The sixth batch
[git.git] / builtin / pack-objects.c
blobde00adbb9e0d517bf44d6318ec8df26b9f82c503
1 #include "builtin.h"
2 #include "cache.h"
3 #include "repository.h"
4 #include "config.h"
5 #include "attr.h"
6 #include "object.h"
7 #include "blob.h"
8 #include "commit.h"
9 #include "tag.h"
10 #include "tree.h"
11 #include "delta.h"
12 #include "pack.h"
13 #include "pack-revindex.h"
14 #include "csum-file.h"
15 #include "tree-walk.h"
16 #include "diff.h"
17 #include "revision.h"
18 #include "list-objects.h"
19 #include "list-objects-filter.h"
20 #include "list-objects-filter-options.h"
21 #include "pack-objects.h"
22 #include "progress.h"
23 #include "refs.h"
24 #include "streaming.h"
25 #include "thread-utils.h"
26 #include "pack-bitmap.h"
27 #include "delta-islands.h"
28 #include "reachable.h"
29 #include "oid-array.h"
30 #include "strvec.h"
31 #include "list.h"
32 #include "packfile.h"
33 #include "object-store.h"
34 #include "dir.h"
35 #include "midx.h"
36 #include "trace2.h"
37 #include "shallow.h"
38 #include "promisor-remote.h"
41 * Objects we are going to pack are collected in the `to_pack` structure.
42 * It contains an array (dynamically expanded) of the object data, and a map
43 * that can resolve SHA1s to their position in the array.
45 static struct packing_data to_pack;
47 static inline struct object_entry *oe_delta(
48 const struct packing_data *pack,
49 const struct object_entry *e)
51 if (!e->delta_idx)
52 return NULL;
53 if (e->ext_base)
54 return &pack->ext_bases[e->delta_idx - 1];
55 else
56 return &pack->objects[e->delta_idx - 1];
59 static inline unsigned long oe_delta_size(struct packing_data *pack,
60 const struct object_entry *e)
62 if (e->delta_size_valid)
63 return e->delta_size_;
66 * pack->delta_size[] can't be NULL because oe_set_delta_size()
67 * must have been called when a new delta is saved with
68 * oe_set_delta().
69 * If oe_delta() returns NULL (i.e. default state, which means
70 * delta_size_valid is also false), then the caller must never
71 * call oe_delta_size().
73 return pack->delta_size[e - pack->objects];
76 unsigned long oe_get_size_slow(struct packing_data *pack,
77 const struct object_entry *e);
79 static inline unsigned long oe_size(struct packing_data *pack,
80 const struct object_entry *e)
82 if (e->size_valid)
83 return e->size_;
85 return oe_get_size_slow(pack, e);
88 static inline void oe_set_delta(struct packing_data *pack,
89 struct object_entry *e,
90 struct object_entry *delta)
92 if (delta)
93 e->delta_idx = (delta - pack->objects) + 1;
94 else
95 e->delta_idx = 0;
98 static inline struct object_entry *oe_delta_sibling(
99 const struct packing_data *pack,
100 const struct object_entry *e)
102 if (e->delta_sibling_idx)
103 return &pack->objects[e->delta_sibling_idx - 1];
104 return NULL;
107 static inline struct object_entry *oe_delta_child(
108 const struct packing_data *pack,
109 const struct object_entry *e)
111 if (e->delta_child_idx)
112 return &pack->objects[e->delta_child_idx - 1];
113 return NULL;
116 static inline void oe_set_delta_child(struct packing_data *pack,
117 struct object_entry *e,
118 struct object_entry *delta)
120 if (delta)
121 e->delta_child_idx = (delta - pack->objects) + 1;
122 else
123 e->delta_child_idx = 0;
126 static inline void oe_set_delta_sibling(struct packing_data *pack,
127 struct object_entry *e,
128 struct object_entry *delta)
130 if (delta)
131 e->delta_sibling_idx = (delta - pack->objects) + 1;
132 else
133 e->delta_sibling_idx = 0;
136 static inline void oe_set_size(struct packing_data *pack,
137 struct object_entry *e,
138 unsigned long size)
140 if (size < pack->oe_size_limit) {
141 e->size_ = size;
142 e->size_valid = 1;
143 } else {
144 e->size_valid = 0;
145 if (oe_get_size_slow(pack, e) != size)
146 BUG("'size' is supposed to be the object size!");
150 static inline void oe_set_delta_size(struct packing_data *pack,
151 struct object_entry *e,
152 unsigned long size)
154 if (size < pack->oe_delta_size_limit) {
155 e->delta_size_ = size;
156 e->delta_size_valid = 1;
157 } else {
158 packing_data_lock(pack);
159 if (!pack->delta_size)
160 ALLOC_ARRAY(pack->delta_size, pack->nr_alloc);
161 packing_data_unlock(pack);
163 pack->delta_size[e - pack->objects] = size;
164 e->delta_size_valid = 0;
168 #define IN_PACK(obj) oe_in_pack(&to_pack, obj)
169 #define SIZE(obj) oe_size(&to_pack, obj)
170 #define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size)
171 #define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj)
172 #define DELTA(obj) oe_delta(&to_pack, obj)
173 #define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj)
174 #define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj)
175 #define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val)
176 #define SET_DELTA_EXT(obj, oid) oe_set_delta_ext(&to_pack, obj, oid)
177 #define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val)
178 #define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val)
179 #define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val)
181 static const char *pack_usage[] = {
182 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
183 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
184 NULL
187 static struct pack_idx_entry **written_list;
188 static uint32_t nr_result, nr_written, nr_seen;
189 static struct bitmap_index *bitmap_git;
190 static uint32_t write_layer;
192 static int non_empty;
193 static int reuse_delta = 1, reuse_object = 1;
194 static int keep_unreachable, unpack_unreachable, include_tag;
195 static timestamp_t unpack_unreachable_expiration;
196 static int pack_loose_unreachable;
197 static int local;
198 static int have_non_local_packs;
199 static int incremental;
200 static int ignore_packed_keep_on_disk;
201 static int ignore_packed_keep_in_core;
202 static int allow_ofs_delta;
203 static struct pack_idx_option pack_idx_opts;
204 static const char *base_name;
205 static int progress = 1;
206 static int window = 10;
207 static unsigned long pack_size_limit;
208 static int depth = 50;
209 static int delta_search_threads;
210 static int pack_to_stdout;
211 static int sparse;
212 static int thin;
213 static int num_preferred_base;
214 static struct progress *progress_state;
216 static struct packed_git *reuse_packfile;
217 static uint32_t reuse_packfile_objects;
218 static struct bitmap *reuse_packfile_bitmap;
220 static int use_bitmap_index_default = 1;
221 static int use_bitmap_index = -1;
222 static int allow_pack_reuse = 1;
223 static enum {
224 WRITE_BITMAP_FALSE = 0,
225 WRITE_BITMAP_QUIET,
226 WRITE_BITMAP_TRUE,
227 } write_bitmap_index;
228 static uint16_t write_bitmap_options = BITMAP_OPT_HASH_CACHE;
230 static int exclude_promisor_objects;
232 static int use_delta_islands;
234 static unsigned long delta_cache_size = 0;
235 static unsigned long max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE;
236 static unsigned long cache_max_small_delta_size = 1000;
238 static unsigned long window_memory_limit = 0;
240 static struct list_objects_filter_options filter_options;
242 static struct string_list uri_protocols = STRING_LIST_INIT_NODUP;
244 enum missing_action {
245 MA_ERROR = 0, /* fail if any missing objects are encountered */
246 MA_ALLOW_ANY, /* silently allow ALL missing objects */
247 MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */
249 static enum missing_action arg_missing_action;
250 static show_object_fn fn_show_object;
252 struct configured_exclusion {
253 struct oidmap_entry e;
254 char *pack_hash_hex;
255 char *uri;
257 static struct oidmap configured_exclusions;
259 static struct oidset excluded_by_config;
262 * stats
264 static uint32_t written, written_delta;
265 static uint32_t reused, reused_delta;
268 * Indexed commits
270 static struct commit **indexed_commits;
271 static unsigned int indexed_commits_nr;
272 static unsigned int indexed_commits_alloc;
274 static void index_commit_for_bitmap(struct commit *commit)
276 if (indexed_commits_nr >= indexed_commits_alloc) {
277 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
278 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
281 indexed_commits[indexed_commits_nr++] = commit;
284 static void *get_delta(struct object_entry *entry)
286 unsigned long size, base_size, delta_size;
287 void *buf, *base_buf, *delta_buf;
288 enum object_type type;
290 buf = read_object_file(&entry->idx.oid, &type, &size);
291 if (!buf)
292 die(_("unable to read %s"), oid_to_hex(&entry->idx.oid));
293 base_buf = read_object_file(&DELTA(entry)->idx.oid, &type,
294 &base_size);
295 if (!base_buf)
296 die("unable to read %s",
297 oid_to_hex(&DELTA(entry)->idx.oid));
298 delta_buf = diff_delta(base_buf, base_size,
299 buf, size, &delta_size, 0);
301 * We successfully computed this delta once but dropped it for
302 * memory reasons. Something is very wrong if this time we
303 * recompute and create a different delta.
305 if (!delta_buf || delta_size != DELTA_SIZE(entry))
306 BUG("delta size changed");
307 free(buf);
308 free(base_buf);
309 return delta_buf;
312 static unsigned long do_compress(void **pptr, unsigned long size)
314 git_zstream stream;
315 void *in, *out;
316 unsigned long maxsize;
318 git_deflate_init(&stream, pack_compression_level);
319 maxsize = git_deflate_bound(&stream, size);
321 in = *pptr;
322 out = xmalloc(maxsize);
323 *pptr = out;
325 stream.next_in = in;
326 stream.avail_in = size;
327 stream.next_out = out;
328 stream.avail_out = maxsize;
329 while (git_deflate(&stream, Z_FINISH) == Z_OK)
330 ; /* nothing */
331 git_deflate_end(&stream);
333 free(in);
334 return stream.total_out;
337 static unsigned long write_large_blob_data(struct git_istream *st, struct hashfile *f,
338 const struct object_id *oid)
340 git_zstream stream;
341 unsigned char ibuf[1024 * 16];
342 unsigned char obuf[1024 * 16];
343 unsigned long olen = 0;
345 git_deflate_init(&stream, pack_compression_level);
347 for (;;) {
348 ssize_t readlen;
349 int zret = Z_OK;
350 readlen = read_istream(st, ibuf, sizeof(ibuf));
351 if (readlen == -1)
352 die(_("unable to read %s"), oid_to_hex(oid));
354 stream.next_in = ibuf;
355 stream.avail_in = readlen;
356 while ((stream.avail_in || readlen == 0) &&
357 (zret == Z_OK || zret == Z_BUF_ERROR)) {
358 stream.next_out = obuf;
359 stream.avail_out = sizeof(obuf);
360 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
361 hashwrite(f, obuf, stream.next_out - obuf);
362 olen += stream.next_out - obuf;
364 if (stream.avail_in)
365 die(_("deflate error (%d)"), zret);
366 if (readlen == 0) {
367 if (zret != Z_STREAM_END)
368 die(_("deflate error (%d)"), zret);
369 break;
372 git_deflate_end(&stream);
373 return olen;
377 * we are going to reuse the existing object data as is. make
378 * sure it is not corrupt.
380 static int check_pack_inflate(struct packed_git *p,
381 struct pack_window **w_curs,
382 off_t offset,
383 off_t len,
384 unsigned long expect)
386 git_zstream stream;
387 unsigned char fakebuf[4096], *in;
388 int st;
390 memset(&stream, 0, sizeof(stream));
391 git_inflate_init(&stream);
392 do {
393 in = use_pack(p, w_curs, offset, &stream.avail_in);
394 stream.next_in = in;
395 stream.next_out = fakebuf;
396 stream.avail_out = sizeof(fakebuf);
397 st = git_inflate(&stream, Z_FINISH);
398 offset += stream.next_in - in;
399 } while (st == Z_OK || st == Z_BUF_ERROR);
400 git_inflate_end(&stream);
401 return (st == Z_STREAM_END &&
402 stream.total_out == expect &&
403 stream.total_in == len) ? 0 : -1;
406 static void copy_pack_data(struct hashfile *f,
407 struct packed_git *p,
408 struct pack_window **w_curs,
409 off_t offset,
410 off_t len)
412 unsigned char *in;
413 unsigned long avail;
415 while (len) {
416 in = use_pack(p, w_curs, offset, &avail);
417 if (avail > len)
418 avail = (unsigned long)len;
419 hashwrite(f, in, avail);
420 offset += avail;
421 len -= avail;
425 static inline int oe_size_greater_than(struct packing_data *pack,
426 const struct object_entry *lhs,
427 unsigned long rhs)
429 if (lhs->size_valid)
430 return lhs->size_ > rhs;
431 if (rhs < pack->oe_size_limit) /* rhs < 2^x <= lhs ? */
432 return 1;
433 return oe_get_size_slow(pack, lhs) > rhs;
436 /* Return 0 if we will bust the pack-size limit */
437 static unsigned long write_no_reuse_object(struct hashfile *f, struct object_entry *entry,
438 unsigned long limit, int usable_delta)
440 unsigned long size, datalen;
441 unsigned char header[MAX_PACK_OBJECT_HEADER],
442 dheader[MAX_PACK_OBJECT_HEADER];
443 unsigned hdrlen;
444 enum object_type type;
445 void *buf;
446 struct git_istream *st = NULL;
447 const unsigned hashsz = the_hash_algo->rawsz;
449 if (!usable_delta) {
450 if (oe_type(entry) == OBJ_BLOB &&
451 oe_size_greater_than(&to_pack, entry, big_file_threshold) &&
452 (st = open_istream(the_repository, &entry->idx.oid, &type,
453 &size, NULL)) != NULL)
454 buf = NULL;
455 else {
456 buf = read_object_file(&entry->idx.oid, &type, &size);
457 if (!buf)
458 die(_("unable to read %s"),
459 oid_to_hex(&entry->idx.oid));
462 * make sure no cached delta data remains from a
463 * previous attempt before a pack split occurred.
465 FREE_AND_NULL(entry->delta_data);
466 entry->z_delta_size = 0;
467 } else if (entry->delta_data) {
468 size = DELTA_SIZE(entry);
469 buf = entry->delta_data;
470 entry->delta_data = NULL;
471 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
472 OBJ_OFS_DELTA : OBJ_REF_DELTA;
473 } else {
474 buf = get_delta(entry);
475 size = DELTA_SIZE(entry);
476 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
477 OBJ_OFS_DELTA : OBJ_REF_DELTA;
480 if (st) /* large blob case, just assume we don't compress well */
481 datalen = size;
482 else if (entry->z_delta_size)
483 datalen = entry->z_delta_size;
484 else
485 datalen = do_compress(&buf, size);
488 * The object header is a byte of 'type' followed by zero or
489 * more bytes of length.
491 hdrlen = encode_in_pack_object_header(header, sizeof(header),
492 type, size);
494 if (type == OBJ_OFS_DELTA) {
496 * Deltas with relative base contain an additional
497 * encoding of the relative offset for the delta
498 * base from this object's position in the pack.
500 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
501 unsigned pos = sizeof(dheader) - 1;
502 dheader[pos] = ofs & 127;
503 while (ofs >>= 7)
504 dheader[--pos] = 128 | (--ofs & 127);
505 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) {
506 if (st)
507 close_istream(st);
508 free(buf);
509 return 0;
511 hashwrite(f, header, hdrlen);
512 hashwrite(f, dheader + pos, sizeof(dheader) - pos);
513 hdrlen += sizeof(dheader) - pos;
514 } else if (type == OBJ_REF_DELTA) {
516 * Deltas with a base reference contain
517 * additional bytes for the base object ID.
