning("object %s cannot be read",
1882 oid_to_hex(&src_entry->idx.oid));
1884 * Those objects are not included in the
1885 * resulting pack. Be resilient and ignore
1886 * them if they can't be read, in case the
1887 * pack could be created nevertheless.
1889 return 0;
1891 die("object %s cannot be read",
1892 oid_to_hex(&src_entry->idx.oid));
1894 if (sz != src_size)
1895 die("object %s inconsistent object length (%lu vs %lu)",
1896 oid_to_hex(&src_entry->idx.oid), sz,
1897 src_size);
1898 *mem_usage += sz;
1900 if (!src->index) {
1901 src->index = create_delta_index(src->data, src_size);
1902 if (!src->index) {
1903 static int warned = 0;
1904 if (!warned++)
1905 warning("suboptimal pack - out of memory");
1906 return 0;
1908 *mem_usage += sizeof_delta_index(src->index);
1911 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1912 if (!delta_buf)
1913 return 0;
1915 if (trg_entry->delta) {
1916 /* Prefer only shallower same-sized deltas. */
1917 if (delta_size == trg_entry->delta_size &&
1918 src->depth + 1 >= trg->depth) {
1919 free(delta_buf);
1920 return 0;
1925 * Handle memory allocation outside of the cache
1926 * accounting lock. Compiler will optimize the strangeness
1927 * away when NO_PTHREADS is defined.
1929 free(trg_entry->delta_data);
1930 cache_lock();
1931 if (trg_entry->delta_data) {
1932 delta_cache_size -= trg_entry->delta_size;
1933 trg_entry->delta_data = NULL;
1935 if (delta_cacheable(src_size, trg_size, delta_size)) {
1936 delta_cache_size += delta_size;
1937 cache_unlock();
1938 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1939 } else {
1940 cache_unlock();
1941 free(delta_buf);
1944 trg_entry->delta = src_entry;
1945 trg_entry->delta_size = delta_size;
1946 trg->depth = src->depth + 1;
1948 return 1;
1951 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1953 struct object_entry *child = me->delta_child;
1954 unsigned int m = n;
1955 while (child) {
1956 unsigned int c = check_delta_limit(child, n + 1);
1957 if (m < c)
1958 m = c;
1959 child = child->delta_sibling;
1961 return m;
1964 static unsigned long free_unpacked(struct unpacked *n)
1966 unsigned long freed_mem = sizeof_delta_index(n->index);
1967 free_delta_index(n->index);
1968 n->index = NULL;
1969 if (n->data) {
1970 freed_mem += n->entry->size;
1971 FREE_AND_NULL(n->data);
1973 n->entry = NULL;
1974 n->depth = 0;
1975 return freed_mem;
1978 static void find_deltas(struct object_entry **list, unsigned *list_size,
1979 int window, int depth, unsigned *processed)
1981 uint32_t i, idx = 0, count = 0;
1982 struct unpacked *array;
1983 unsigned long mem_usage = 0;
1985 array = xcalloc(window, sizeof(struct unpacked));
1987 for (;;) {
1988 struct object_entry *entry;
1989 struct unpacked *n = array + idx;
1990 int j, max_depth, best_base = -1;
1992 progress_lock();
1993 if (!*list_size) {
1994 progress_unlock();
1995 break;
1997 entry = *list++;
1998 (*list_size)--;
1999 if (!entry->preferred_base) {
2000 (*processed)++;
2001 display_progress(progress_state, *processed);
2003 progress_unlock();
2005 mem_usage -= free_unpacked(n);
2006 n->entry = entry;
2008 while (window_memory_limit &&
2009 mem_usage > window_memory_limit &&
2010 count > 1) {
2011 uint32_t tail = (idx + window - count) % window;
2012 mem_usage -= free_unpacked(array + tail);
2013 count--;
2016 /* We do not compute delta to *create* objects we are not
2017 * going to pack.
2019 if (entry->preferred_base)
2020 goto next;
2023 * If the current object is at pack edge, take the depth the
2024 * objects that depend on the current object into account
2025 * otherwise they would become too deep.
2027 max_depth = depth;
2028 if (entry->delta_child) {
2029 max_depth -= check_delta_limit(entry, 0);
2030 if (max_depth <= 0)
2031 goto next;
2034 j = window;
2035 while (--j > 0) {
2036 int ret;
2037 uint32_t other_idx = idx + j;
2038 struct unpacked *m;
2039 if (other_idx >= window)
2040 other_idx -= window;
2041 m = array + other_idx;
2042 if (!m->entry)
2043 break;
2044 ret = try_delta(n, m, max_depth, &mem_usage);
2045 if (ret < 0)
2046 break;
2047 else if (ret > 0)
2048 best_base = other_idx;
2052 * If we decided to cache the delta data, then it is best
2053 * to compress it right away. First because we have to do
2054 * it anyway, and doing it here while we're threaded will
2055 * save a lot of time in the non threaded write phase,
2056 * as well as allow for caching more deltas within
2057 * the same cache size limit.