519 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) {
520 if (st)
521 close_istream(st);
522 free(buf);
523 return 0;
525 hashwrite(f, header, hdrlen);
526 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz);
527 hdrlen += hashsz;
528 } else {
529 if (limit && hdrlen + datalen + hashsz >= limit) {
530 if (st)
531 close_istream(st);
532 free(buf);
533 return 0;
535 hashwrite(f, header, hdrlen);
537 if (st) {
538 datalen = write_large_blob_data(st, f, &entry->idx.oid);
539 close_istream(st);
540 } else {
541 hashwrite(f, buf, datalen);
542 free(buf);
545 return hdrlen + datalen;
548 /* Return 0 if we will bust the pack-size limit */
549 static off_t write_reuse_object(struct hashfile *f, struct object_entry *entry,
550 unsigned long limit, int usable_delta)
552 struct packed_git *p = IN_PACK(entry);
553 struct pack_window *w_curs = NULL;
554 uint32_t pos;
555 off_t offset;
556 enum object_type type = oe_type(entry);
557 off_t datalen;
558 unsigned char header[MAX_PACK_OBJECT_HEADER],
559 dheader[MAX_PACK_OBJECT_HEADER];
560 unsigned hdrlen;
561 const unsigned hashsz = the_hash_algo->rawsz;
562 unsigned long entry_size = SIZE(entry);
564 if (DELTA(entry))
565 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
566 OBJ_OFS_DELTA : OBJ_REF_DELTA;
567 hdrlen = encode_in_pack_object_header(header, sizeof(header),
568 type, entry_size);
570 offset = entry->in_pack_offset;
571 if (offset_to_pack_pos(p, offset, &pos) < 0)
572 die(_("write_reuse_object: could not locate %s, expected at "
573 "offset %"PRIuMAX" in pack %s"),
574 oid_to_hex(&entry->idx.oid), (uintmax_t)offset,
575 p->pack_name);
576 datalen = pack_pos_to_offset(p, pos + 1) - offset;
577 if (!pack_to_stdout && p->index_version > 1 &&
578 check_pack_crc(p, &w_curs, offset, datalen,
579 pack_pos_to_index(p, pos))) {
580 error(_("bad packed object CRC for %s"),
581 oid_to_hex(&entry->idx.oid));
582 unuse_pack(&w_curs);
583 return write_no_reuse_object(f, entry, limit, usable_delta);
586 offset += entry->in_pack_header_size;
587 datalen -= entry->in_pack_header_size;
589 if (!pack_to_stdout && p->index_version == 1 &&
590 check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) {
591 error(_("corrupt packed object for %s"),
592 oid_to_hex(&entry->idx.oid));
593 unuse_pack(&w_curs);
594 return write_no_reuse_object(f, entry, limit, usable_delta);
597 if (type == OBJ_OFS_DELTA) {
598 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
599 unsigned pos = sizeof(dheader) - 1;
600 dheader[pos] = ofs & 127;
601 while (ofs >>= 7)
602 dheader[--pos] = 128 | (--ofs & 127);
603 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) {
604 unuse_pack(&w_curs);
605 return 0;
607 hashwrite(f, header, hdrlen);
608 hashwrite(f, dheader + pos, sizeof(dheader) - pos);
609 hdrlen += sizeof(dheader) - pos;
610 reused_delta++;
611 } else if (type == OBJ_REF_DELTA) {
612 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) {
613 unuse_pack(&w_curs);
614 return 0;
616 hashwrite(f, header, hdrlen);
617 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz);
618 hdrlen += hashsz;
619 reused_delta++;
620 } else {
621 if (limit && hdrlen + datalen + hashsz >= limit) {
622 unuse_pack(&w_curs);
623 return 0;
625 hashwrite(f, header, hdrlen);
627 copy_pack_data(f, p, &w_curs, offset, datalen);
628 unuse_pack(&w_curs);
629 reused++;
630 return hdrlen + datalen;
633 /* Return 0 if we will bust the pack-size limit */
634 static off_t write_object(struct hashfile *f,
635 struct object_entry *entry,
636 off_t write_offset)
638 unsigned long limit;
639 off_t len;
640 int usable_delta, to_reuse;
642 if (!pack_to_stdout)
643 crc32_begin(f);
645 /* apply size limit if limited packsize and not first object */
646 if (!pack_size_limit || !nr_written)
647 limit = 0;
648 else if (pack_size_limit <= write_offset)
650 * the earlier object did not fit the limit; avoid
651 * mistaking this with unlimited (i.e. limit = 0).
653 limit = 1;
654 else
655 limit = pack_size_limit - write_offset;
657 if (!DELTA(entry))
658 usable_delta = 0; /* no delta */
659 else if (!pack_size_limit)
660 usable_delta = 1; /* unlimited packfile */
661 else if (DELTA(entry)->idx.offset == (off_t)-1)
662 usable_delta = 0; /* base was written to another pack */
663 else if (DELTA(entry)->idx.offset)
664 usable_delta = 1; /* base already exists in this pack */
665 else
666 usable_delta = 0; /* base could end up in another pack */
668 if (!reuse_object)
669 to_reuse = 0; /* explicit */
670 else if (!IN_PACK(entry))
671 to_reuse = 0; /* can't reuse what we don't have */
672 else if (oe_type(entry) == OBJ_REF_DELTA ||
673 oe_type(entry) == OBJ_OFS_DELTA)
674 /* check_object() decided it for us ... */
675 to_reuse = usable_delta;
676 /* ... but pack split may override that */
677 else if (oe_type(entry) != entry->in_pack_type)
678 to_reuse = 0; /* pack has delta which is unusable */
679 else if (DELTA(entry))
680 to_reuse = 0; /* we want to pack afresh */
681 else
682 to_reuse = 1; /* we have it in-pack undeltified,
683 * and we do not need to deltify it.
686 if (!to_reuse)
687 len = write_no_reuse_object(f, entry, limit, usable_delta);
688 else
689 len = write_reuse_object(f, entry, limit, usable_delta);
690 if (!len)
691 return 0;
693 if (usable_delta)
694 written_delta++;
695 written++;
696 if (!pack_to_stdout)
697 entry->idx.crc32 = crc32_end(f);
698 return len;
701 enum write_one_status {
702 WRITE_ONE_SKIP = -1, /* already written */
703 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
704 WRITE_ONE_WRITTEN = 1, /* normal */
705 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
708 static enum write_one_status write_one(struct hashfile *f,
709 struct object_entry *e,
710 off_t *offset)
712 off_t size;
713 int recursing;
716 * we set offset to 1 (which is an impossible value) to mark
717 * the fact that this object is involved in "write its base
718 * first before writing a deltified object" recursion.
720 recursing = (e->idx.offset == 1);
721 if (recursing) {
722 warning(_("recursive delta detected for object %s"),
723 oid_to_hex(&e->idx.oid));
724 return WRITE_ONE_RECURSIVE;
725 } else if (e->idx.offset || e->preferred_base) {
726 /* offset is non zero if object is written already. */
727 return WRITE_ONE_SKIP;
730 /* if we are deltified, write out base object first. */
731 if (DELTA(e)) {
732 e->idx.offset = 1; /* now recurse */
733 switch (write_one(f, DELTA(e), offset)) {
734 case WRITE_ONE_RECURSIVE:
735 /* we cannot depend on this one */
736 SET_DELTA(e, NULL);
737 break;
738 default:
739 break;
740 case WRITE_ONE_BREAK:
741 e->idx.offset = recursing;
742 return WRITE_ONE_BREAK;
746 e->idx.offset = *offset;
747 size = write_object(f, e, *offset);
748 if (!size) {
749 e->idx.offset = recursing;
750 return WRITE_ONE_BREAK;
752 written_list[nr_written++] = &e->idx;
754 /* make sure off_t is sufficiently large not to wrap */
755 if (signed_add_overflows(*offset, size))
756 die(_("pack too large for current definition of off_t"));
757 *offset += size;
758 return WRITE_ONE_WRITTEN;
761 static int mark_tagged(const char *path, const struct object_id *oid, int flag,
762 void *cb_data)
764 struct object_id peeled;
765 struct object_entry *entry = packlist_find(&to_pack, oid);
767 if (entry)
768 entry->tagged = 1;
769 if (!peel_iterated_oid(oid, &peeled)) {
770 entry = packlist_find(&to_pack, &peeled);
771 if (entry)
772 entry->tagged = 1;
774 return 0;
777 static inline unsigned char oe_layer(struct packing_data *pack,
778 struct object_entry *e)
780 if (!pack->layer)
781 return 0;
782 return pack->layer[e - pack->objects];
785 static inline void add_to_write_order(struct object_entry **wo,
786 unsigned int *endp,
787 struct object_entry *e)
789 if (e->filled || oe_layer(&to_pack, e) != write_layer)
790 return;
791 wo[(*endp)++] = e;
792 e->filled = 1;
795 static void add_descendants_to_write_order(struct object_entry **wo,
796 unsigned int *endp,
797 struct object_entry *e)
799 int add_to_order = 1;
800 while (e) {
801 if (add_to_order) {
802 struct object_entry *s;
803 /* add this node... */
804 add_to_write_order(wo, endp, e);
805 /* all its siblings... */
806 for (s = DELTA_SIBLING(e); s; s = DELTA_SIBLING(s)) {
807 add_to_write_order(wo, endp, s);
810 /* drop down a level to add left subtree nodes if possible */
811 if (DELTA_CHILD(e)) {
812 add_to_order = 1;
813 e = DELTA_CHILD(e);
814 } else {
815 add_to_order = 0;
816 /* our sibling might have some children, it is next */
817 if (DELTA_SIBLING(e)) {
818 e = DELTA_SIBLING(e);
819 continue;
821 /* go back to our parent node */
822 e = DELTA(e);
823 while (e && !DELTA_SIBLING(e)) {
824 /* we're on the right side of a subtree, keep
825 * going up until we can go right again */
826 e = DELTA(e);
828 if (!e) {
829 /* done- we hit our original root node */
830 return;
832 /* pass it off to sibling at this level */
833 e = DELTA_SIBLING(e);
838 static void add_family_to_write_order(struct object_entry **wo,
839 unsigned int *endp,
840 struct object_entry *e)
842 struct object_entry *root;
844 for (root = e; DELTA(root); root = DELTA(root))
845 ; /* nothing */
846 add_descendants_to_write_order(wo, endp, root);
849 static void compute_layer_order(struct object_entry **wo, unsigned int *wo_end)
851 unsigned int i, last_untagged;
852 struct object_entry *objects = to_pack.objects;
854 for (i = 0; i < to_pack.nr_objects; i++) {
855 if (objects[i].tagged)
856 break;
857 add_to_write_order(wo, wo_end, &objects[i]);
859 last_untagged = i;
862 * Then fill all the tagged tips.
864 for (; i < to_pack.nr_objects; i++) {
865 if (objects[i].tagged)
866 add_to_write_order(wo, wo_end, &objects[i]);
870 * And then all remaining commits and tags.
872 for (i = last_untagged; i < to_pack.nr_objects; i++) {
873 if (oe_type(&objects[i]) != OBJ_COMMIT &&
874 oe_type(&objects[i]) != OBJ_TAG)
875 continue;
876 add_to_write_order(wo, wo_end, &objects[i]);
880 * And then all the trees.
882 for (i = last_untagged; i < to_pack.nr_objects; i++) {
883 if (oe_type(&objects[i]) != OBJ_TREE)
884 continue;
885 add_to_write_order(wo, wo_end, &objects[i]);
889 * Finally all the rest in really tight order
891 for (i = last_untagged; i < to_pack.nr_objects; i++) {
892 if (!objects[i].filled && oe_layer(&to_pack, &objects[i]) == write_layer)
893 add_family_to_write_order(wo, wo_end, &objects[i]);
897 static struct object_entry **compute_write_order(void)
899 uint32_t max_layers = 1;
900 unsigned int i, wo_end;
902 struct object_entry **wo;
903 struct object_entry *objects = to_pack.objects;
905 for (i = 0; i < to_pack.nr_objects; i++) {
906 objects[i].tagged = 0;
907 objects[i].filled = 0;
908 SET_DELTA_CHILD(&objects[i], NULL);
909 SET_DELTA_SIBLING(&objects[i], NULL);
913 * Fully connect delta_child/delta_sibling network.
914 * Make sure delta_sibling is sorted in the original
915 * recency order.
917 for (i = to_pack.nr_objects; i > 0;) {
918 struct object_entry *e = &objects[--i];
919 if (!DELTA(e))
920 continue;
921 /* Mark me as the first child */
922 e->delta_sibling_idx = DELTA(e)->delta_child_idx;
923 SET_DELTA_CHILD(DELTA(e), e);
927 * Mark objects that are at the tip of tags.
929 for_each_tag_ref(mark_tagged, NULL);
931 if (use_delta_islands)
932 max_layers = compute_pack_layers(&to_pack);
934 ALLOC_ARRAY(wo, to_pack.nr_objects);
935 wo_end = 0;
937 for (; write_layer < max_layers; ++write_layer)
938 compute_layer_order(wo, &wo_end);
940 if (wo_end != to_pack.nr_objects)
941 die(_("ordered %u objects, expected %"PRIu32),
942 wo_end, to_pack.nr_objects);
944 return wo;
949 * A reused set of objects. All objects in a chunk have the same
950 * relative position in the original packfile and the generated
951 * packfile.
954 static struct reused_chunk {
955 /* The offset of the first object of this chunk in the original
956 * packfile. */
957 off_t original;
958 /* The difference for "original" minus the offset of the first object of
959 * this chunk in the generated packfile. */
960 off_t difference;
961 } *reused_chunks;
962 static int reused_chunks_nr;
963 static int reused_chunks_alloc;
965 static void record_reused_object(off_t where, off_t offset)
967 if (reused_chunks_nr && reused_chunks[reused_chunks_nr-1].difference == offset)
968 return;
970 ALLOC_GROW(reused_chunks, reused_chunks_nr + 1,
971 reused_chunks_alloc);
972 reused_chunks[reused_chunks_nr].original = where;
973 reused_chunks[reused_chunks_nr].difference = offset;
974 reused_chunks_nr++;
978 * Binary search to find the chunk that "where" is in. Note
979 * that we're not looking for an exact match, just the first
980 * chunk that contains it (which implicitly ends at the start
981 * of the next chunk.
983 static off_t find_reused_offset(off_t where)
985 int lo = 0, hi = reused_chunks_nr;
986 while (lo < hi) {
987 int mi = lo + ((hi - lo) / 2);
988 if (where == reused_chunks[mi].original)
989 return reused_chunks[mi].difference;
990 if (where < reused_chunks[mi].original)
991 hi = mi;
992 else
993 lo = mi + 1;
997 * The first chunk starts at zero, so we can't have gone below
998 * there.
1000 assert(lo);
1001 return reused_chunks[lo-1].difference;
1004 static void write_reused_pack_one(size_t pos, struct hashfile *out,
1005 struct pack_window **w_curs)
1007 off_t offset, next, cur;
1008 enum object_type type;
1009 unsigned long size;
1011 offset = pack_pos_to_offset(reuse_packfile, pos);
1012 next = pack_pos_to_offset(reuse_packfile, pos + 1);
1014 record_reused_object(offset, offset - hashfile_total(out));
1016 cur = offset;
1017 type = unpack_object_header(reuse_packfile, w_curs, &cur, &size);
1018 assert(type >= 0);
1020 if (type == OBJ_OFS_DELTA) {
1021 off_t base_offset;
1022 off_t fixup;
1024 unsigned char header[MAX_PACK_OBJECT_HEADER];
1025 unsigned len;
1027 base_offset = get_delta_base(reuse_packfile, w_curs, &cur, type, offset);
1028 assert(base_offset != 0);
1030 /* Convert to REF_DELTA if we must... */
1031 if (!allow_ofs_delta) {
1032 uint32_t base_pos;
1033 struct object_id base_oid;
1035 if (offset_to_pack_pos(reuse_packfile, base_offset, &base_pos) < 0)
1036 die(_("expected object at offset %"PRIuMAX" "
1037 "in pack %s"),
1038 (uintmax_t)base_offset,
1039 reuse_packfile->pack_name);
1041 nth_packed_object_id(&base_oid, reuse_packfile,
1042 pack_pos_to_index(reuse_packfile, base_pos));
1044 len = encode_in_pack_object_header(header, sizeof(header),
1045 OBJ_REF_DELTA, size);
1046 hashwrite(out, header, len);
1047 hashwrite(out, base_oid.hash, the_hash_algo->rawsz);
1048 copy_pack_data(out, reuse_packfile, w_curs, cur, next - cur);
1049 return;
1052 /* Otherwise see if we need to rewrite the offset... */
1053 fixup = find_reused_offset(offset) -
1054 find_reused_offset(base_offset);
1055 if (fixup) {
1056 unsigned char ofs_header[10];
1057 unsigned i, ofs_len;
1058 off_t ofs = offset - base_offset - fixup;
1060 len = encode_in_pack_object_header(header, sizeof(header),
1061 OBJ_OFS_DELTA, size);
1063 i = sizeof(ofs_header) - 1;
1064 ofs_header[i] = ofs & 127;
1065 while (ofs >>= 7)
1066 ofs_header[--i] = 128 | (--ofs & 127);
1068 ofs_len = sizeof(ofs_header) - i;
1070 hashwrite(out, header, len);
1071 hashwrite(out, ofs_header + sizeof(ofs_header) - ofs_len, ofs_len);
1072 copy_pack_data(out, reuse_packfile, w_curs, cur, next - cur);
1073 return;
1076 /* ...otherwise we have no fixup, and can write it verbatim */
1079 copy_pack_data(out, reuse_packfile, w_curs, offset, next - offset);
1082 static size_t write_reused_pack_verbatim(struct hashfile *out,
1083 struct pack_window **w_curs)
1085 size_t pos = 0;
1087 while (pos < reuse_packfile_bitmap->word_alloc &&
1088 reuse_packfile_bitmap->words[pos] == (eword_t)~0)
1089 pos++;
1091 if (pos) {
1092 off_t to_write;
1094 written = (pos * BITS_IN_EWORD);
1095 to_write = pack_pos_to_offset(reuse_packfile, written)
1096 - sizeof(struct pack_header);
1098 /* We're recording one chunk, not one object. */
1099 record_reused_object(sizeof(struct pack_header), 0);
1100 hashflush(out);
1101 copy_pack_data(out, reuse_packfile, w_curs,
1102 sizeof(struct pack_header), to_write);
1104 display_progress(progress_state, written);
1106 return pos;
1109 static void write_reused_pack(struct hashfile *f)
1111 size_t i = 0;
1112 uint32_t offset;
1113 struct pack_window *w_curs = NULL;
1115 if (allow_ofs_delta)
1116 i = write_reused_pack_verbatim(f, &w_curs);
1118 for (; i < reuse_packfile_bitmap->word_alloc; ++i) {
1119 eword_t word = reuse_packfile_bitmap->words[i];
1120 size_t pos = (i * BITS_IN_EWORD);
1122 for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
1123 if ((word >> offset) == 0)
1124 break;
1126 offset += ewah_bit_ctz64(word >> offset);
1127 write_reused_pack_one(pos + offset, f, &w_curs);
1128 display_progress(progress_state, ++written);
1132 unuse_pack(&w_curs);
1135 static void write_excluded_by_configs(void)
1137 struct oidset_iter iter;
1138 const struct object_id *oid;
1140 oidset_iter_init(&excluded_by_config, &iter);
1141 while ((oid = oidset_iter_next(&iter))) {
1142 struct configured_exclusion *ex =
1143 oidmap_get(&configured_exclusions, oid);
1145 if (!ex)
1146 BUG("configured exclusion wasn't configured");
1147 write_in_full(1, ex->pack_hash_hex, strlen(ex->pack_hash_hex));
1148 write_in_full(1, " ", 1);
1149 write_in_full(1, ex->uri, strlen(ex->uri));
1150 write_in_full(1, "\n", 1);
1154 static const char no_split_warning[] = N_(
1155 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
1158 static void write_pack_file(void)
1160 uint32_t i = 0, j;
1161 struct hashfile *f;
1162 off_t offset;
1163 uint32_t nr_remaining = nr_result;
1164 time_t last_mtime = 0;
1165 struct object_entry **write_order;
1167 if (progress > pack_to_stdout)
1168 progress_state = start_progress(_("Writing objects"), nr_result);
1169 ALLOC_ARRAY(written_list, to_pack.nr_objects);
1170 write_order = compute_write_order();
1172 do {
1173 unsigned char hash[GIT_MAX_RAWSZ];
1174 char *pack_tmp_name = NULL;
1176 if (pack_to_stdout)
1177 f = hashfd_throughput(1, "<stdout>", progress_state);
1178 else
1179 f = create_tmp_packfile(&pack_tmp_name);
1181 offset = write_pack_header(f, nr_remaining);
1183 if (reuse_packfile) {
1184 assert(pack_to_stdout);
1185 write_reused_pack(f);
1186 offset = hashfile_total(f);
1189 nr_written = 0;
1190 for (; i < to_pack.nr_objects; i++) {
1191 struct object_entry *e = write_order[i];
1192 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
1193 break;
1194 display_progress(progress_state, written);
1198 * Did we write the wrong # entries in the header?
1199 * If so, rewrite it like in fast-import
1201 if (pack_to_stdout) {
1202 finalize_hashfile(f, hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE);
1203 } else if (nr_written == nr_remaining) {
1204 finalize_hashfile(f, hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE);
1205 } else {
1206 int fd = finalize_hashfile(f, hash, 0);
1207 fixup_pack_header_footer(fd, hash, pack_tmp_name,
1208 nr_written, hash, offset);
1209 close(fd);
1210 if (write_bitmap_index) {
1211 if (write_bitmap_index != WRITE_BITMAP_QUIET)
1212 warning(_(no_split_warning));
1213 write_bitmap_index = 0;
1217 if (!pack_to_stdout) {
1218 struct stat st;
1219 struct strbuf tmpname = STRBUF_INIT;
1222 * Packs are runtime accessed in their mtime
1223 * order since newer packs are more likely to contain
1224 * younger objects. So if we are creating multiple
1225 * packs then we should modify the mtime of later ones
1226 * to preserve this property.
1228 if (stat(pack_tmp_name, &st) < 0) {
1229 warning_errno(_("failed to stat %s"), pack_tmp_name);
1230 } else if (!last_mtime) {
1231 last_mtime = st.st_mtime;
1232 } else {
1233 struct utimbuf utb;
1234 utb.actime = st.st_atime;
1235 utb.modtime = --last_mtime;
1236 if (utime(pack_tmp_name, &utb) < 0)
1237 warning_errno(_("failed utime() on %s"), pack_tmp_name);
1240 strbuf_addf(&tmpname, "%s-", base_name);
1242 if (write_bitmap_index) {
1243 bitmap_writer_set_checksum(hash);
1244 bitmap_writer_build_type_index(
1245 &to_pack, written_list, nr_written);
1248 finish_tmp_packfile(&tmpname, pack_tmp_name,
1249 written_list, nr_written,
1250 &pack_idx_opts, hash);
1252 if (write_bitmap_index) {
1253 strbuf_addf(&tmpname, "%s.bitmap", hash_to_hex(hash));
1255 stop_progress(&progress_state);
1257 bitmap_writer_show_progress(progress);
1258 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
1259 bitmap_writer_build(&to_pack);
1260 bitmap_writer_finish(written_list, nr_written,
1261 tmpname.buf, write_bitmap_options);
1262 write_bitmap_index = 0;
1265 strbuf_release(&tmpname);
1266 free(pack_tmp_name);
1267 puts(hash_to_hex(hash));
1270 /* mark written objects as written to previous pack */
1271 for (j = 0; j < nr_written; j++) {
1272 written_list[j]->offset = (off_t)-1;
1274 nr_remaining -= nr_written;
1275 } while (nr_remaining && i < to_pack.nr_objects);
1277 free(written_list);
1278 free(write_order);
1279 stop_progress(&progress_state);
1280 if (written != nr_result)
1281 die(_("wrote %"PRIu32" objects while expecting %"PRIu32),
1282 written, nr_result);
1283 trace2_data_intmax("pack-objects", the_repository,
1284 "write_pack_file/wrote", nr_result);
1287 static int no_try_delta(const char *path)
1289 static struct attr_check *check;
1291 if (!check)
1292 check = attr_check_initl("delta", NULL);
1293 git_check_attr(the_repository->index, path, check);
1294 if (ATTR_FALSE(check->items[0].value))
1295 return 1;
1296 return 0;
1300 * When adding an object, check whether we have already added it
1301 * to our packing list. If so, we can skip. However, if we are
1302 * being asked to excludei t, but the previous mention was to include
1303 * it, make sure to adjust its flags and tweak our numbers accordingly.
1305 * As an optimization, we pass out the index position where we would have
1306 * found the item, since that saves us from having to look it up again a
1307 * few lines later when we want to add the new entry.
1309 static int have_duplicate_entry(const struct object_id *oid,
1310 int exclude)
1312 struct object_entry *entry;
1314 if (reuse_packfile_bitmap &&
1315 bitmap_walk_contains(bitmap_git, reuse_packfile_bitmap, oid))
1316 return 1;
1318 entry = packlist_find(&to_pack, oid);
1319 if (!entry)
1320 return 0;
1322 if (exclude) {
1323 if (!entry->preferred_base)
1324 nr_result--;
1325 entry->preferred_base = 1;
1328 return 1;
1331 static int want_found_object(const struct object_id *oid, int exclude,
1332 struct packed_git *p)
1334 if (exclude)
1335 return 1;
1336 if (incremental)
1337 return 0;
1340 * When asked to do --local (do not include an object that appears in a
1341 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
1342 * an object that appears in a pack marked with .keep), finding a pack
1343 * that matches the criteria is sufficient for us to decide to omit it.
1344 * However, even if this pack does not satisfy the criteria, we need to
1345 * make sure no copy of this object appears in _any_ pack that makes us
1346 * to omit the object, so we need to check all the packs.
1348 * We can however first check whether these options can possibly matter;
1349 * if they do not matter we know we want the object in generated pack.
1350 * Otherwise, we signal "-1" at the end to tell the caller that we do
1351 * not know either way, and it needs to check more packs.
1355 * Objects in packs borrowed from elsewhere are discarded regardless of
1356 * if they appear in other packs that weren't borrowed.
1358 if (local && !p->pack_local)
1359 return 0;
1362 * Then handle .keep first, as we have a fast(er) path there.
1364 if (ignore_packed_keep_on_disk || ignore_packed_keep_in_core) {
1366 * Set the flags for the kept-pack cache to be the ones we want
1367 * to ignore.
1369 * That is, if we are ignoring objects in on-disk keep packs,
1370 * then we want to search through the on-disk keep and ignore
1371 * the in-core ones.
1373 unsigned flags = 0;
1374 if (ignore_packed_keep_on_disk)
1375 flags |= ON_DISK_KEEP_PACKS;
1376 if (ignore_packed_keep_in_core)
1377 flags |= IN_CORE_KEEP_PACKS;
1379 if (ignore_packed_keep_on_disk && p->pack_keep)
1380 return 0;
1381 if (ignore_packed_keep_in_core && p->pack_keep_in_core)
1382 return 0;
1383 if (has_object_kept_pack(oid, flags))
1384 return 0;
1388 * At this point we know definitively that either we don't care about
1389 * keep-packs, or the object is not in one. Keep checking other
1390 * conditions...
1392 if (!local || !have_non_local_packs)
1393 return 1;
1395 /* we don't know yet; keep looking for more packs */
1396 return -1;
1399 static int want_object_in_pack_one(struct packed_git *p,
1400 const struct object_id *oid,
1401 int exclude,
1402 struct packed_git **found_pack,
1403 off_t *found_offset)
1405 off_t offset;
1407 if (p == *found_pack)
1408 offset = *found_offset;
1409 else
1410 offset = find_pack_entry_one(oid->hash, p);
1412 if (offset) {
1413 if (!*found_pack) {
1414 if (!is_pack_valid(p))
1415 return -1;
1416 *found_offset = offset;
1417 *found_pack = p;
1419 return want_found_object(oid, exclude, p);
1421 return -1;
1425 * Check whether we want the object in the pack (e.g., we do not want
1426 * objects found in non-local stores if the "--local" option was used).
1428 * If the caller already knows an existing pack it wants to take the object
1429 * from, that is passed in *found_pack and *found_offset; otherwise this
1430 * function finds if there is any pack that has the object and returns the pack
1431 * and its offset in these variables.
1433 static int want_object_in_pack(const struct object_id *oid,
1434 int exclude,
1435 struct packed_git **found_pack,
1436 off_t *found_offset)
1438 int want;
1439 struct list_head *pos;
1440 struct multi_pack_index *m;
1442 if (!exclude && local && has_loose_object_nonlocal(oid))
1443 return 0;
1446 * If we already know the pack object lives in, start checks from that
1447 * pack - in the usual case when neither --local was given nor .keep files
1448 * are present we will determine the answer right now.
1450 if (*found_pack) {
1451 want = want_found_object(oid, exclude, *found_pack);
1452 if (want != -1)
1453 return want;
1456 for (m = get_multi_pack_index(the_repository); m; m = m->next) {
1457 struct pack_entry e;
1458 if (fill_midx_entry(the_repository, oid, &e, m)) {
1459 want = want_object_in_pack_one(e.p, oid, exclude, found_pack, found_offset);
1460 if (want != -1)
1461 return want;
1465 list_for_each(pos, get_packed_git_mru(the_repository)) {
1466 struct packed_git *p = list_entry(pos, struct packed_git, mru);
1467 want = want_object_in_pack_one(p, oid, exclude, found_pack, found_offset);
1468 if (!exclude && want > 0)
1469 list_move(&p->mru,
1470 get_packed_git_mru(the_repository));
1471 if (want != -1)
1472 return want;
1475 if (uri_protocols.nr) {
1476 struct configured_exclusion *ex =
1477 oidmap_get(&configured_exclusions, oid);
1478 int i;
1479 const char *p;
1481 if (ex) {
1482 for (i = 0; i < uri_protocols.nr; i++) {
1483 if (skip_prefix(ex->uri,
1484 uri_protocols.items[i].string,
1485 &p) &&
1486 *p == ':') {
1487 oidset_insert(&excluded_by_config, oid);
1488 return 0;
1494 return 1;
1497 static void create_object_entry(const struct object_id *oid,
1498 enum object_type type,
1499 uint32_t hash,
1500 int exclude,
1501 int no_try_delta,
1502 struct packed_git *found_pack,
1503 off_t found_offset)
1505 struct object_entry *entry;
1507 entry = packlist_alloc(&to_pack, oid);
1508 entry->hash = hash;
1509 oe_set_type(entry, type);
1510 if (exclude)
1511 entry->preferred_base = 1;
1512 else
1513 nr_result++;
1514 if (found_pack) {
1515 oe_set_in_pack(&to_pack, entry, found_pack);
1516 entry->in_pack_offset = found_offset;
1519 entry->no_try_delta = no_try_delta;
1522 static const char no_closure_warning[] = N_(
1523 "disabling bitmap writing, as some objects are not being packed"
1526 static int add_object_entry(const struct object_id *oid, enum object_type type,
1527 const char *name, int exclude)
1529 struct packed_git *found_pack = NULL;
1530 off_t found_offset = 0;
1532 display_progress(progress_state, ++nr_seen);
1534 if (have_duplicate_entry(oid, exclude))
1535 return 0;
1537 if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {
1538 /* The pack is missing an object, so it will not have closure */
1539 if (write_bitmap_index) {
1540 if (write_bitmap_index != WRITE_BITMAP_QUIET)
1541 warning(_(no_closure_warning));
1542 write_bitmap_index = 0;
1544 return 0;
1547 create_object_entry(oid, type, pack_name_hash(name),
1548 exclude, name && no_try_delta(name),
1549 found_pack, found_offset);
1550 return 1;
1553 static int add_object_entry_from_bitmap(const struct object_id *oid,
1554 enum object_type type,
1555 int flags, uint32_t name_hash,
1556 struct packed_git *pack, off_t offset)
1558 display_progress(progress_state, ++nr_seen);
1560 if (have_duplicate_entry(oid, 0))
1561 return 0;
1563 if (!want_object_in_pack(oid, 0, &pack, &offset))
1564 return 0;
1566 create_object_entry(oid, type, name_hash, 0, 0, pack, offset);
1567 return 1;
1570 struct pbase_tree_cache {
1571 struct object_id oid;
1572 int ref;
1573 int temporary;
1574 void *tree_data;
1575 unsigned long tree_size;
1578 static struct pbase_tree_cache *(pbase_tree_cache[256]);
1579 static int pbase_tree_cache_ix(const struct object_id *oid)
1581 return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);
1583 static int pbase_tree_cache_ix_incr(int ix)
1585 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1588 static struct pbase_tree {
1589 struct pbase_tree *next;
1590 /* This is a phony "cache" entry; we are not
1591 * going to evict it or find it through _get()
1592 * mechanism -- this is for the toplevel node that
1593 * would almost always change with any commit.
1595 struct pbase_tree_cache pcache;
1596 } *pbase_tree;
1598 static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)
1600 struct pbase_tree_cache *ent, *nent;
1601 void *data;
1602 unsigned long size;
1603 enum object_type type;
1604 int neigh;
1605 int my_ix = pbase_tree_cache_ix(oid);
1606 int available_ix = -1;
1608 /* pbase-tree-cache acts as a limited hashtable.
1609 * your object will be found at your index or within a few
1610 * slots after that slot if it is cached.
1612 for (neigh = 0; neigh < 8; neigh++) {
1613 ent = pbase_tree_cache[my_ix];
1614 if (ent && oideq(&ent->oid, oid)) {
1615 ent->ref++;
1616 return ent;
1618 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1619 ((0 <= available_ix) &&
1620 (!ent && pbase_tree_cache[available_ix])))
1621 available_ix = my_ix;
1622 if (!ent)
1623 break;
1624 my_ix = pbase_tree_cache_ix_incr(my_ix);
1627 /* Did not find one. Either we got a bogus request or
1628 * we need to read and perhaps cache.