2058 * ...
2059 * But only if not writing to stdout, since in that case
2060 * the network is most likely throttling writes anyway,
2061 * and therefore it is best to go to the write phase ASAP
2062 * instead, as we can afford spending more time compressing
2063 * between writes at that moment.
2065 if (entry->delta_data && !pack_to_stdout) {
2066 entry->z_delta_size = do_compress(&entry->delta_data,
2067 entry->delta_size);
2068 cache_lock();
2069 delta_cache_size -= entry->delta_size;
2070 delta_cache_size += entry->z_delta_size;
2071 cache_unlock();
2074 /* if we made n a delta, and if n is already at max
2075 * depth, leaving it in the window is pointless. we
2076 * should evict it first.
2078 if (entry->delta && max_depth <= n->depth)
2079 continue;
2082 * Move the best delta base up in the window, after the
2083 * currently deltified object, to keep it longer. It will
2084 * be the first base object to be attempted next.
2086 if (entry->delta) {
2087 struct unpacked swap = array[best_base];
2088 int dist = (window + idx - best_base) % window;
2089 int dst = best_base;
2090 while (dist--) {
2091 int src = (dst + 1) % window;
2092 array[dst] = array[src];
2093 dst = src;
2095 array[dst] = swap;
2098 next:
2099 idx++;
2100 if (count + 1 < window)
2101 count++;
2102 if (idx >= window)
2103 idx = 0;
2106 for (i = 0; i < window; ++i) {
2107 free_delta_index(array[i].index);
2108 free(array[i].data);
2110 free(array);
2113 #ifndef NO_PTHREADS
2115 static void try_to_free_from_threads(size_t size)
2117 read_lock();
2118 release_pack_memory(size);
2119 read_unlock();
2122 static try_to_free_t old_try_to_free_routine;
2125 * The main thread waits on the condition that (at least) one of the workers
2126 * has stopped working (which is indicated in the .working member of
2127 * struct thread_params).
2128 * When a work thread has completed its work, it sets .working to 0 and
2129 * signals the main thread and waits on the condition that .data_ready
2130 * becomes 1.
2133 struct thread_params {
2134 pthread_t thread;
2135 struct object_entry **list;
2136 unsigned list_size;
2137 unsigned remaining;
2138 int window;
2139 int depth;
2140 int working;
2141 int data_ready;
2142 pthread_mutex_t mutex;
2143 pthread_cond_t cond;
2144 unsigned *processed;
2147 static pthread_cond_t progress_cond;
2150 * Mutex and conditional variable can't be statically-initialized on Windows.
2152 static void init_threaded_search(void)
2154 init_recursive_mutex(&read_mutex);
2155 pthread_mutex_init(&cache_mutex, NULL);
2156 pthread_mutex_init(&progress_mutex, NULL);
2157 pthread_cond_init(&progress_cond, NULL);
2158 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
2161 static void cleanup_threaded_search(void)
2163 set_try_to_free_routine(old_try_to_free_routine);
2164 pthread_cond_destroy(&progress_cond);
2165 pthread_mutex_destroy(&read_mutex);
2166 pthread_mutex_destroy(&cache_mutex);
2167 pthread_mutex_destroy(&progress_mutex);
2170 static void *threaded_find_deltas(void *arg)
2172 struct thread_params *me = arg;
2174 progress_lock();
2175 while (me->remaining) {
2176 progress_unlock();
2178 find_deltas(me->list, &me->remaining,
2179 me->window, me->depth, me->processed);
2181 progress_lock();
2182 me->working = 0;
2183 pthread_cond_signal(&progress_cond);
2184 progress_unlock();
2187 * We must not set ->data_ready before we wait on the
2188 * condition because the main thread may have set it to 1
2189 * before we get here. In order to be sure that new
2190 * work is available if we see 1 in ->data_ready, it
2191 * was initialized to 0 before this thread was spawned
2192 * and we reset it to 0 right away.