1630 data = read_object_file(oid, &type, &size);
1631 if (!data)
1632 return NULL;
1633 if (type != OBJ_TREE) {
1634 free(data);
1635 return NULL;
1638 /* We need to either cache or return a throwaway copy */
1640 if (available_ix < 0)
1641 ent = NULL;
1642 else {
1643 ent = pbase_tree_cache[available_ix];
1644 my_ix = available_ix;
1647 if (!ent) {
1648 nent = xmalloc(sizeof(*nent));
1649 nent->temporary = (available_ix < 0);
1651 else {
1652 /* evict and reuse */
1653 free(ent->tree_data);
1654 nent = ent;
1656 oidcpy(&nent->oid, oid);
1657 nent->tree_data = data;
1658 nent->tree_size = size;
1659 nent->ref = 1;
1660 if (!nent->temporary)
1661 pbase_tree_cache[my_ix] = nent;
1662 return nent;
1665 static void pbase_tree_put(struct pbase_tree_cache *cache)
1667 if (!cache->temporary) {
1668 cache->ref--;
1669 return;
1671 free(cache->tree_data);
1672 free(cache);
1675 static int name_cmp_len(const char *name)
1677 int i;
1678 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1680 return i;
1683 static void add_pbase_object(struct tree_desc *tree,
1684 const char *name,
1685 int cmplen,
1686 const char *fullname)
1688 struct name_entry entry;
1689 int cmp;
1691 while (tree_entry(tree,&entry)) {
1692 if (S_ISGITLINK(entry.mode))
1693 continue;
1694 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1695 memcmp(name, entry.path, cmplen);
1696 if (cmp > 0)
1697 continue;
1698 if (cmp < 0)
1699 return;
1700 if (name[cmplen] != '/') {
1701 add_object_entry(&entry.oid,
1702 object_type(entry.mode),
1703 fullname, 1);
1704 return;
1706 if (S_ISDIR(entry.mode)) {
1707 struct tree_desc sub;
1708 struct pbase_tree_cache *tree;
1709 const char *down = name+cmplen+1;
1710 int downlen = name_cmp_len(down);
1712 tree = pbase_tree_get(&entry.oid);
1713 if (!tree)
1714 return;
1715 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1717 add_pbase_object(&sub, down, downlen, fullname);
1718 pbase_tree_put(tree);
1723 static unsigned *done_pbase_paths;
1724 static int done_pbase_paths_num;
1725 static int done_pbase_paths_alloc;
1726 static int done_pbase_path_pos(unsigned hash)
1728 int lo = 0;
1729 int hi = done_pbase_paths_num;
1730 while (lo < hi) {
1731 int mi = lo + (hi - lo) / 2;
1732 if (done_pbase_paths[mi] == hash)
1733 return mi;
1734 if (done_pbase_paths[mi] < hash)
1735 hi = mi;
1736 else
1737 lo = mi + 1;
1739 return -lo-1;
1742 static int check_pbase_path(unsigned hash)
1744 int pos = done_pbase_path_pos(hash);
1745 if (0 <= pos)
1746 return 1;
1747 pos = -pos - 1;
1748 ALLOC_GROW(done_pbase_paths,
1749 done_pbase_paths_num + 1,
1750 done_pbase_paths_alloc);
1751 done_pbase_paths_num++;
1752 if (pos < done_pbase_paths_num)
1753 MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,
1754 done_pbase_paths_num - pos - 1);
1755 done_pbase_paths[pos] = hash;
1756 return 0;
1759 static void add_preferred_base_object(const char *name)
1761 struct pbase_tree *it;
1762 int cmplen;
1763 unsigned hash = pack_name_hash(name);
1765 if (!num_preferred_base || check_pbase_path(hash))
1766 return;
1768 cmplen = name_cmp_len(name);
1769 for (it = pbase_tree; it; it = it->next) {
1770 if (cmplen == 0) {
1771 add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);
1773 else {
1774 struct tree_desc tree;
1775 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1776 add_pbase_object(&tree, name, cmplen, name);
1781 static void add_preferred_base(struct object_id *oid)
1783 struct pbase_tree *it;
1784 void *data;
1785 unsigned long size;
1786 struct object_id tree_oid;
1788 if (window <= num_preferred_base++)
1789 return;
1791 data = read_object_with_reference(the_repository, oid,
1792 tree_type, &size, &tree_oid);
1793 if (!data)
1794 return;
1796 for (it = pbase_tree; it; it = it->next) {
1797 if (oideq(&it->pcache.oid, &tree_oid)) {
1798 free(data);
1799 return;
1803 CALLOC_ARRAY(it, 1);
1804 it->next = pbase_tree;
1805 pbase_tree = it;
1807 oidcpy(&it->pcache.oid, &tree_oid);
1808 it->pcache.tree_data = data;
1809 it->pcache.tree_size = size;
1812 static void cleanup_preferred_base(void)
1814 struct pbase_tree *it;
1815 unsigned i;
1817 it = pbase_tree;
1818 pbase_tree = NULL;
1819 while (it) {
1820 struct pbase_tree *tmp = it;
1821 it = tmp->next;
1822 free(tmp->pcache.tree_data);
1823 free(tmp);
1826 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1827 if (!pbase_tree_cache[i])
1828 continue;
1829 free(pbase_tree_cache[i]->tree_data);
1830 FREE_AND_NULL(pbase_tree_cache[i]);
1833 FREE_AND_NULL(done_pbase_paths);
1834 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1838 * Return 1 iff the object specified by "delta" can be sent
1839 * literally as a delta against the base in "base_sha1". If
1840 * so, then *base_out will point to the entry in our packing
1841 * list, or NULL if we must use the external-base list.
1843 * Depth value does not matter - find_deltas() will
1844 * never consider reused delta as the base object to
1845 * deltify other objects against, in order to avoid
1846 * circular deltas.
1848 static int can_reuse_delta(const struct object_id *base_oid,
1849 struct object_entry *delta,
1850 struct object_entry **base_out)
1852 struct object_entry *base;
1855 * First see if we're already sending the base (or it's explicitly in
1856 * our "excluded" list).
1858 base = packlist_find(&to_pack, base_oid);
1859 if (base) {
1860 if (!in_same_island(&delta->idx.oid, &base->idx.oid))
1861 return 0;
1862 *base_out = base;
1863 return 1;
1867 * Otherwise, reachability bitmaps may tell us if the receiver has it,
1868 * even if it was buried too deep in history to make it into the
1869 * packing list.
1871 if (thin && bitmap_has_oid_in_uninteresting(bitmap_git, base_oid)) {
1872 if (use_delta_islands) {
1873 if (!in_same_island(&delta->idx.oid, base_oid))
1874 return 0;
1876 *base_out = NULL;
1877 return 1;
1880 return 0;
1883 static void prefetch_to_pack(uint32_t object_index_start) {
1884 struct oid_array to_fetch = OID_ARRAY_INIT;
1885 uint32_t i;
1887 for (i = object_index_start; i < to_pack.nr_objects; i++) {
1888 struct object_entry *entry = to_pack.objects + i;
1890 if (!oid_object_info_extended(the_repository,
1891 &entry->idx.oid,
1892 NULL,
1893 OBJECT_INFO_FOR_PREFETCH))
1894 continue;
1895 oid_array_append(&to_fetch, &entry->idx.oid);
1897 promisor_remote_get_direct(the_repository,
1898 to_fetch.oid, to_fetch.nr);
1899 oid_array_clear(&to_fetch);
1902 static void check_object(struct object_entry *entry, uint32_t object_index)
1904 unsigned long canonical_size;
1905 enum object_type type;
1906 struct object_info oi = {.typep = &type, .sizep = &canonical_size};
1908 if (IN_PACK(entry)) {
1909 struct packed_git *p = IN_PACK(entry);
1910 struct pack_window *w_curs = NULL;
1911 int have_base = 0;
1912 struct object_id base_ref;
1913 struct object_entry *base_entry;
1914 unsigned long used, used_0;
1915 unsigned long avail;
1916 off_t ofs;
1917 unsigned char *buf, c;
1918 enum object_type type;
1919 unsigned long in_pack_size;
1921 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1924 * We want in_pack_type even if we do not reuse delta
1925 * since non-delta representations could still be reused.
1927 used = unpack_object_header_buffer(buf, avail,
1928 &type,
1929 &in_pack_size);
1930 if (used == 0)
1931 goto give_up;
1933 if (type < 0)
1934 BUG("invalid type %d", type);
1935 entry->in_pack_type = type;
1938 * Determine if this is a delta and if so whether we can
1939 * reuse it or not. Otherwise let's find out as cheaply as
1940 * possible what the actual type and size for this object is.
1942 switch (entry->in_pack_type) {
1943 default:
1944 /* Not a delta hence we've already got all we need. */
1945 oe_set_type(entry, entry->in_pack_type);
1946 SET_SIZE(entry, in_pack_size);
1947 entry->in_pack_header_size = used;
1948 if (oe_type(entry) < OBJ_COMMIT || oe_type(entry) > OBJ_BLOB)
1949 goto give_up;
1950 unuse_pack(&w_curs);
1951 return;
1952 case OBJ_REF_DELTA:
1953 if (reuse_delta && !entry->preferred_base) {
1954 oidread(&base_ref,
1955 use_pack(p, &w_curs,
1956 entry->in_pack_offset + used,
1957 NULL));
1958 have_base = 1;
1960 entry->in_pack_header_size = used + the_hash_algo->rawsz;
1961 break;
1962 case OBJ_OFS_DELTA:
1963 buf = use_pack(p, &w_curs,
1964 entry->in_pack_offset + used, NULL);
1965 used_0 = 0;
1966 c = buf[used_0++];
1967 ofs = c & 127;
1968 while (c & 128) {
1969 ofs += 1;
1970 if (!ofs || MSB(ofs, 7)) {
1971 error(_("delta base offset overflow in pack for %s"),
1972 oid_to_hex(&entry->idx.oid));
1973 goto give_up;
1975 c = buf[used_0++];
1976 ofs = (ofs << 7) + (c & 127);
1978 ofs = entry->in_pack_offset - ofs;
1979 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1980 error(_("delta base offset out of bound for %s"),
1981 oid_to_hex(&entry->idx.oid));
1982 goto give_up;
1984 if (reuse_delta && !entry->preferred_base) {
1985 uint32_t pos;
1986 if (offset_to_pack_pos(p, ofs, &pos) < 0)
1987 goto give_up;
1988 if (!nth_packed_object_id(&base_ref, p,
1989 pack_pos_to_index(p, pos)))
1990 have_base = 1;
1992 entry->in_pack_header_size = used + used_0;
1993 break;
1996 if (have_base &&
1997 can_reuse_delta(&base_ref, entry, &base_entry)) {
1998 oe_set_type(entry, entry->in_pack_type);
1999 SET_SIZE(entry, in_pack_size); /* delta size */
2000 SET_DELTA_SIZE(entry, in_pack_size);
2002 if (base_entry) {
2003 SET_DELTA(entry, base_entry);
2004 entry->delta_sibling_idx = base_entry->delta_child_idx;
2005 SET_DELTA_CHILD(base_entry, entry);
2006 } else {
2007 SET_DELTA_EXT(entry, &base_ref);
2010 unuse_pack(&w_curs);
2011 return;
2014 if (oe_type(entry)) {
2015 off_t delta_pos;
2018 * This must be a delta and we already know what the
2019 * final object type is. Let's extract the actual
2020 * object size from the delta header.
2022 delta_pos = entry->in_pack_offset + entry->in_pack_header_size;
2023 canonical_size = get_size_from_delta(p, &w_curs, delta_pos);
2024 if (canonical_size == 0)
2025 goto give_up;
2026 SET_SIZE(entry, canonical_size);
2027 unuse_pack(&w_curs);
2028 return;
2032 * No choice but to fall back to the recursive delta walk
2033 * with oid_object_info() to find about the object type
2034 * at this point...
2036 give_up:
2037 unuse_pack(&w_curs);
2040 if (oid_object_info_extended(the_repository, &entry->idx.oid, &oi,
2041 OBJECT_INFO_SKIP_FETCH_OBJECT | OBJECT_INFO_LOOKUP_REPLACE) < 0) {
2042 if (has_promisor_remote()) {
2043 prefetch_to_pack(object_index);
2044 if (oid_object_info_extended(the_repository, &entry->idx.oid, &oi,
2045 OBJECT_INFO_SKIP_FETCH_OBJECT | OBJECT_INFO_LOOKUP_REPLACE) < 0)
2046 type = -1;
2047 } else {
2048 type = -1;
2051 oe_set_type(entry, type);
2052 if (entry->type_valid) {
2053 SET_SIZE(entry, canonical_size);
2054 } else {
2056 * Bad object type is checked in prepare_pack(). This is
2057 * to permit a missing preferred base object to be ignored
2058 * as a preferred base. Doing so can result in a larger
2059 * pack file, but the transfer will still take place.
2064 static int pack_offset_sort(const void *_a, const void *_b)
2066 const struct object_entry *a = *(struct object_entry **)_a;
2067 const struct object_entry *b = *(struct object_entry **)_b;
2068 const struct packed_git *a_in_pack = IN_PACK(a);
2069 const struct packed_git *b_in_pack = IN_PACK(b);
2071 /* avoid filesystem trashing with loose objects */
2072 if (!a_in_pack && !b_in_pack)
2073 return oidcmp(&a->idx.oid, &b->idx.oid);
2075 if (a_in_pack < b_in_pack)
2076 return -1;
2077 if (a_in_pack > b_in_pack)
2078 return 1;
2079 return a->in_pack_offset < b->in_pack_offset ? -1 :
2080 (a->in_pack_offset > b->in_pack_offset);
2084 * Drop an on-disk delta we were planning to reuse. Naively, this would
2085 * just involve blanking out the "delta" field, but we have to deal
2086 * with some extra book-keeping:
2088 * 1. Removing ourselves from the delta_sibling linked list.
2090 * 2. Updating our size/type to the non-delta representation. These were
2091 * either not recorded initially (size) or overwritten with the delta type
2092 * (type) when check_object() decided to reuse the delta.
2094 * 3. Resetting our delta depth, as we are now a base object.
2096 static void drop_reused_delta(struct object_entry *entry)
2098 unsigned *idx = &to_pack.objects[entry->delta_idx - 1].delta_child_idx;
2099 struct object_info oi = OBJECT_INFO_INIT;
2100 enum object_type type;
2101 unsigned long size;
2103 while (*idx) {
2104 struct object_entry *oe = &to_pack.objects[*idx - 1];
2106 if (oe == entry)
2107 *idx = oe->delta_sibling_idx;
2108 else
2109 idx = &oe->delta_sibling_idx;
2111 SET_DELTA(entry, NULL);
2112 entry->depth = 0;
2114 oi.sizep = &size;
2115 oi.typep = &type;
2116 if (packed_object_info(the_repository, IN_PACK(entry), entry->in_pack_offset, &oi) < 0) {
2118 * We failed to get the info from this pack for some reason;
2119 * fall back to oid_object_info, which may find another copy.
2120 * And if that fails, the error will be recorded in oe_type(entry)
2121 * and dealt with in prepare_pack().
2123 oe_set_type(entry,
2124 oid_object_info(the_repository, &entry->idx.oid, &size));
2125 } else {
2126 oe_set_type(entry, type);
2128 SET_SIZE(entry, size);
2132 * Follow the chain of deltas from this entry onward, throwing away any links
2133 * that cause us to hit a cycle (as determined by the DFS state flags in
2134 * the entries).
2136 * We also detect too-long reused chains that would violate our --depth
2137 * limit.
2139 static void break_delta_chains(struct object_entry *entry)
2142 * The actual depth of each object we will write is stored as an int,
2143 * as it cannot exceed our int "depth" limit. But before we break
2144 * changes based no that limit, we may potentially go as deep as the
2145 * number of objects, which is elsewhere bounded to a uint32_t.
2147 uint32_t total_depth;
2148 struct object_entry *cur, *next;
2150 for (cur = entry, total_depth = 0;
2151 cur;
2152 cur = DELTA(cur), total_depth++) {
2153 if (cur->dfs_state == DFS_DONE) {
2155 * We've already seen this object and know it isn't
2156 * part of a cycle. We do need to append its depth
2157 * to our count.
2159 total_depth += cur->depth;
2160 break;
2164 * We break cycles before looping, so an ACTIVE state (or any
2165 * other cruft which made its way into the state variable)
2166 * is a bug.
2168 if (cur->dfs_state != DFS_NONE)
2169 BUG("confusing delta dfs state in first pass: %d",
2170 cur->dfs_state);
2173 * Now we know this is the first time we've seen the object. If
2174 * it's not a delta, we're done traversing, but we'll mark it
2175 * done to save time on future traversals.
2177 if (!DELTA(cur)) {
2178 cur->dfs_state = DFS_DONE;
2179 break;
2183 * Mark ourselves as active and see if the next step causes
2184 * us to cycle to another active object. It's important to do
2185 * this _before_ we loop, because it impacts where we make the
2186 * cut, and thus how our total_depth counter works.
2187 * E.g., We may see a partial loop like:
2189 * A -> B -> C -> D -> B
2191 * Cutting B->C breaks the cycle. But now the depth of A is
2192 * only 1, and our total_depth counter is at 3. The size of the
2193 * error is always one less than the size of the cycle we
2194 * broke. Commits C and D were "lost" from A's chain.
2196 * If we instead cut D->B, then the depth of A is correct at 3.
2197 * We keep all commits in the chain that we examined.
2199 cur->dfs_state = DFS_ACTIVE;
2200 if (DELTA(cur)->dfs_state == DFS_ACTIVE) {
2201 drop_reused_delta(cur);
2202 cur->dfs_state = DFS_DONE;
2203 break;
2208 * And now that we've gone all the way to the bottom of the chain, we
2209 * need to clear the active flags and set the depth fields as
2210 * appropriate. Unlike the loop above, which can quit when it drops a
2211 * delta, we need to keep going to look for more depth cuts. So we need
2212 * an extra "next" pointer to keep going after we reset cur->delta.
2214 for (cur = entry; cur; cur = next) {
2215 next = DELTA(cur);
2218 * We should have a chain of zero or more ACTIVE states down to
2219 * a final DONE. We can quit after the DONE, because either it
2220 * has no bases, or we've already handled them in a previous
2221 * call.
2223 if (cur->dfs_state == DFS_DONE)
2224 break;
2225 else if (cur->dfs_state != DFS_ACTIVE)
2226 BUG("confusing delta dfs state in second pass: %d",
2227 cur->dfs_state);
2230 * If the total_depth is more than depth, then we need to snip
2231 * the chain into two or more smaller chains that don't exceed
2232 * the maximum depth. Most of the resulting chains will contain
2233 * (depth + 1) entries (i.e., depth deltas plus one base), and
2234 * the last chain (i.e., the one containing entry) will contain
2235 * whatever entries are left over, namely
2236 * (total_depth % (depth + 1)) of them.
2238 * Since we are iterating towards decreasing depth, we need to
2239 * decrement total_depth as we go, and we need to write to the
2240 * entry what its final depth will be after all of the
2241 * snipping. Since we're snipping into chains of length (depth
2242 * + 1) entries, the final depth of an entry will be its
2243 * original depth modulo (depth + 1). Any time we encounter an
2244 * entry whose final depth is supposed to be zero, we snip it
2245 * from its delta base, thereby making it so.
2247 cur->depth = (total_depth--) % (depth + 1);
2248 if (!cur->depth)
2249 drop_reused_delta(cur);
2251 cur->dfs_state = DFS_DONE;
2255 static void get_object_details(void)
2257 uint32_t i;
2258 struct object_entry **sorted_by_offset;
2260 if (progress)
2261 progress_state = start_progress(_("Counting objects"),
2262 to_pack.nr_objects);
2264 CALLOC_ARRAY(sorted_by_offset, to_pack.nr_objects);
2265 for (i = 0; i < to_pack.nr_objects; i++)
2266 sorted_by_offset[i] = to_pack.objects + i;
2267 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
2269 for (i = 0; i < to_pack.nr_objects; i++) {
2270 struct object_entry *entry = sorted_by_offset[i];
2271 check_object(entry, i);
2272 if (entry->type_valid &&
2273 oe_size_greater_than(&to_pack, entry, big_file_threshold))
2274 entry->no_try_delta = 1;
2275 display_progress(progress_state, i + 1);
2277 stop_progress(&progress_state);
2280 * This must happen in a second pass, since we rely on the delta
2281 * information for the whole list being completed.
2283 for (i = 0; i < to_pack.nr_objects; i++)
2284 break_delta_chains(&to_pack.objects[i]);
2286 free(sorted_by_offset);
2290 * We search for deltas in a list sorted by type, by filename hash, and then
2291 * by size, so that we see progressively smaller and smaller files.
2292 * That's because we prefer deltas to be from the bigger file
2293 * to the smaller -- deletes are potentially cheaper, but perhaps
2294 * more importantly, the bigger file is likely the more recent
2295 * one. The deepest deltas are therefore the oldest objects which are
2296 * less susceptible to be accessed often.
2298 static int type_size_sort(const void *_a, const void *_b)
2300 const struct object_entry *a = *(struct object_entry **)_a;
2301 const struct object_entry *b = *(struct object_entry **)_b;
2302 const enum object_type a_type = oe_type(a);
2303 const enum object_type b_type = oe_type(b);
2304 const unsigned long a_size = SIZE(a);
2305 const unsigned long b_size = SIZE(b);
2307 if (a_type > b_type)
2308 return -1;
2309 if (a_type < b_type)
2310 return 1;
2311 if (a->hash > b->hash)
2312 return -1;
2313 if (a->hash < b->hash)
2314 return 1;
2315 if (a->preferred_base > b->preferred_base)
2316 return -1;
2317 if (a->preferred_base < b->preferred_base)
2318 return 1;
2319 if (use_delta_islands) {
2320 const int island_cmp = island_delta_cmp(&a->idx.oid, &b->idx.oid);
2321 if (island_cmp)
2322 return island_cmp;
2324 if (a_size > b_size)
2325 return -1;
2326 if (a_size < b_size)
2327 return 1;
2328 return a < b ? -1 : (a > b); /* newest first */
2331 struct unpacked {
2332 struct object_entry *entry;
2333 void *data;
2334 struct delta_index *index;
2335 unsigned depth;
2338 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
2339 unsigned long delta_size)
2341 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
2342 return 0;
2344 if (delta_size < cache_max_small_delta_size)
2345 return 1;
2347 /* cache delta, if objects are large enough compared to delta size */
2348 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
2349 return 1;
2351 return 0;
2354 /* Protect delta_cache_size */
2355 static pthread_mutex_t cache_mutex;
2356 #define cache_lock() pthread_mutex_lock(&cache_mutex)
2357 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
2360 * Protect object list partitioning (e.g. struct thread_param) and
2361 * progress_state
2363 static pthread_mutex_t progress_mutex;
2364 #define progress_lock() pthread_mutex_lock(&progress_mutex)
2365 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
2368 * Access to struct object_entry is unprotected since each thread owns
2369 * a portion of the main object list. Just don't access object entries
2370 * ahead in the list because they can be stolen and would need
2371 * progress_mutex for protection.
2374 static inline int oe_size_less_than(struct packing_data *pack,
2375 const struct object_entry *lhs,
2376 unsigned long rhs)
2378 if (lhs->size_valid)
2379 return lhs->size_ < rhs;
2380 if (rhs < pack->oe_size_limit) /* rhs < 2^x <= lhs ? */
2381 return 0;
2382 return oe_get_size_slow(pack, lhs) < rhs;
2385 static inline void oe_set_tree_depth(struct packing_data *pack,
2386 struct object_entry *e,
2387 unsigned int tree_depth)
2389 if (!pack->tree_depth)
2390 CALLOC_ARRAY(pack->tree_depth, pack->nr_alloc);
2391 pack->tree_depth[e - pack->objects] = tree_depth;
2395 * Return the size of the object without doing any delta
2396 * reconstruction (so non-deltas are true object sizes, but deltas
2397 * return the size of the delta data).
2399 unsigned long oe_get_size_slow(struct packing_data *pack,
2400 const struct object_entry *e)
2402 struct packed_git *p;
2403 struct pack_window *w_curs;
2404 unsigned char *buf;
2405 enum object_type type;
2406 unsigned long used, avail, size;
2408 if (e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {
2409 packing_data_lock(&to_pack);
2410 if (oid_object_info(the_repository, &e->idx.oid, &size) < 0)
2411 die(_("unable to get size of %s"),
2412 oid_to_hex(&e->idx.oid));
2413 packing_data_unlock(&to_pack);
2414 return size;
2417 p = oe_in_pack(pack, e);
2418 if (!p)
2419 BUG("when e->type is a delta, it must belong to a pack");
2421 packing_data_lock(&to_pack);
2422 w_curs = NULL;
2423 buf = use_pack(p, &w_curs, e->in_pack_offset, &avail);
2424 used = unpack_object_header_buffer(buf, avail, &type, &size);
2425 if (used == 0)
2426 die(_("unable to parse object header of %s"),
2427 oid_to_hex(&e->idx.oid));
2429 unuse_pack(&w_curs);
2430 packing_data_unlock(&to_pack);
2431 return size;
2434 static int try_delta(struct unpacked *trg, struct unpacked *src,
2435 unsigned max_depth, unsigned long *mem_usage)
2437 struct object_entry *trg_entry = trg->entry;
2438 struct object_entry *src_entry = src->entry;
2439 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
2440 unsigned ref_depth;
2441 enum object_type type;
2442 void *delta_buf;
2444 /* Don't bother doing diffs between different types */
2445 if (oe_type(trg_entry) != oe_type(src_entry))
2446 return -1;
2449 * We do not bother to try a delta that we discarded on an
2450 * earlier try, but only when reusing delta data. Note that
2451 * src_entry that is marked as the preferred_base should always
2452 * be considered, as even if we produce a suboptimal delta against
2453 * it, we will still save the transfer cost, as we already know
2454 * the other side has it and we won't send src_entry at all.
2456 if (reuse_delta && IN_PACK(trg_entry) &&
2457 IN_PACK(trg_entry) == IN_PACK(src_entry) &&
2458 !src_entry->preferred_base &&
2459 trg_entry->in_pack_type != OBJ_REF_DELTA &&
2460 trg_entry->in_pack_type != OBJ_OFS_DELTA)
2461 return 0;
2463 /* Let's not bust the allowed depth. */
2464 if (src->depth >= max_depth)
2465 return 0;
2467 /* Now some size filtering heuristics. */
2468 trg_size = SIZE(trg_entry);
2469 if (!DELTA(trg_entry)) {
2470 max_size = trg_size/2 - the_hash_algo->rawsz;
2471 ref_depth = 1;
2472 } else {
2473 max_size = DELTA_SIZE(trg_entry);
2474 ref_depth = trg->depth;
2476 max_size = (uint64_t)max_size * (max_depth - src->depth) /
2477 (max_depth - ref_depth + 1);
2478 if (max_size == 0)
2479 return 0;
2480 src_size = SIZE(src_entry);
2481 sizediff = src_size < trg_size ? trg_size - src_size : 0;
2482 if (sizediff >= max_size)
2483 return 0;
2484 if (trg_size < src_size / 32)
2485 return 0;
2487 if (!in_same_island(&trg->entry->idx.oid, &src->entry->idx.oid))
2488 return 0;
2490 /* Load data if not already done */
2491 if (!trg->data) {
2492 packing_data_lock(&to_pack);
2493 trg->data = read_object_file(&trg_entry->idx.oid, &type, &sz);
2494 packing_data_unlock(&to_pack);
2495 if (!trg->data)
2496 die(_("object %s cannot be read"),
2497 oid_to_hex(&trg_entry->idx.oid));
2498 if (sz != trg_size)
2499 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),
2500 oid_to_hex(&trg_entry->idx.oid), (uintmax_t)sz,
2501 (uintmax_t)trg_size);
2502 *mem_usage += sz;
2504 if (!src->data) {
2505 packing_data_lock(&to_pack);
2506 src->data = read_object_file(&src_entry->idx.oid, &type, &sz);
2507 packing_data_unlock(&to_pack);
2508 if (!src->data) {
2509 if (src_entry->preferred_base) {
2510 static int warned = 0;
2511 if (!warned++)
2512 warning(_("object %s cannot be read"),
2513 oid_to_hex(&src_entry->idx.oid));
2515 * Those objects are not included in the
2516 * resulting pack. Be resilient and ignore
2517 * them if they can't be read, in case the
2518 * pack could be created nevertheless.
2520 return 0;
2522 die(_("object %s cannot be read"),
2523 oid_to_hex(&src_entry->idx.oid));
2525 if (sz != src_size)
2526 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),
2527 oid_to_hex(&src_entry->idx.oid), (uintmax_t)sz,
2528 (uintmax_t)src_size);
2529 *mem_usage += sz;
2531 if (!src->index) {
2532 src->index = create_delta_index(src->data, src_size);
2533 if (!src->index) {
2534 static int warned = 0;
2535 if (!warned++)
2536 warning(_("suboptimal pack - out of memory"));
2537 return 0;
2539 *mem_usage += sizeof_delta_index(src->index);
2542 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
2543 if (!delta_buf)
2544 return 0;
2546 if (DELTA(trg_entry)) {
2547 /* Prefer only shallower same-sized deltas. */
2548 if (delta_size == DELTA_SIZE(trg_entry) &&
2549 src->depth + 1 >= trg->depth) {
2550 free(delta_buf);
2551 return 0;
2556 * Handle memory allocation outside of the cache
2557 * accounting lock. Compiler will optimize the strangeness
2558 * away when NO_PTHREADS is defined.
2560 free(trg_entry->delta_data);
2561 cache_lock();
2562 if (trg_entry->delta_data) {
2563 delta_cache_size -= DELTA_SIZE(trg_entry);
2564 trg_entry->delta_data = NULL;
2566 if (delta_cacheable(src_size, trg_size, delta_size)) {
2567 delta_cache_size += delta_size;
2568 cache_unlock();
2569 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
2570 } else {
2571 cache_unlock();
2572 free(delta_buf);
2575 SET_DELTA(trg_entry, src_entry);
2576 SET_DELTA_SIZE(trg_entry, delta_size);
2577 trg->depth = src->depth + 1;
2579 return 1;
2582 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
2584 struct object_entry *child = DELTA_CHILD(me);
2585 unsigned int m = n;
2586 while (child) {
2587 const unsigned int c = check_delta_limit(child, n + 1);
2588 if (m < c)
2589 m = c;
2590 child = DELTA_SIBLING(child);
2592 return m;
2595 static unsigned long free_unpacked(struct unpacked *n)
2597 unsigned long freed_mem = sizeof_delta_index(n->index);
2598 free_delta_index(n->index);
2599 n->index = NULL;
2600 if (n->data) {
2601 freed_mem += SIZE(n->entry);
2602 FREE_AND_NULL(n->data);
2604 n->entry = NULL;
2605 n->depth = 0;
2606 return freed_mem;
2609 static void find_deltas(struct object_entry **list, unsigned *list_size,
2610 int window, int depth, unsigned *processed)
2612 uint32_t i, idx = 0, count = 0;
2613 struct unpacked *array;
2614 unsigned long mem_usage = 0;
2616 CALLOC_ARRAY(array, window);
2618 for (;;) {
2619 struct object_entry *entry;
2620 struct unpacked *n = array + idx;
2621 int j, max_depth, best_base = -1;
2623 progress_lock();
2624 if (!*list_size) {
2625 progress_unlock();
2626 break;
2628 entry = *list++;
2629 (*list_size)--;
2630 if (!entry->preferred_base) {
2631 (*processed)++;
2632 display_progress(progress_state, *processed);
2634 progress_unlock();
2636 mem_usage -= free_unpacked(n);
2637 n->entry = entry;
2639 while (window_memory_limit &&
2640 mem_usage > window_memory_limit &&
2641 count > 1) {
2642 const uint32_t tail = (idx + window - count) % window;
2643 mem_usage -= free_unpacked(array + tail);
2644 count--;
2647 /* We do not compute delta to *create* objects we are not
2648 * going to pack.
2650 if (entry->preferred_base)
2651 goto next;
2654 * If the current object is at pack edge, take the depth the
2655 * objects that depend on the current object into account
2656 * otherwise they would become too deep.
2658 max_depth = depth;
2659 if (DELTA_CHILD(entry)) {
2660 max_depth -= check_delta_limit(entry, 0);
2661 if (max_depth <= 0)
2662 goto next;
2665 j = window;
2666 while (--j > 0) {
2667 int ret;
2668 uint32_t other_idx = idx + j;
2669 struct unpacked *m;
2670 if (other_idx >= window)
2671 other_idx -= window;
2672 m = array + other_idx;
2673 if (!m->entry)
2674 break;
2675 ret = try_delta(n, m, max_depth, &mem_usage);
2676 if (ret < 0)
2677 break;
2678 else if (ret > 0)
2679 best_base = other_idx;
2683 * If we decided to cache the delta data, then it is best
2684 * to compress it right away. First because we have to do
2685 * it anyway, and doing it here while we're threaded will
2686 * save a lot of time in the non threaded write phase,
2687 * as well as allow for caching more deltas within
2688 * the same cache size limit.