2194 pthread_mutex_lock(&me->mutex);
2195 while (!me->data_ready)
2196 pthread_cond_wait(&me->cond, &me->mutex);
2197 me->data_ready = 0;
2198 pthread_mutex_unlock(&me->mutex);
2200 progress_lock();
2202 progress_unlock();
2203 /* leave ->working 1 so that this doesn't get more work assigned */
2204 return NULL;
2207 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2208 int window, int depth, unsigned *processed)
2210 struct thread_params *p;
2211 int i, ret, active_threads = 0;
2213 init_threaded_search();
2215 if (delta_search_threads <= 1) {
2216 find_deltas(list, &list_size, window, depth, processed);
2217 cleanup_threaded_search();
2218 return;
2220 if (progress > pack_to_stdout)
2221 fprintf(stderr, "Delta compression using up to %d threads.\n",
2222 delta_search_threads);
2223 p = xcalloc(delta_search_threads, sizeof(*p));
2225 /* Partition the work amongst work threads. */
2226 for (i = 0; i < delta_search_threads; i++) {
2227 unsigned sub_size = list_size / (delta_search_threads - i);
2229 /* don't use too small segments or no deltas will be found */
2230 if (sub_size < 2*window && i+1 < delta_search_threads)
2231 sub_size = 0;
2233 p[i].window = window;
2234 p[i].depth = depth;
2235 p[i].processed = processed;
2236 p[i].working = 1;
2237 p[i].data_ready = 0;
2239 /* try to split chunks on "path" boundaries */
2240 while (sub_size && sub_size < list_size &&
2241 list[sub_size]->hash &&
2242 list[sub_size]->hash == list[sub_size-1]->hash)
2243 sub_size++;
2245 p[i].list = list;
2246 p[i].list_size = sub_size;
2247 p[i].remaining = sub_size;
2249 list += sub_size;
2250 list_size -= sub_size;
2253 /* Start work threads. */
2254 for (i = 0; i < delta_search_threads; i++) {
2255 if (!p[i].list_size)
2256 continue;
2257 pthread_mutex_init(&p[i].mutex, NULL);
2258 pthread_cond_init(&p[i].cond, NULL);
2259 ret = pthread_create(&p[i].thread, NULL,
2260 threaded_find_deltas, &p[i]);
2261 if (ret)
2262 die("unable to create thread: %s", strerror(ret));
2263 active_threads++;
2267 * Now let's wait for work completion. Each time a thread is done
2268 * with its work, we steal half of the remaining work from the
2269 * thread with the largest number of unprocessed objects and give
2270 * it to that newly idle thread. This ensure good load balancing
2271 * until the remaining object list segments are simply too short
2272 * to be worth splitting anymore.
2274 while (active_threads) {
2275 struct thread_params *target = NULL;
2276 struct thread_params *victim = NULL;
2277 unsigned sub_size = 0;
2279 progress_lock();
2280 for (;;) {
2281 for (i = 0; !target && i < delta_search_threads; i++)
2282 if (!p[i].working)
2283 target = &p[i];
2284 if (target)
2285 break;
2286 pthread_cond_wait(&progress_cond, &progress_mutex);
2289 for (i = 0; i < delta_search_threads; i++)
2290 if (p[i].remaining > 2*window &&
2291 (!victim || victim->remaining < p[i].remaining))
2292 victim = &p[i];
2293 if (victim) {
2294 sub_size = victim->remaining / 2;
2295 list = victim->list + victim->list_size - sub_size;
2296 while (sub_size && list[0]->hash &&
2297 list[0]->hash == list[-1]->hash) {
2298 list++;
2299 sub_size--;
2301 if (!sub_size) {
2303 * It is possible for some "paths" to have
2304 * so many objects that no hash boundary
2305 * might be found. Let's just steal the
2306 * exact half in that case.
2308 sub_size = victim->remaining / 2;
2309 list -= sub_size;
2311 target->list = list;
2312 victim->list_size -= sub_size;
2313 victim->remaining -= sub_size;
2315 target->list_size = sub_size;
2316 target->remaining = sub_size;
2317 target->working = 1;
2318 progress_unlock();
2320 pthread_mutex_lock(&target->mutex);
2321 target->data_ready = 1;
2322 pthread_cond_signal(&target->cond);
2323 pthread_mutex_unlock(&target->mutex);
2325 if (!sub_size) {
2326 pthread_join(target->thread, NULL);
2327 pthread_cond_destroy(&target->cond);
2328 pthread_mutex_destroy(&target->mutex);
2329 active_threads--;
2332 cleanup_threaded_search();
2333 free(p);
2336 #else
2337 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2338 #endif
2340 static void add_tag_chain(const struct object_id *oid)
2342 struct tag *tag;
2345 * We catch duplicates already in add_object_entry(), but we'd
2346 * prefer to do this extra check to avoid having to parse the
2347 * tag at all if we already know that it's being packed (e.g., if
2348 * it was included via bitmaps, we would not have parsed it
2349 * previously).
2351 if (packlist_find(&to_pack, oid->hash, NULL))
2352 return;
2354 tag = lookup_tag(oid);
2355 while (1) {
2356 if (!tag || parse_tag(tag) || !tag->tagged)
2357 die("unable to pack objects reachable from tag %s",
2358 oid_to_hex(oid));
2360 add_object_entry(tag->object.oid.hash, OBJ_TAG, NULL, 0);
2362 if (tag->tagged->type != OBJ_TAG)
2363 return;
2365 tag = (struct tag *)tag->tagged;
2369 static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2371 struct object_id peeled;
2373 if (starts_with(path, "refs/tags/") && /* is a tag? */
2374 !peel_ref(path, peeled.hash) && /* peelable? */
2375 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */
2376 add_tag_chain(oid);
2377 return 0;
2380 static void prepare_pack(int window, int depth)
2382 struct object_entry **delta_list;
2383 uint32_t i, nr_deltas;
2384 unsigned n;
2386 get_object_details();
2389 * If we're locally repacking then we need to be doubly careful
2390 * from now on in order to make sure no stealth corruption gets
2391 * propagated to the new pack. Clients receiving streamed packs
2392 * should validate everything they get anyway so no need to incur
2393 * the additional cost here in that case.
2395 if (!pack_to_stdout)
2396 do_check_packed_object_crc = 1;
2398 if (!to_pack.nr_objects || !window || !depth)
2399 return;
2401 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2402 nr_deltas = n = 0;
2404 for (i = 0; i < to_pack.nr_objects; i++) {
2405 struct object_entry *entry = to_pack.objects + i;
2407 if (entry->delta)
2408 /* This happens if we decided to reuse existing
2409 * delta from a pack. "reuse_delta &&" is implied.
2411 continue;
2413 if (entry->size < 50)
2414 continue;
2416 if (entry->no_try_delta)
2417 continue;
2419 if (!entry->preferred_base) {
2420 nr_deltas++;
2421 if (entry->type < 0)
2422 die("unable to get type of object %s",
2423 oid_to_hex(&entry->idx.oid));
2424 } else {
2425 if (entry->type < 0) {
2427 * This object is not found, but we
2428 * don't have to include it anyway.
2430 continue;
2434 delta_list[n++] = entry;
2437 if (nr_deltas && n > 1) {
2438 unsigned nr_done = 0;
2439 if (progress)
2440 progress_state = start_progress(_("Compressing objects"),
2441 nr_deltas);
2442 QSORT(delta_list, n, type_size_sort);
2443 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2444 stop_progress(&progress_state);
2445 if (nr_done != nr_deltas)
2446 die("inconsistency with delta count");
2448 free(delta_list);
2451 static int git_pack_config(const char *k, const char *v, void *cb)
2453 if (!strcmp(k, "pack.window")) {
2454 window = git_config_int(k, v);
2455 return 0;
2457 if (!strcmp(k, "pack.windowmemory")) {
2458 window_memory_limit = git_config_ulong(k, v);
2459 return 0;
2461 if (!strcmp(k, "pack.depth")) {
2462 depth = git_config_int(k, v);
2463 return 0;
2465 if (!strcmp(k, "pack.deltacachesize")) {
2466 max_delta_cache_size = git_config_int(k, v);
2467 return 0;
2469 if (!strcmp(k, "pack.deltacachelimit")) {
2470 cache_max_small_delta_size = git_config_int(k, v);
2471 return 0;
2473 if (!strcmp(k, "pack.writebitmaphashcache")) {
2474 if (git_config_bool(k, v))
2475 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2476 else
2477 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2479 if (!strcmp(k, "pack.usebitmaps")) {
2480 use_bitmap_index_default = git_config_bool(k, v);
2481 return 0;
2483 if (!strcmp(k, "pack.threads")) {
2484 delta_search_threads = git_config_int(k, v);
2485 if (delta_search_threads < 0)
2486 die("invalid number of threads specified (%d)",
2487 delta_search_threads);
2488 #ifdef NO_PTHREADS
2489 if (delta_search_threads != 1) {
2490 warning("no threads support, ignoring %s", k);
2491 delta_search_threads = 0;
2493 #endif
2494 return 0;
2496 if (!strcmp(k, "pack.indexversion")) {
2497 pack_idx_opts.version = git_config_int(k, v);
2498 if (pack_idx_opts.version > 2)
2499 die("bad pack.indexversion=%"PRIu32,
2500 pack_idx_opts.version);
2501 return 0;
2503 return git_default_config(k, v, cb);
2506 static void read_object_list_from_stdin(void)
2508 char line[40 + 1 + PATH_MAX + 2];
2509 unsigned char sha1[20];
2511 for (;;) {
2512 if (!fgets(line, sizeof(line), stdin)) {
2513 if (feof(stdin))