2689 * ...
2690 * But only if not writing to stdout, since in that case
2691 * the network is most likely throttling writes anyway,
2692 * and therefore it is best to go to the write phase ASAP
2693 * instead, as we can afford spending more time compressing
2694 * between writes at that moment.
2696 if (entry->delta_data && !pack_to_stdout) {
2697 unsigned long size;
2699 size = do_compress(&entry->delta_data, DELTA_SIZE(entry));
2700 if (size < (1U << OE_Z_DELTA_BITS)) {
2701 entry->z_delta_size = size;
2702 cache_lock();
2703 delta_cache_size -= DELTA_SIZE(entry);
2704 delta_cache_size += entry->z_delta_size;
2705 cache_unlock();
2706 } else {
2707 FREE_AND_NULL(entry->delta_data);
2708 entry->z_delta_size = 0;
2712 /* if we made n a delta, and if n is already at max
2713 * depth, leaving it in the window is pointless. we
2714 * should evict it first.
2716 if (DELTA(entry) && max_depth <= n->depth)
2717 continue;
2720 * Move the best delta base up in the window, after the
2721 * currently deltified object, to keep it longer. It will
2722 * be the first base object to be attempted next.
2724 if (DELTA(entry)) {
2725 struct unpacked swap = array[best_base];
2726 int dist = (window + idx - best_base) % window;
2727 int dst = best_base;
2728 while (dist--) {
2729 int src = (dst + 1) % window;
2730 array[dst] = array[src];
2731 dst = src;
2733 array[dst] = swap;
2736 next:
2737 idx++;
2738 if (count + 1 < window)
2739 count++;
2740 if (idx >= window)
2741 idx = 0;
2744 for (i = 0; i < window; ++i) {
2745 free_delta_index(array[i].index);
2746 free(array[i].data);
2748 free(array);
2752 * The main object list is split into smaller lists, each is handed to
2753 * one worker.
2755 * The main thread waits on the condition that (at least) one of the workers
2756 * has stopped working (which is indicated in the .working member of
2757 * struct thread_params).
2759 * When a work thread has completed its work, it sets .working to 0 and
2760 * signals the main thread and waits on the condition that .data_ready
2761 * becomes 1.
2763 * The main thread steals half of the work from the worker that has
2764 * most work left to hand it to the idle worker.
2767 struct thread_params {
2768 pthread_t thread;
2769 struct object_entry **list;
2770 unsigned list_size;
2771 unsigned remaining;
2772 int window;
2773 int depth;
2774 int working;
2775 int data_ready;
2776 pthread_mutex_t mutex;
2777 pthread_cond_t cond;
2778 unsigned *processed;
2781 static pthread_cond_t progress_cond;
2784 * Mutex and conditional variable can't be statically-initialized on Windows.
2786 static void init_threaded_search(void)
2788 pthread_mutex_init(&cache_mutex, NULL);
2789 pthread_mutex_init(&progress_mutex, NULL);
2790 pthread_cond_init(&progress_cond, NULL);
2793 static void cleanup_threaded_search(void)
2795 pthread_cond_destroy(&progress_cond);
2796 pthread_mutex_destroy(&cache_mutex);
2797 pthread_mutex_destroy(&progress_mutex);
2800 static void *threaded_find_deltas(void *arg)
2802 struct thread_params *me = arg;
2804 progress_lock();
2805 while (me->remaining) {
2806 progress_unlock();
2808 find_deltas(me->list, &me->remaining,
2809 me->window, me->depth, me->processed);
2811 progress_lock();
2812 me->working = 0;
2813 pthread_cond_signal(&progress_cond);
2814 progress_unlock();
2817 * We must not set ->data_ready before we wait on the
2818 * condition because the main thread may have set it to 1
2819 * before we get here. In order to be sure that new
2820 * work is available if we see 1 in ->data_ready, it
2821 * was initialized to 0 before this thread was spawned
2822 * and we reset it to 0 right away.
2824 pthread_mutex_lock(&me->mutex);
2825 while (!me->data_ready)
2826 pthread_cond_wait(&me->cond, &me->mutex);
2827 me->data_ready = 0;
2828 pthread_mutex_unlock(&me->mutex);
2830 progress_lock();
2832 progress_unlock();
2833 /* leave ->working 1 so that this doesn't get more work assigned */
2834 return NULL;
2837 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2838 int window, int depth, unsigned *processed)
2840 struct thread_params *p;
2841 int i, ret, active_threads = 0;
2843 init_threaded_search();
2845 if (delta_search_threads <= 1) {
2846 find_deltas(list, &list_size, window, depth, processed);
2847 cleanup_threaded_search();
2848 return;
2850 if (progress > pack_to_stdout)
2851 fprintf_ln(stderr, _("Delta compression using up to %d threads"),
2852 delta_search_threads);
2853 CALLOC_ARRAY(p, delta_search_threads);
2855 /* Partition the work amongst work threads. */
2856 for (i = 0; i < delta_search_threads; i++) {
2857 unsigned sub_size = list_size / (delta_search_threads - i);
2859 /* don't use too small segments or no deltas will be found */
2860 if (sub_size < 2*window && i+1 < delta_search_threads)
2861 sub_size = 0;
2863 p[i].window = window;
2864 p[i].depth = depth;
2865 p[i].processed = processed;
2866 p[i].working = 1;
2867 p[i].data_ready = 0;
2869 /* try to split chunks on "path" boundaries */
2870 while (sub_size && sub_size < list_size &&
2871 list[sub_size]->hash &&
2872 list[sub_size]->hash == list[sub_size-1]->hash)
2873 sub_size++;
2875 p[i].list = list;
2876 p[i].list_size = sub_size;
2877 p[i].remaining = sub_size;
2879 list += sub_size;
2880 list_size -= sub_size;
2883 /* Start work threads. */
2884 for (i = 0; i < delta_search_threads; i++) {
2885 if (!p[i].list_size)
2886 continue;
2887 pthread_mutex_init(&p[i].mutex, NULL);
2888 pthread_cond_init(&p[i].cond, NULL);
2889 ret = pthread_create(&p[i].thread, NULL,
2890 threaded_find_deltas, &p[i]);
2891 if (ret)
2892 die(_("unable to create thread: %s"), strerror(ret));
2893 active_threads++;
2897 * Now let's wait for work completion. Each time a thread is done
2898 * with its work, we steal half of the remaining work from the
2899 * thread with the largest number of unprocessed objects and give
2900 * it to that newly idle thread. This ensure good load balancing
2901 * until the remaining object list segments are simply too short
2902 * to be worth splitting anymore.
2904 while (active_threads) {
2905 struct thread_params *target = NULL;
2906 struct thread_params *victim = NULL;
2907 unsigned sub_size = 0;
2909 progress_lock();
2910 for (;;) {
2911 for (i = 0; !target && i < delta_search_threads; i++)
2912 if (!p[i].working)
2913 target = &p[i];
2914 if (target)
2915 break;
2916 pthread_cond_wait(&progress_cond, &progress_mutex);
2919 for (i = 0; i < delta_search_threads; i++)
2920 if (p[i].remaining > 2*window &&
2921 (!victim || victim->remaining < p[i].remaining))
2922 victim = &p[i];
2923 if (victim) {
2924 sub_size = victim->remaining / 2;
2925 list = victim->list + victim->list_size - sub_size;
2926 while (sub_size && list[0]->hash &&
2927 list[0]->hash == list[-1]->hash) {
2928 list++;
2929 sub_size--;
2931 if (!sub_size) {
2933 * It is possible for some "paths" to have
2934 * so many objects that no hash boundary
2935 * might be found. Let's just steal the
2936 * exact half in that case.
2938 sub_size = victim->remaining / 2;
2939 list -= sub_size;
2941 target->list = list;
2942 victim->list_size -= sub_size;
2943 victim->remaining -= sub_size;
2945 target->list_size = sub_size;
2946 target->remaining = sub_size;
2947 target->working = 1;
2948 progress_unlock();
2950 pthread_mutex_lock(&target->mutex);
2951 target->data_ready = 1;
2952 pthread_cond_signal(&target->cond);
2953 pthread_mutex_unlock(&target->mutex);
2955 if (!sub_size) {
2956 pthread_join(target->thread, NULL);
2957 pthread_cond_destroy(&target->cond);
2958 pthread_mutex_destroy(&target->mutex);
2959 active_threads--;
2962 cleanup_threaded_search();
2963 free(p);
2966 static int obj_is_packed(const struct object_id *oid)
2968 return packlist_find(&to_pack, oid) ||
2969 (reuse_packfile_bitmap &&
2970 bitmap_walk_contains(bitmap_git, reuse_packfile_bitmap, oid));
2973 static void add_tag_chain(const struct object_id *oid)
2975 struct tag *tag;
2978 * We catch duplicates already in add_object_entry(), but we'd
2979 * prefer to do this extra check to avoid having to parse the
2980 * tag at all if we already know that it's being packed (e.g., if
2981 * it was included via bitmaps, we would not have parsed it
2982 * previously).
2984 if (obj_is_packed(oid))
2985 return;
2987 tag = lookup_tag(the_repository, oid);
2988 while (1) {
2989 if (!tag || parse_tag(tag) || !tag->tagged)
2990 die(_("unable to pack objects reachable from tag %s"),
2991 oid_to_hex(oid));
2993 add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);
2995 if (tag->tagged->type != OBJ_TAG)
2996 return;
2998 tag = (struct tag *)tag->tagged;
3002 static int add_ref_tag(const char *tag, const struct object_id *oid, int flag, void *cb_data)
3004 struct object_id peeled;
3006 if (!peel_iterated_oid(oid, &peeled) && obj_is_packed(&peeled))
3007 add_tag_chain(oid);
3008 return 0;
3011 static void prepare_pack(int window, int depth)
3013 struct object_entry **delta_list;
3014 uint32_t i, nr_deltas;
3015 unsigned n;
3017 if (use_delta_islands)
3018 resolve_tree_islands(the_repository, progress, &to_pack);
3020 get_object_details();
3023 * If we're locally repacking then we need to be doubly careful
3024 * from now on in order to make sure no stealth corruption gets
3025 * propagated to the new pack. Clients receiving streamed packs
3026 * should validate everything they get anyway so no need to incur
3027 * the additional cost here in that case.
3029 if (!pack_to_stdout)
3030 do_check_packed_object_crc = 1;
3032 if (!to_pack.nr_objects || !window || !depth)
3033 return;
3035 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
3036 nr_deltas = n = 0;
3038 for (i = 0; i < to_pack.nr_objects; i++) {
3039 struct object_entry *entry = to_pack.objects + i;
3041 if (DELTA(entry))
3042 /* This happens if we decided to reuse existing
3043 * delta from a pack. "reuse_delta &&" is implied.
3045 continue;
3047 if (!entry->type_valid ||
3048 oe_size_less_than(&to_pack, entry, 50))
3049 continue;
3051 if (entry->no_try_delta)
3052 continue;
3054 if (!entry->preferred_base) {
3055 nr_deltas++;
3056 if (oe_type(entry) < 0)
3057 die(_("unable to get type of object %s"),
3058 oid_to_hex(&entry->idx.oid));
3059 } else {
3060 if (oe_type(entry) < 0) {
3062 * This object is not found, but we
3063 * don't have to include it anyway.
3065 continue;
3069 delta_list[n++] = entry;
3072 if (nr_deltas && n > 1) {
3073 unsigned nr_done = 0;
3075 if (progress)
3076 progress_state = start_progress(_("Compressing objects"),
3077 nr_deltas);
3078 QSORT(delta_list, n, type_size_sort);
3079 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
3080 stop_progress(&progress_state);
3081 if (nr_done != nr_deltas)
3082 die(_("inconsistency with delta count"));
3084 free(delta_list);
3087 static int git_pack_config(const char *k, const char *v, void *cb)
3089 if (!strcmp(k, "pack.window")) {
3090 window = git_config_int(k, v);
3091 return 0;
3093 if (!strcmp(k, "pack.windowmemory")) {
3094 window_memory_limit = git_config_ulong(k, v);
3095 return 0;
3097 if (!strcmp(k, "pack.depth")) {
3098 depth = git_config_int(k, v);
3099 return 0;
3101 if (!strcmp(k, "pack.deltacachesize")) {
3102 max_delta_cache_size = git_config_int(k, v);
3103 return 0;
3105 if (!strcmp(k, "pack.deltacachelimit")) {
3106 cache_max_small_delta_size = git_config_int(k, v);
3107 return 0;
3109 if (!strcmp(k, "pack.writebitmaphashcache")) {
3110 if (git_config_bool(k, v))
3111 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
3112 else
3113 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
3115 if (!strcmp(k, "pack.usebitmaps")) {
3116 use_bitmap_index_default = git_config_bool(k, v);
3117 return 0;
3119 if (!strcmp(k, "pack.allowpackreuse")) {
3120 allow_pack_reuse = git_config_bool(k, v);
3121 return 0;
3123 if (!strcmp(k, "pack.threads")) {
3124 delta_search_threads = git_config_int(k, v);
3125 if (delta_search_threads < 0)
3126 die(_("invalid number of threads specified (%d)"),
3127 delta_search_threads);
3128 if (!HAVE_THREADS && delta_search_threads != 1) {
3129 warning(_("no threads support, ignoring %s"), k);
3130 delta_search_threads = 0;
3132 return 0;
3134 if (!strcmp(k, "pack.indexversion")) {
3135 pack_idx_opts.version = git_config_int(k, v);
3136 if (pack_idx_opts.version > 2)
3137 die(_("bad pack.indexversion=%"PRIu32),
3138 pack_idx_opts.version);
3139 return 0;
3141 if (!strcmp(k, "pack.writereverseindex")) {
3142 if (git_config_bool(k, v))
3143 pack_idx_opts.flags |= WRITE_REV;
3144 else
3145 pack_idx_opts.flags &= ~WRITE_REV;
3146 return 0;
3148 if (!strcmp(k, "uploadpack.blobpackfileuri")) {
3149 struct configured_exclusion *ex = xmalloc(sizeof(*ex));
3150 const char *oid_end, *pack_end;
3152 * Stores the pack hash. This is not a true object ID, but is
3153 * of the same form.
3155 struct object_id pack_hash;
3157 if (parse_oid_hex(v, &ex->e.oid, &oid_end) ||
3158 *oid_end != ' ' ||
3159 parse_oid_hex(oid_end + 1, &pack_hash, &pack_end) ||
3160 *pack_end != ' ')
3161 die(_("value of uploadpack.blobpackfileuri must be "
3162 "of the form '<object-hash> <pack-hash> <uri>' (got '%s')"), v);
3163 if (oidmap_get(&configured_exclusions, &ex->e.oid))
3164 die(_("object already configured in another "
3165 "uploadpack.blobpackfileuri (got '%s')"), v);
3166 ex->pack_hash_hex = xcalloc(1, pack_end - oid_end);
3167 memcpy(ex->pack_hash_hex, oid_end + 1, pack_end - oid_end - 1);
3168 ex->uri = xstrdup(pack_end + 1);
3169 oidmap_put(&configured_exclusions, ex);
3171 return git_default_config(k, v, cb);
3174 /* Counters for trace2 output when in --stdin-packs mode. */
3175 static int stdin_packs_found_nr;
3176 static int stdin_packs_hints_nr;
3178 static int add_object_entry_from_pack(const struct object_id *oid,
3179 struct packed_git *p,
3180 uint32_t pos,
3181 void *_data)
3183 struct rev_info *revs = _data;
3184 struct object_info oi = OBJECT_INFO_INIT;
3185 off_t ofs;
3186 enum object_type type;
3188 display_progress(progress_state, ++nr_seen);
3190 if (have_duplicate_entry(oid, 0))
3191 return 0;
3193 ofs = nth_packed_object_offset(p, pos);
3194 if (!want_object_in_pack(oid, 0, &p, &ofs))
3195 return 0;
3197 oi.typep = &type;
3198 if (packed_object_info(the_repository, p, ofs, &oi) < 0)
3199 die(_("could not get type of object %s in pack %s"),
3200 oid_to_hex(oid), p->pack_name);
3201 else if (type == OBJ_COMMIT) {
3203 * commits in included packs are used as starting points for the
3204 * subsequent revision walk
3206 add_pending_oid(revs, NULL, oid, 0);
3209 stdin_packs_found_nr++;
3211 create_object_entry(oid, type, 0, 0, 0, p, ofs);
3213 return 0;
3216 static void show_commit_pack_hint(struct commit *commit, void *_data)
3218 /* nothing to do; commits don't have a namehash */
3221 static void show_object_pack_hint(struct object *object, const char *name,
3222 void *_data)
3224 struct object_entry *oe = packlist_find(&to_pack, &object->oid);
3225 if (!oe)
3226 return;
3229 * Our 'to_pack' list was constructed by iterating all objects packed in
3230 * included packs, and so doesn't have a non-zero hash field that you
3231 * would typically pick up during a reachability traversal.