2514 break;
2515 if (!ferror(stdin))
2516 die("fgets returned NULL, not EOF, not error!");
2517 if (errno != EINTR)
2518 die_errno("fgets");
2519 clearerr(stdin);
2520 continue;
2522 if (line[0] == '-') {
2523 if (get_sha1_hex(line+1, sha1))
2524 die("expected edge sha1, got garbage:\n %s",
2525 line);
2526 add_preferred_base(sha1);
2527 continue;
2529 if (get_sha1_hex(line, sha1))
2530 die("expected sha1, got garbage:\n %s", line);
2532 add_preferred_base_object(line+41);
2533 add_object_entry(sha1, 0, line+41, 0);
2537 #define OBJECT_ADDED (1u<<20)
2539 static void show_commit(struct commit *commit, void *data)
2541 add_object_entry(commit->object.oid.hash, OBJ_COMMIT, NULL, 0);
2542 commit->object.flags |= OBJECT_ADDED;
2544 if (write_bitmap_index)
2545 index_commit_for_bitmap(commit);
2548 static void show_object(struct object *obj, const char *name, void *data)
2550 add_preferred_base_object(name);
2551 add_object_entry(obj->oid.hash, obj->type, name, 0);
2552 obj->flags |= OBJECT_ADDED;
2555 static void show_edge(struct commit *commit)
2557 add_preferred_base(commit->object.oid.hash);
2560 struct in_pack_object {
2561 off_t offset;
2562 struct object *object;
2565 struct in_pack {
2566 unsigned int alloc;
2567 unsigned int nr;
2568 struct in_pack_object *array;
2571 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2573 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2574 in_pack->array[in_pack->nr].object = object;
2575 in_pack->nr++;
2579 * Compare the objects in the offset order, in order to emulate the
2580 * "git rev-list --objects" output that produced the pack originally.
2582 static int ofscmp(const void *a_, const void *b_)
2584 struct in_pack_object *a = (struct in_pack_object *)a_;
2585 struct in_pack_object *b = (struct in_pack_object *)b_;
2587 if (a->offset < b->offset)
2588 return -1;
2589 else if (a->offset > b->offset)
2590 return 1;
2591 else
2592 return oidcmp(&a->object->oid, &b->object->oid);
2595 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2597 struct packed_git *p;
2598 struct in_pack in_pack;
2599 uint32_t i;
2601 memset(&in_pack, 0, sizeof(in_pack));
2603 for (p = packed_git; p; p = p->next) {
2604 const unsigned char *sha1;
2605 struct object *o;
2607 if (!p->pack_local || p->pack_keep)
2608 continue;
2609 if (open_pack_index(p))
2610 die("cannot open pack index");
2612 ALLOC_GROW(in_pack.array,
2613 in_pack.nr + p->num_objects,
2614 in_pack.alloc);
2616 for (i = 0; i < p->num_objects; i++) {
2617 sha1 = nth_packed_object_sha1(p, i);
2618 o = lookup_unknown_object(sha1);
2619 if (!(o->flags & OBJECT_ADDED))
2620 mark_in_pack_object(o, p, &in_pack);
2621 o->flags |= OBJECT_ADDED;
2625 if (in_pack.nr) {
2626 QSORT(in_pack.array, in_pack.nr, ofscmp);
2627 for (i = 0; i < in_pack.nr; i++) {
2628 struct object *o = in_pack.array[i].object;
2629 add_object_entry(o->oid.hash, o->type, "", 0);
2632 free(in_pack.array);
2635 static int add_loose_object(const struct object_id *oid, const char *path,
2636 void *data)
2638 enum object_type type = sha1_object_info(oid->hash, NULL);
2640 if (type < 0) {
2641 warning("loose object at %s could not be examined", path);
2642 return 0;
2645 add_object_entry(oid->hash, type, "", 0);
2646 return 0;
2650 * We actually don't even have to worry about reachability here.
2651 * add_object_entry will weed out duplicates, so we just add every
2652 * loose object we find.
2654 static void add_unreachable_loose_objects(void)
2656 for_each_loose_file_in_objdir(get_object_directory(),
2657 add_loose_object,
2658 NULL, NULL, NULL);
2661 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2663 static struct packed_git *last_found = (void *)1;
2664 struct packed_git *p;
2666 p = (last_found != (void *)1) ? last_found : packed_git;
2668 while (p) {
2669 if ((!p->pack_local || p->pack_keep) &&
2670 find_pack_entry_one(sha1, p)) {
2671 last_found = p;
2672 return 1;
2674 if (p == last_found)
2675 p = packed_git;
2676 else
2677 p = p->next;
2678 if (p == last_found)
2679 p = p->next;
2681 return 0;
2685 * Store a list of sha1s that are should not be discarded
2686 * because they are either written too recently, or are
2687 * reachable from another object that was.
2689 * This is filled by get_object_list.
2691 static struct oid_array recent_objects;
2693 static int loosened_object_can_be_discarded(const struct object_id *oid,
2694 timestamp_t mtime)
2696 if (!unpack_unreachable_expiration)
2697 return 0;
2698 if (mtime > unpack_unreachable_expiration)
2699 return 0;
2700 if (oid_array_lookup(&recent_objects, oid) >= 0)
2701 return 0;
2702 return 1;
2705 static void loosen_unused_packed_objects(struct rev_info *revs)
2707 struct packed_git *p;
2708 uint32_t i;
2709 struct object_id oid;
2711 for (p = packed_git; p; p = p->next) {
2712 if (!p->pack_local || p->pack_keep)
2713 continue;
2715 if (open_pack_index(p))
2716 die("cannot open pack index");
2718 for (i = 0; i < p->num_objects; i++) {
2719 nth_packed_object_oid(&oid, p, i);
2720 if (!packlist_find(&to_pack, oid.hash, NULL) &&
2721 !has_sha1_pack_kept_or_nonlocal(oid.hash) &&
2722 !loosened_object_can_be_discarded(&oid, p->mtime))
2723 if (force_object_loose(oid.hash, p->mtime))
2724 die("unable to force loose object");
2730 * This tracks any options which pack-reuse code expects to be on, or which a
2731 * reader of the pack might not understand, and which would therefore prevent
2732 * blind reuse of what we have on disk.
2734 static int pack_options_allow_reuse(void)
2736 return pack_to_stdout &&
2737 allow_ofs_delta &&
2738 !ignore_packed_keep &&
2739 (!local || !have_non_local_packs) &&
2740 !incremental;
2743 static int get_object_list_from_bitmap(struct rev_info *revs)
2745 if (prepare_bitmap_walk(revs) < 0)
2746 return -1;
2748 if (pack_options_allow_reuse() &&
2749 !reuse_partial_packfile_from_bitmap(
2750 &reuse_packfile,
2751 &reuse_packfile_objects,
2752 &reuse_packfile_offset)) {
2753 assert(reuse_packfile_objects);
2754 nr_result += reuse_packfile_objects;
2755 display_progress(progress_state, nr_result);
2758 traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2759 return 0;
2762 static void record_recent_object(struct object *obj,
2763 const char *name,
2764 void *data)
2766 oid_array_append(&recent_objects, &obj->oid);
2769 static void record_recent_commit(struct commit *commit, void *data)
2771 oid_array_append(&recent_objects, &commit->object.oid);
2774 static void get_object_list(int ac, const char **av)
2776 struct rev_info revs;
2777 char line[1000];
2778 int flags = 0;
2780 init_revisions(&revs, NULL);
2781 save_commit_buffer = 0;
2782 setup_revisions(ac, av, &revs, NULL);
2784 /* make sure shallows are read */
2785 is_repository_shallow();
2787 while (fgets(line, sizeof(line), stdin) != NULL) {
2788 int len = strlen(line);
2789 if (len && line[len - 1] == '\n')
2790 line[--len] = 0;
2791 if (!len)
2792 break;
2793 if (*line == '-') {
2794 if (!strcmp(line, "--not")) {
2795 flags ^= UNINTERESTING;
2796 write_bitmap_index = 0;
2797 continue;
2799 if (starts_with(line, "--shallow ")) {
2800 struct object_id oid;
2801 if (get_oid_hex(line + 10, &oid))
2802 die("not an SHA-1 '%s'", line + 10);
2803 register_shallow(&oid);
2804 use_bitmap_index = 0;
2805 continue;
2807 die("not a rev '%s'", line);
2809 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2810 die("bad revision '%s'", line);
2813 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2814 return;
2816 if (prepare_revision_walk(&revs))
2817 die("revision walk setup failed");
2818 mark_edges_uninteresting(&revs, show_edge);
2819 traverse_commit_list(&revs, show_commit, show_object, NULL);
2821 if (unpack_unreachable_expiration) {
2822 revs.ignore_missing_links = 1;
2823 if (add_unseen_recent_objects_to_traversal(&revs,
2824 unpack_unreachable_expiration))
2825 die("unable to add recent objects");
2826 if (prepare_revision_walk(&revs))
2827 die("revision walk setup failed");
2828 traverse_commit_list(&revs, record_recent_commit,
2829 record_recent_object, NULL);
2832 if (keep_unreachable)
2833 add_objects_in_unpacked_packs(&revs);
2834 if (pack_loose_unreachable)
2835 add_unreachable_loose_objects();
2836 if (unpack_unreachable)
2837 loosen_unused_packed_objects(&revs);
2839 oid_array_clear(&recent_objects);
2842 static int option_parse_index_version(const struct option *opt,
2843 const char *arg, int unset)
2845 char *c;
2846 const char *val = arg;
2847 pack_idx_opts.version = strtoul(val, &c, 10);
2848 if (pack_idx_opts.version > 2)
2849 die(_("unsupported index version %s"), val);
2850 if (*c == ',' && c[1])