3233 * Make a best-effort attempt to fill in the ->hash and ->no_try_delta
3234 * here using a now in order to perhaps improve the delta selection
3235 * process.
3237 oe->hash = pack_name_hash(name);
3238 oe->no_try_delta = name && no_try_delta(name);
3240 stdin_packs_hints_nr++;
3243 static int pack_mtime_cmp(const void *_a, const void *_b)
3245 struct packed_git *a = ((const struct string_list_item*)_a)->util;
3246 struct packed_git *b = ((const struct string_list_item*)_b)->util;
3249 * order packs by descending mtime so that objects are laid out
3250 * roughly as newest-to-oldest
3252 if (a->mtime < b->mtime)
3253 return 1;
3254 else if (b->mtime < a->mtime)
3255 return -1;
3256 else
3257 return 0;
3260 static void read_packs_list_from_stdin(void)
3262 struct strbuf buf = STRBUF_INIT;
3263 struct string_list include_packs = STRING_LIST_INIT_DUP;
3264 struct string_list exclude_packs = STRING_LIST_INIT_DUP;
3265 struct string_list_item *item = NULL;
3267 struct packed_git *p;
3268 struct rev_info revs;
3270 repo_init_revisions(the_repository, &revs, NULL);
3272 * Use a revision walk to fill in the namehash of objects in the include
3273 * packs. To save time, we'll avoid traversing through objects that are
3274 * in excluded packs.
3276 * That may cause us to avoid populating all of the namehash fields of
3277 * all included objects, but our goal is best-effort, since this is only
3278 * an optimization during delta selection.
3280 revs.no_kept_objects = 1;
3281 revs.keep_pack_cache_flags |= IN_CORE_KEEP_PACKS;
3282 revs.blob_objects = 1;
3283 revs.tree_objects = 1;
3284 revs.tag_objects = 1;
3285 revs.ignore_missing_links = 1;
3287 while (strbuf_getline(&buf, stdin) != EOF) {
3288 if (!buf.len)
3289 continue;
3291 if (*buf.buf == '^')
3292 string_list_append(&exclude_packs, buf.buf + 1);
3293 else
3294 string_list_append(&include_packs, buf.buf);
3296 strbuf_reset(&buf);
3299 string_list_sort(&include_packs);
3300 string_list_sort(&exclude_packs);
3302 for (p = get_all_packs(the_repository); p; p = p->next) {
3303 const char *pack_name = pack_basename(p);
3305 item = string_list_lookup(&include_packs, pack_name);
3306 if (!item)
3307 item = string_list_lookup(&exclude_packs, pack_name);
3309 if (item)
3310 item->util = p;
3314 * First handle all of the excluded packs, marking them as kept in-core
3315 * so that later calls to add_object_entry() discards any objects that
3316 * are also found in excluded packs.
3318 for_each_string_list_item(item, &exclude_packs) {
3319 struct packed_git *p = item->util;
3320 if (!p)
3321 die(_("could not find pack '%s'"), item->string);
3322 p->pack_keep_in_core = 1;
3326 * Order packs by ascending mtime; use QSORT directly to access the
3327 * string_list_item's ->util pointer, which string_list_sort() does not
3328 * provide.
3330 QSORT(include_packs.items, include_packs.nr, pack_mtime_cmp);
3332 for_each_string_list_item(item, &include_packs) {
3333 struct packed_git *p = item->util;
3334 if (!p)
3335 die(_("could not find pack '%s'"), item->string);
3336 for_each_object_in_pack(p,
3337 add_object_entry_from_pack,
3338 &revs,
3339 FOR_EACH_OBJECT_PACK_ORDER);
3342 if (prepare_revision_walk(&revs))
3343 die(_("revision walk setup failed"));
3344 traverse_commit_list(&revs,
3345 show_commit_pack_hint,
3346 show_object_pack_hint,
3347 NULL);
3349 trace2_data_intmax("pack-objects", the_repository, "stdin_packs_found",
3350 stdin_packs_found_nr);
3351 trace2_data_intmax("pack-objects", the_repository, "stdin_packs_hints",
3352 stdin_packs_hints_nr);
3354 strbuf_release(&buf);
3355 string_list_clear(&include_packs, 0);
3356 string_list_clear(&exclude_packs, 0);
3359 static void read_object_list_from_stdin(void)
3361 char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];
3362 struct object_id oid;
3363 const char *p;
3365 for (;;) {
3366 if (!fgets(line, sizeof(line), stdin)) {
3367 if (feof(stdin))
3368 break;
3369 if (!ferror(stdin))
3370 die("BUG: fgets returned NULL, not EOF, not error!");
3371 if (errno != EINTR)
3372 die_errno("fgets");
3373 clearerr(stdin);
3374 continue;
3376 if (line[0] == '-') {
3377 if (get_oid_hex(line+1, &oid))
3378 die(_("expected edge object ID, got garbage:\n %s"),
3379 line);
3380 add_preferred_base(&oid);
3381 continue;
3383 if (parse_oid_hex(line, &oid, &p))
3384 die(_("expected object ID, got garbage:\n %s"), line);
3386 add_preferred_base_object(p + 1);
3387 add_object_entry(&oid, OBJ_NONE, p + 1, 0);
3391 /* Remember to update object flag allocation in object.h */
3392 #define OBJECT_ADDED (1u<<20)
3394 static void show_commit(struct commit *commit, void *data)
3396 add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);
3397 commit->object.flags |= OBJECT_ADDED;
3399 if (write_bitmap_index)
3400 index_commit_for_bitmap(commit);
3402 if (use_delta_islands)
3403 propagate_island_marks(commit);
3406 static void show_object(struct object *obj, const char *name, void *data)
3408 add_preferred_base_object(name);
3409 add_object_entry(&obj->oid, obj->type, name, 0);
3410 obj->flags |= OBJECT_ADDED;
3412 if (use_delta_islands) {
3413 const char *p;
3414 unsigned depth;
3415 struct object_entry *ent;
3417 /* the empty string is a root tree, which is depth 0 */
3418 depth = *name ? 1 : 0;
3419 for (p = strchr(name, '/'); p; p = strchr(p + 1, '/'))
3420 depth++;
3422 ent = packlist_find(&to_pack, &obj->oid);
3423 if (ent && depth > oe_tree_depth(&to_pack, ent))
3424 oe_set_tree_depth(&to_pack, ent, depth);
3428 static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)
3430 assert(arg_missing_action == MA_ALLOW_ANY);
3433 * Quietly ignore ALL missing objects. This avoids problems with
3434 * staging them now and getting an odd error later.
3436 if (!has_object(the_repository, &obj->oid, 0))
3437 return;
3439 show_object(obj, name, data);
3442 static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)
3444 assert(arg_missing_action == MA_ALLOW_PROMISOR);
3447 * Quietly ignore EXPECTED missing objects. This avoids problems with
3448 * staging them now and getting an odd error later.
3450 if (!has_object(the_repository, &obj->oid, 0) && is_promisor_object(&obj->oid))
3451 return;
3453 show_object(obj, name, data);
3456 static int option_parse_missing_action(const struct option *opt,
3457 const char *arg, int unset)
3459 assert(arg);
3460 assert(!unset);
3462 if (!strcmp(arg, "error")) {
3463 arg_missing_action = MA_ERROR;
3464 fn_show_object = show_object;
3465 return 0;
3468 if (!strcmp(arg, "allow-any")) {
3469 arg_missing_action = MA_ALLOW_ANY;
3470 fetch_if_missing = 0;
3471 fn_show_object = show_object__ma_allow_any;
3472 return 0;
3475 if (!strcmp(arg, "allow-promisor")) {
3476 arg_missing_action = MA_ALLOW_PROMISOR;
3477 fetch_if_missing = 0;
3478 fn_show_object = show_object__ma_allow_promisor;
3479 return 0;
3482 die(_("invalid value for --missing"));
3483 return 0;
3486 static void show_edge(struct commit *commit)
3488 add_preferred_base(&commit->object.oid);
3491 struct in_pack_object {
3492 off_t offset;
3493 struct object *object;
3496 struct in_pack {
3497 unsigned int alloc;
3498 unsigned int nr;
3499 struct in_pack_object *array;
3502 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
3504 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
3505 in_pack->array[in_pack->nr].object = object;
3506 in_pack->nr++;
3510 * Compare the objects in the offset order, in order to emulate the
3511 * "git rev-list --objects" output that produced the pack originally.
3513 static int ofscmp(const void *a_, const void *b_)
3515 struct in_pack_object *a = (struct in_pack_object *)a_;
3516 struct in_pack_object *b = (struct in_pack_object *)b_;
3518 if (a->offset < b->offset)
3519 return -1;
3520 else if (a->offset > b->offset)
3521 return 1;
3522 else
3523 return oidcmp(&a->object->oid, &b->object->oid);
3526 static void add_objects_in_unpacked_packs(void)
3528 struct packed_git *p;
3529 struct in_pack in_pack;
3530 uint32_t i;
3532 memset(&in_pack, 0, sizeof(in_pack));
3534 for (p = get_all_packs(the_repository); p; p = p->next) {
3535 struct object_id oid;
3536 struct object *o;
3538 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)
3539 continue;
3540 if (open_pack_index(p))
3541 die(_("cannot open pack index"));
3543 ALLOC_GROW(in_pack.array,
3544 in_pack.nr + p->num_objects,
3545 in_pack.alloc);
3547 for (i = 0; i < p->num_objects; i++) {
3548 nth_packed_object_id(&oid, p, i);
3549 o = lookup_unknown_object(the_repository, &oid);
3550 if (!(o->flags & OBJECT_ADDED))
3551 mark_in_pack_object(o, p, &in_pack);
3552 o->flags |= OBJECT_ADDED;
3556 if (in_pack.nr) {
3557 QSORT(in_pack.array, in_pack.nr, ofscmp);
3558 for (i = 0; i < in_pack.nr; i++) {
3559 struct object *o = in_pack.array[i].object;
3560 add_object_entry(&o->oid, o->type, "", 0);
3563 free(in_pack.array);
3566 static int add_loose_object(const struct object_id *oid, const char *path,
3567 void *data)
3569 enum object_type type = oid_object_info(the_repository, oid, NULL);
3571 if (type < 0) {
3572 warning(_("loose object at %s could not be examined"), path);
3573 return 0;
3576 add_object_entry(oid, type, "", 0);
3577 return 0;
3581 * We actually don't even have to worry about reachability here.
3582 * add_object_entry will weed out duplicates, so we just add every
3583 * loose object we find.
3585 static void add_unreachable_loose_objects(void)
3587 for_each_loose_file_in_objdir(get_object_directory(),
3588 add_loose_object,
3589 NULL, NULL, NULL);
3592 static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)
3594 static struct packed_git *last_found = (void *)1;
3595 struct packed_git *p;
3597 p = (last_found != (void *)1) ? last_found :
3598 get_all_packs(the_repository);
3600 while (p) {
3601 if ((!p->pack_local || p->pack_keep ||
3602 p->pack_keep_in_core) &&
3603 find_pack_entry_one(oid->hash, p)) {
3604 last_found = p;
3605 return 1;
3607 if (p == last_found)
3608 p = get_all_packs(the_repository);
3609 else
3610 p = p->next;
3611 if (p == last_found)
3612 p = p->next;
3614 return 0;
3618 * Store a list of sha1s that are should not be discarded
3619 * because they are either written too recently, or are
3620 * reachable from another object that was.
3622 * This is filled by get_object_list.
3624 static struct oid_array recent_objects;
3626 static int loosened_object_can_be_discarded(const struct object_id *oid,
3627 timestamp_t mtime)
3629 if (!unpack_unreachable_expiration)
3630 return 0;
3631 if (mtime > unpack_unreachable_expiration)
3632 return 0;
3633 if (oid_array_lookup(&recent_objects, oid) >= 0)
3634 return 0;
3635 return 1;
3638 static void loosen_unused_packed_objects(void)
3640 struct packed_git *p;
3641 uint32_t i;
3642 uint32_t loosened_objects_nr = 0;
3643 struct object_id oid;
3645 for (p = get_all_packs(the_repository); p; p = p->next) {
3646 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)
3647 continue;
3649 if (open_pack_index(p))
3650 die(_("cannot open pack index"));
3652 for (i = 0; i < p->num_objects; i++) {
3653 nth_packed_object_id(&oid, p, i);
3654 if (!packlist_find(&to_pack, &oid) &&
3655 !has_sha1_pack_kept_or_nonlocal(&oid) &&
3656 !loosened_object_can_be_discarded(&oid, p->mtime)) {
3657 if (force_object_loose(&oid, p->mtime))
3658 die(_("unable to force loose object"));
3659 loosened_objects_nr++;
3664 trace2_data_intmax("pack-objects", the_repository,
3665 "loosen_unused_packed_objects/loosened", loosened_objects_nr);
3669 * This tracks any options which pack-reuse code expects to be on, or which a
3670 * reader of the pack might not understand, and which would therefore prevent
3671 * blind reuse of what we have on disk.