2851 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2852 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2853 die(_("bad index version '%s'"), val);
2854 return 0;
2857 static int option_parse_unpack_unreachable(const struct option *opt,
2858 const char *arg, int unset)
2860 if (unset) {
2861 unpack_unreachable = 0;
2862 unpack_unreachable_expiration = 0;
2864 else {
2865 unpack_unreachable = 1;
2866 if (arg)
2867 unpack_unreachable_expiration = approxidate(arg);
2869 return 0;
2872 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2874 int use_internal_rev_list = 0;
2875 int thin = 0;
2876 int shallow = 0;
2877 int all_progress_implied = 0;
2878 struct argv_array rp = ARGV_ARRAY_INIT;
2879 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2880 int rev_list_index = 0;
2881 struct option pack_objects_options[] = {
2882 OPT_SET_INT('q', "quiet", &progress,
2883 N_("do not show progress meter"), 0),
2884 OPT_SET_INT(0, "progress", &progress,
2885 N_("show progress meter"), 1),
2886 OPT_SET_INT(0, "all-progress", &progress,
2887 N_("show progress meter during object writing phase"), 2),
2888 OPT_BOOL(0, "all-progress-implied",
2889 &all_progress_implied,
2890 N_("similar to --all-progress when progress meter is shown")),
2891 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2892 N_("write the pack index file in the specified idx format version"),
2893 0, option_parse_index_version },
2894 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
2895 N_("maximum size of each output pack file")),
2896 OPT_BOOL(0, "local", &local,
2897 N_("ignore borrowed objects from alternate object store")),
2898 OPT_BOOL(0, "incremental", &incremental,
2899 N_("ignore packed objects")),
2900 OPT_INTEGER(0, "window", &window,
2901 N_("limit pack window by objects")),
2902 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
2903 N_("limit pack window by memory in addition to object limit")),
2904 OPT_INTEGER(0, "depth", &depth,
2905 N_("maximum length of delta chain allowed in the resulting pack")),
2906 OPT_BOOL(0, "reuse-delta", &reuse_delta,
2907 N_("reuse existing deltas")),
2908 OPT_BOOL(0, "reuse-object", &reuse_object,
2909 N_("reuse existing objects")),
2910 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
2911 N_("use OFS_DELTA objects")),
2912 OPT_INTEGER(0, "threads", &delta_search_threads,
2913 N_("use threads when searching for best delta matches")),
2914 OPT_BOOL(0, "non-empty", &non_empty,
2915 N_("do not create an empty pack output")),
2916 OPT_BOOL(0, "revs", &use_internal_rev_list,
2917 N_("read revision arguments from standard input")),
2918 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
2919 N_("limit the objects to those that are not yet packed"),
2920 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2921 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
2922 N_("include objects reachable from any reference"),
2923 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2924 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
2925 N_("include objects referred by reflog entries"),
2926 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2927 { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL,
2928 N_("include objects referred to by the index"),
2929 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2930 OPT_BOOL(0, "stdout", &pack_to_stdout,
2931 N_("output pack to stdout")),
2932 OPT_BOOL(0, "include-tag", &include_tag,
2933 N_("include tag objects that refer to objects to be packed")),
2934 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
2935 N_("keep unreachable objects")),
2936 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
2937 N_("pack loose unreachable objects")),
2938 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
2939 N_("unpack unreachable objects newer than <time>"),
2940 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
2941 OPT_BOOL(0, "thin", &thin,
2942 N_("create thin packs")),
2943 OPT_BOOL(0, "shallow", &shallow,
2944 N_("create packs suitable for shallow fetches")),
2945 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
2946 N_("ignore packs that have companion .keep file")),
2947 OPT_INTEGER(0, "compression", &pack_compression_level,
2948 N_("pack compression level")),
2949 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
2950 N_("do not hide commits by grafts"), 0),
2951 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