3673 static int pack_options_allow_reuse(void)
3675 return allow_pack_reuse &&
3676 pack_to_stdout &&
3677 !ignore_packed_keep_on_disk &&
3678 !ignore_packed_keep_in_core &&
3679 (!local || !have_non_local_packs) &&
3680 !incremental;
3683 static int get_object_list_from_bitmap(struct rev_info *revs)
3685 if (!(bitmap_git = prepare_bitmap_walk(revs, &filter_options, 0)))
3686 return -1;
3688 if (pack_options_allow_reuse() &&
3689 !reuse_partial_packfile_from_bitmap(
3690 bitmap_git,
3691 &reuse_packfile,
3692 &reuse_packfile_objects,
3693 &reuse_packfile_bitmap)) {
3694 assert(reuse_packfile_objects);
3695 nr_result += reuse_packfile_objects;
3696 nr_seen += reuse_packfile_objects;
3697 display_progress(progress_state, nr_seen);
3700 traverse_bitmap_commit_list(bitmap_git, revs,
3701 &add_object_entry_from_bitmap);
3702 return 0;
3705 static void record_recent_object(struct object *obj,
3706 const char *name,
3707 void *data)
3709 oid_array_append(&recent_objects, &obj->oid);
3712 static void record_recent_commit(struct commit *commit, void *data)
3714 oid_array_append(&recent_objects, &commit->object.oid);
3717 static int mark_bitmap_preferred_tip(const char *refname,
3718 const struct object_id *oid, int flags,
3719 void *_data)
3721 struct object_id peeled;
3722 struct object *object;
3724 if (!peel_iterated_oid(oid, &peeled))
3725 oid = &peeled;
3727 object = parse_object_or_die(oid, refname);
3728 if (object->type == OBJ_COMMIT)
3729 object->flags |= NEEDS_BITMAP;
3731 return 0;
3734 static void mark_bitmap_preferred_tips(void)
3736 struct string_list_item *item;
3737 const struct string_list *preferred_tips;
3739 preferred_tips = bitmap_preferred_tips(the_repository);
3740 if (!preferred_tips)
3741 return;
3743 for_each_string_list_item(item, preferred_tips) {
3744 for_each_ref_in(item->string, mark_bitmap_preferred_tip, NULL);
3748 static void get_object_list(int ac, const char **av)
3750 struct rev_info revs;
3751 struct setup_revision_opt s_r_opt = {
3752 .allow_exclude_promisor_objects = 1,
3754 char line[1000];
3755 int flags = 0;
3756 int save_warning;
3758 repo_init_revisions(the_repository, &revs, NULL);
3759 save_commit_buffer = 0;
3760 setup_revisions(ac, av, &revs, &s_r_opt);
3762 /* make sure shallows are read */
3763 is_repository_shallow(the_repository);
3765 save_warning = warn_on_object_refname_ambiguity;
3766 warn_on_object_refname_ambiguity = 0;
3768 while (fgets(line, sizeof(line), stdin) != NULL) {
3769 int len = strlen(line);
3770 if (len && line[len - 1] == '\n')
3771 line[--len] = 0;
3772 if (!len)
3773 break;
3774 if (*line == '-') {
3775 if (!strcmp(line, "--not")) {
3776 flags ^= UNINTERESTING;
3777 write_bitmap_index = 0;
3778 continue;
3780 if (starts_with(line, "--shallow ")) {
3781 struct object_id oid;
3782 if (get_oid_hex(line + 10, &oid))
3783 die("not an object name '%s'", line + 10);
3784 register_shallow(the_repository, &oid);
3785 use_bitmap_index = 0;
3786 continue;
3788 die(_("not a rev '%s'"), line);
3790 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
3791 die(_("bad revision '%s'"), line);
3794 warn_on_object_refname_ambiguity = save_warning;
3796 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
3797 return;
3799 if (use_delta_islands)
3800 load_delta_islands(the_repository, progress);
3802 if (write_bitmap_index)
3803 mark_bitmap_preferred_tips();
3805 if (prepare_revision_walk(&revs))
3806 die(_("revision walk setup failed"));
3807 mark_edges_uninteresting(&revs, show_edge, sparse);
3809 if (!fn_show_object)
3810 fn_show_object = show_object;
3811 traverse_commit_list_filtered(&filter_options, &revs,
3812 show_commit, fn_show_object, NULL,
3813 NULL);
3815 if (unpack_unreachable_expiration) {
3816 revs.ignore_missing_links = 1;
3817 if (add_unseen_recent_objects_to_traversal(&revs,
3818 unpack_unreachable_expiration))
3819 die(_("unable to add recent objects"));
3820 if (prepare_revision_walk(&revs))
3821 die(_("revision walk setup failed"));
3822 traverse_commit_list(&revs, record_recent_commit,
3823 record_recent_object, NULL);
3826 if (keep_unreachable)
3827 add_objects_in_unpacked_packs();
3828 if (pack_loose_unreachable)
3829 add_unreachable_loose_objects();
3830 if (unpack_unreachable)
3831 loosen_unused_packed_objects();
3833 oid_array_clear(&recent_objects);
3836 static void add_extra_kept_packs(const struct string_list *names)
3838 struct packed_git *p;
3840 if (!names->nr)
3841 return;
3843 for (p = get_all_packs(the_repository); p; p = p->next) {
3844 const char *name = basename(p->pack_name);
3845 int i;
3847 if (!p->pack_local)
3848 continue;
3850 for (i = 0; i < names->nr; i++)
3851 if (!fspathcmp(name, names->items[i].string))
3852 break;
3854 if (i < names->nr) {
3855 p->pack_keep_in_core = 1;
3856 ignore_packed_keep_in_core = 1;
3857 continue;
3862 static int option_parse_index_version(const struct option *opt,
3863 const char *arg, int unset)
3865 char *c;
3866 const char *val = arg;
3868 BUG_ON_OPT_NEG(unset);
3870 pack_idx_opts.version = strtoul(val, &c, 10);
3871 if (pack_idx_opts.version > 2)
3872 die(_("unsupported index version %s"), val);
3873 if (*c == ',' && c[1])
3874 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
3875 if (*c || pack_idx_opts.off32_limit & 0x80000000)
3876 die(_("bad index version '%s'"), val);
3877 return 0;
3880 static int option_parse_unpack_unreachable(const struct option *opt,
3881 const char *arg, int unset)
3883 if (unset) {
3884 unpack_unreachable = 0;
3885 unpack_unreachable_expiration = 0;
3887 else {
3888 unpack_unreachable = 1;
3889 if (arg)
3890 unpack_unreachable_expiration = approxidate(arg);
3892 return 0;
3895 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
3897 int use_internal_rev_list = 0;
3898 int shallow = 0;
3899 int all_progress_implied = 0;
3900 struct strvec rp = STRVEC_INIT;
3901 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
3902 int rev_list_index = 0;
3903 int stdin_packs = 0;
3904 struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;
3905 struct option pack_objects_options[] = {
3906 OPT_SET_INT('q', "quiet", &progress,
3907 N_("do not show progress meter"), 0),
3908 OPT_SET_INT(0, "progress", &progress,
3909 N_("show progress meter"), 1),
3910 OPT_SET_INT(0, "all-progress", &progress,
3911 N_("show progress meter during object writing phase"), 2),
3912 OPT_BOOL(0, "all-progress-implied",
3913 &all_progress_implied,
3914 N_("similar to --all-progress when progress meter is shown")),
3915 OPT_CALLBACK_F(0, "index-version", NULL, N_("<version>[,<offset>]"),
3916 N_("write the pack index file in the specified idx format version"),
3917 PARSE_OPT_NONEG, option_parse_index_version),
3918 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
3919 N_("maximum size of each output pack file")),
3920 OPT_BOOL(0, "local", &local,
3921 N_("ignore borrowed objects from alternate object store")),
3922 OPT_BOOL(0, "incremental", &incremental,
3923 N_("ignore packed objects")),
3924 OPT_INTEGER(0, "window", &window,
3925 N_("limit pack window by objects")),
3926 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
3927 N_("limit pack window by memory in addition to object limit")),
3928 OPT_INTEGER(0, "depth", &depth,
3929 N_("maximum length of delta chain allowed in the resulting pack")),
3930 OPT_BOOL(0, "reuse-delta", &reuse_delta,
3931 N_("reuse existing deltas")),
3932 OPT_BOOL(0, "reuse-object", &reuse_object,
3933 N_("reuse existing objects")),
3934 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
3935 N_("use OFS_DELTA objects")),
3936 OPT_INTEGER(0, "threads", &delta_search_threads,
3937 N_("use threads when searching for best delta matches")),
3938 OPT_BOOL(0, "non-empty", &non_empty,
3939 N_("do not create an empty pack output")),
3940 OPT_BOOL(0, "revs", &use_internal_rev_list,
3941 N_("read revision arguments from standard input")),
3942 OPT_SET_INT_F(0, "unpacked", &rev_list_unpacked,
3943 N_("limit the objects to those that are not yet packed"),
3944 1, PARSE_OPT_NONEG),
3945 OPT_SET_INT_F(0, "all", &rev_list_all,
3946 N_("include objects reachable from any reference"),
3947 1, PARSE_OPT_NONEG),
3948 OPT_SET_INT_F(0, "reflog", &rev_list_reflog,
3949 N_("include objects referred by reflog entries"),
3950 1, PARSE_OPT_NONEG),
3951 OPT_SET_INT_F(0, "indexed-objects", &rev_list_index,
3952 N_("include objects referred to by the index"),
3953 1, PARSE_OPT_NONEG),
3954 OPT_BOOL(0, "stdin-packs", &stdin_packs,
3955 N_("read packs from stdin")),
3956 OPT_BOOL(0, "stdout", &pack_to_stdout,
3957 N_("output pack to stdout")),
3958 OPT_BOOL(0, "include-tag", &include_tag,
3959 N_("include tag objects that refer to objects to be packed")),
3960 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
3961 N_("keep unreachable objects")),
3962 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
3963 N_("pack loose unreachable objects")),
3964 OPT_CALLBACK_F(0, "unpack-unreachable", NULL, N_("time"),
3965 N_("unpack unreachable objects newer than <time>"),
3966 PARSE_OPT_OPTARG, option_parse_unpack_unreachable),
3967 OPT_BOOL(0, "sparse", &sparse,
3968 N_("use the sparse reachability algorithm")),
3969 OPT_BOOL(0, "thin", &thin,
3970 N_("create thin packs")),
3971 OPT_BOOL(0, "shallow", &shallow,
3972 N_("create packs suitable for shallow fetches")),
3973 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk,
3974 N_("ignore packs that have companion .keep file")),
3975 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list, N_("name"),
3976 N_("ignore this pack")),
3977 OPT_INTEGER(0, "compression", &pack_compression_level,
3978 N_("pack compression level")),
3979 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
3980 N_("do not hide commits by grafts"), 0),
3981 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
3982 N_("use a bitmap index if available to speed up counting objects")),
3983 OPT_SET_INT(0, "write-bitmap-index", &write_bitmap_index,
3984 N_("write a bitmap index together with the pack index"),
3985 WRITE_BITMAP_TRUE),
3986 OPT_SET_INT_F(0, "write-bitmap-index-quiet",
3987 &write_bitmap_index,
3988 N_("write a bitmap index if possible"),
3989 WRITE_BITMAP_QUIET, PARSE_OPT_HIDDEN),
3990 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),
3991 OPT_CALLBACK_F(0, "missing", NULL, N_("action"),
3992 N_("handling for missing objects"), PARSE_OPT_NONEG,
3993 option_parse_missing_action),
3994 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,
3995 N_("do not pack objects in promisor packfiles")),
3996 OPT_BOOL(0, "delta-islands", &use_delta_islands,
3997 N_("respect islands during delta compression")),
3998 OPT_STRING_LIST(0, "uri-protocol", &uri_protocols,
3999 N_("protocol"),
4000 N_("exclude any configured uploadpack.blobpackfileuri with this protocol")),
4001 OPT_END(),
4004 if (DFS_NUM_STATES > (1 << OE_DFS_STATE_BITS))
4005 BUG("too many dfs states, increase OE_DFS_STATE_BITS");
4007 read_replace_refs = 0;
4009 sparse = git_env_bool("GIT_TEST_PACK_SPARSE", -1);
4010 prepare_repo_settings(the_repository);
4011 if (sparse < 0)
4012 sparse = the_repository->settings.pack_use_sparse;
4014 reset_pack_idx_option(&pack_idx_opts);
4015 git_config(git_pack_config, NULL);
4016 if (git_env_bool(GIT_TEST_WRITE_REV_INDEX, 0))
4017 pack_idx_opts.flags |= WRITE_REV;
4019 progress = isatty(2);
4020 argc = parse_options(argc, argv, prefix, pack_objects_options,
4021 pack_usage, 0);
4023 if (argc) {
4024 base_name = argv[0];
4025 argc--;
4027 if (pack_to_stdout != !base_name || argc)
4028 usage_with_options(pack_usage, pack_objects_options);
4030 if (depth < 0)
4031 depth = 0;
4032 if (depth >= (1 << OE_DEPTH_BITS)) {
4033 warning(_("delta chain depth %d is too deep, forcing %d"),
4034 depth, (1 << OE_DEPTH_BITS) - 1);
4035 depth = (1 << OE_DEPTH_BITS) - 1;
4037 if (cache_max_small_delta_size >= (1U << OE_Z_DELTA_BITS)) {
4038 warning(_("pack.deltaCacheLimit is too high, forcing %d"),
4039 (1U << OE_Z_DELTA_BITS) - 1);
4040 cache_max_small_delta_size = (1U << OE_Z_DELTA_BITS) - 1;
4042 if (window < 0)
4043 window = 0;
4045 strvec_push(&rp, "pack-objects");
4046 if (thin) {
4047 use_internal_rev_list = 1;
4048 strvec_push(&rp, shallow
4049 ? "--objects-edge-aggressive"
4050 : "--objects-edge");
4051 } else
4052 strvec_push(&rp, "--objects");
4054 if (rev_list_all) {
4055 use_internal_rev_list = 1;
4056 strvec_push(&rp, "--all");
4058 if (rev_list_reflog) {
4059 use_internal_rev_list = 1;
4060 strvec_push(&rp, "--reflog");
4062 if (rev_list_index) {
4063 use_internal_rev_list = 1;
4064 strvec_push(&rp, "--indexed-objects");
4066 if (rev_list_unpacked && !stdin_packs) {
4067 use_internal_rev_list = 1;
4068 strvec_push(&rp, "--unpacked");
4071 if (exclude_promisor_objects) {
4072 use_internal_rev_list = 1;
4073 fetch_if_missing = 0;
4074 strvec_push(&rp, "--exclude-promisor-objects");
4076 if (unpack_unreachable || keep_unreachable || pack_loose_unreachable)
4077 use_internal_rev_list = 1;
4079 if (!reuse_object)
4080 reuse_delta = 0;
4081 if (pack_compression_level == -1)
4082 pack_compression_level = Z_DEFAULT_COMPRESSION;
4083 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
4084 die(_("bad pack compression level %d"), pack_compression_level);
4086 if (!delta_search_threads) /* --threads=0 means autodetect */
4087 delta_search_threads = online_cpus();
4089 if (!HAVE_THREADS && delta_search_threads != 1)
4090 warning(_("no threads support, ignoring --threads"));
4091 if (!pack_to_stdout && !pack_size_limit)
4092 pack_size_limit = pack_size_limit_cfg;
4093 if (pack_to_stdout && pack_size_limit)
4094 die(_("--max-pack-size cannot be used to build a pack for transfer"));
4095 if (pack_size_limit && pack_size_limit < 1024*1024) {
4096 warning(_("minimum pack size limit is 1 MiB"));
4097 pack_size_limit = 1024*1024;
4100 if (!pack_to_stdout && thin)
4101 die(_("--thin cannot be used to build an indexable pack"));
4103 if (keep_unreachable && unpack_unreachable)
4104 die(_("--keep-unreachable and --unpack-unreachable are incompatible"));
4105 if (!rev_list_all || !rev_list_reflog || !rev_list_index)
4106 unpack_unreachable_expiration = 0;
4108 if (filter_options.choice) {
4109 if (!pack_to_stdout)
4110 die(_("cannot use --filter without --stdout"));
4111 if (stdin_packs)
4112 die(_("cannot use --filter with --stdin-packs"));
4115 if (stdin_packs && use_internal_rev_list)
4116 die(_("cannot use internal rev list with --stdin-packs"));
4119 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
4121 * - to produce good pack (with bitmap index not-yet-packed objects are
4122 * packed in suboptimal order).
4124 * - to use more robust pack-generation codepath (avoiding possible
4125 * bugs in bitmap code and possible bitmap index corruption).
4127 if (!pack_to_stdout)
4128 use_bitmap_index_default = 0;
4130 if (use_bitmap_index < 0)
4131 use_bitmap_index = use_bitmap_index_default;
4133 /* "hard" reasons not to use bitmaps; these just won't work at all */
4134 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow(the_repository))
4135 use_bitmap_index = 0;
4137 if (pack_to_stdout || !rev_list_all)
4138 write_bitmap_index = 0;
4140 if (use_delta_islands)
4141 strvec_push(&rp, "--topo-order");
4143 if (progress && all_progress_implied)
4144 progress = 2;
4146 add_extra_kept_packs(&keep_pack_list);
4147 if (ignore_packed_keep_on_disk) {
4148 struct packed_git *p;
4149 for (p = get_all_packs(the_repository); p; p = p->next)
4150 if (p->pack_local && p->pack_keep)
4151 break;
4152 if (!p) /* no keep-able packs found */
4153 ignore_packed_keep_on_disk = 0;
4155 if (local) {
4157 * unlike ignore_packed_keep_on_disk above, we do not
4158 * want to unset "local" based on looking at packs, as
4159 * it also covers non-local objects
4161 struct packed_git *p;
4162 for (p = get_all_packs(the_repository); p; p = p->next) {
4163 if (!p->pack_local) {
4164 have_non_local_packs = 1;
4165 break;
4170 trace2_region_enter("pack-objects", "enumerate-objects",
4171 the_repository);
4172 prepare_packing_data(the_repository, &to_pack);
4174 if (progress)
4175 progress_state = start_progress(_("Enumerating objects"), 0);
4176 if (stdin_packs) {
4177 /* avoids adding objects in excluded packs */
4178 ignore_packed_keep_in_core = 1;
4179 read_packs_list_from_stdin();
4180 if (rev_list_unpacked)
4181 add_unreachable_loose_objects();
4182 } else if (!use_internal_rev_list)
4183 read_object_list_from_stdin();
4184 else {
4185 get_object_list(rp.nr, rp.v);
4186 strvec_clear(&rp);
4188 cleanup_preferred_base();
4189 if (include_tag && nr_result)
4190 for_each_tag_ref(add_ref_tag, NULL);
4191 stop_progress(&progress_state);
4192 trace2_region_leave("pack-objects", "enumerate-objects",
4193 the_repository);
4195 if (non_empty && !nr_result)
4196 return 0;
4197 if (nr_result) {
4198 trace2_region_enter("pack-objects", "prepare-pack",
4199 the_repository);
4200 prepare_pack(window, depth);
4201 trace2_region_leave("pack-objects", "prepare-pack",
4202 the_repository);
4205 trace2_region_enter("pack-objects", "write-pack-file", the_repository);
4206 write_excluded_by_configs();
4207 write_pack_file();
4208 trace2_region_leave("pack-objects", "write-pack-file", the_repository);
4210 if (progress)
4211 fprintf_ln(stderr,
4212 _("Total %"PRIu32" (delta %"PRIu32"),"
4213 " reused %"PRIu32" (delta %"PRIu32"),"
4214 " pack-reused %"PRIu32),
4215 written, written_delta, reused, reused_delta,
4216 reuse_packfile_objects);
4217 return 0;