2952 N_("use a bitmap index if available to speed up counting objects")),
2953 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
2954 N_("write a bitmap index together with the pack index")),
2955 OPT_END(),
2958 check_replace_refs = 0;
2960 reset_pack_idx_option(&pack_idx_opts);
2961 git_config(git_pack_config, NULL);
2963 progress = isatty(2);
2964 argc = parse_options(argc, argv, prefix, pack_objects_options,
2965 pack_usage, 0);
2967 if (argc) {
2968 base_name = argv[0];
2969 argc--;
2971 if (pack_to_stdout != !base_name || argc)
2972 usage_with_options(pack_usage, pack_objects_options);
2974 argv_array_push(&rp, "pack-objects");
2975 if (thin) {
2976 use_internal_rev_list = 1;
2977 argv_array_push(&rp, shallow
2978 ? "--objects-edge-aggressive"
2979 : "--objects-edge");
2980 } else
2981 argv_array_push(&rp, "--objects");
2983 if (rev_list_all) {
2984 use_internal_rev_list = 1;
2985 argv_array_push(&rp, "--all");
2987 if (rev_list_reflog) {
2988 use_internal_rev_list = 1;
2989 argv_array_push(&rp, "--reflog");
2991 if (rev_list_index) {
2992 use_internal_rev_list = 1;
2993 argv_array_push(&rp, "--indexed-objects");
2995 if (rev_list_unpacked) {
2996 use_internal_rev_list = 1;
2997 argv_array_push(&rp, "--unpacked");
3000 if (!reuse_object)
3001 reuse_delta = 0;
3002 if (pack_compression_level == -1)
3003 pack_compression_level = Z_DEFAULT_COMPRESSION;
3004 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
3005 die("bad pack compression level %d", pack_compression_level);
3007 if (!delta_search_threads) /* --threads=0 means autodetect */
3008 delta_search_threads = online_cpus();
3010 #ifdef NO_PTHREADS
3011 if (delta_search_threads != 1)
3012 warning("no threads support, ignoring --threads");
3013 #endif
3014 if (!pack_to_stdout && !pack_size_limit)
3015 pack_size_limit = pack_size_limit_cfg;
3016 if (pack_to_stdout && pack_size_limit)
3017 die("--max-pack-size cannot be used to build a pack for transfer.");
3018 if (pack_size_limit && pack_size_limit < 1024*1024) {
3019 warning("minimum pack size limit is 1 MiB");
3020 pack_size_limit = 1024*1024;
3023 if (!pack_to_stdout && thin)
3024 die("--thin cannot be used to build an indexable pack.");
3026 if (keep_unreachable && unpack_unreachable)
3027 die("--keep-unreachable and --unpack-unreachable are incompatible.");
3028 if (!rev_list_all || !rev_list_reflog || !rev_list_index)
3029 unpack_unreachable_expiration = 0;
3032 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3034 * - to produce good pack (with bitmap index not-yet-packed objects are
3035 * packed in suboptimal order).
3037 * - to use more robust pack-generation codepath (avoiding possible
3038 * bugs in bitmap code and possible bitmap index corruption).
3040 if (!pack_to_stdout)
3041 use_bitmap_index_default = 0;
3043 if (use_bitmap_index < 0)
3044 use_bitmap_index = use_bitmap_index_default;
3046 /* "hard" reasons not to use bitmaps; these just won't work at all */
3047 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow())
3048 use_bitmap_index = 0;
3050 if (pack_to_stdout || !rev_list_all)
3051 write_bitmap_index = 0;
3053 if (progress && all_progress_implied)
3054 progress = 2;
3056 prepare_packed_git();
3057 if (ignore_packed_keep) {
3058 struct packed_git *p;
3059 for (p = packed_git; p; p = p->next)
3060 if (p->pack_local && p->pack_keep)
3061 break;
3062 if (!p) /* no keep-able packs found */
3063 ignore_packed_keep = 0;
3065 if (local) {
3067 * unlike ignore_packed_keep above, we do not want to
3068 * unset "local" based on looking at packs, as it
3069 * also covers non-local objects
3071 struct packed_git *p;
3072 for (p = packed_git; p; p = p->next) {
3073 if (!p->pack_local) {
3074 have_non_local_packs = 1;
3075 break;
3080 if (progress)
3081 progress_state = start_progress(_("Counting objects"), 0);
3082 if (!use_internal_rev_list)
3083 read_object_list_from_stdin();
3084 else {
3085 get_object_list(rp.argc, rp.argv);
3086 argv_array_clear(&rp);
3088 cleanup_preferred_base();
3089 if (include_tag && nr_result)
3090 for_each_ref(add_ref_tag, NULL);
3091 stop_progress(&progress_state);
3093 if (non_empty && !nr_result)
3094 return 0;
3095 if (nr_result)
3096 prepare_pack(window, depth);
3097 write_pack_file();
3098 if (progress)
3099 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
3100 " reused %"PRIu32" (delta %"PRIu32")\n",
3101 written, written_delta, reused, reused_delta);
3102 return 0;