3 #include "repository.h"
13 #include "pack-revindex.h"
14 #include "csum-file.h"
15 #include "tree-walk.h"
18 #include "list-objects.h"
19 #include "list-objects-filter.h"
20 #include "list-objects-filter-options.h"
21 #include "pack-objects.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 "sha1-array.h"
30 #include "argv-array.h"
33 #include "object-store.h"
38 #define IN_PACK(obj) oe_in_pack(&to_pack, obj)
39 #define SIZE(obj) oe_size(&to_pack, obj)
40 #define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size)
41 #define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj)
42 #define DELTA(obj) oe_delta(&to_pack, obj)
43 #define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj)
44 #define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj)
45 #define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val)
46 #define SET_DELTA_EXT(obj, oid) oe_set_delta_ext(&to_pack, obj, oid)
47 #define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val)
48 #define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val)
49 #define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val)
51 static const char *pack_usage[] = {
52 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
53 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
58 * Objects we are going to pack are collected in the `to_pack` structure.
59 * It contains an array (dynamically expanded) of the object data, and a map
60 * that can resolve SHA1s to their position in the array.
62 static struct packing_data to_pack;
64 static struct pack_idx_entry **written_list;
65 static uint32_t nr_result, nr_written, nr_seen;
66 static struct bitmap_index *bitmap_git;
67 static uint32_t write_layer;
70 static int reuse_delta = 1, reuse_object = 1;
71 static int keep_unreachable, unpack_unreachable, include_tag;
72 static timestamp_t unpack_unreachable_expiration;
73 static int pack_loose_unreachable;
75 static int have_non_local_packs;
76 static int incremental;
77 static int ignore_packed_keep_on_disk;
78 static int ignore_packed_keep_in_core;
79 static int allow_ofs_delta;
80 static struct pack_idx_option pack_idx_opts;
81 static const char *base_name;
82 static int progress = 1;
83 static int window = 10;
84 static unsigned long pack_size_limit;
85 static int depth = 50;
86 static int delta_search_threads;
87 static int pack_to_stdout;
90 static int num_preferred_base;
91 static struct progress *progress_state;
93 static struct packed_git *reuse_packfile;
94 static uint32_t reuse_packfile_objects;
95 static off_t reuse_packfile_offset;
97 static int use_bitmap_index_default = 1;
98 static int use_bitmap_index = -1;
100 WRITE_BITMAP_FALSE = 0,
103 } write_bitmap_index;
104 static uint16_t write_bitmap_options = BITMAP_OPT_HASH_CACHE;
106 static int exclude_promisor_objects;
108 static int use_delta_islands;
110 static unsigned long delta_cache_size = 0;
111 static unsigned long max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE;
112 static unsigned long cache_max_small_delta_size = 1000;
114 static unsigned long window_memory_limit = 0;
116 static struct list_objects_filter_options filter_options;
118 enum missing_action {
119 MA_ERROR = 0, /* fail if any missing objects are encountered */
120 MA_ALLOW_ANY, /* silently allow ALL missing objects */
121 MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */
123 static enum missing_action arg_missing_action;
124 static show_object_fn fn_show_object;
129 static uint32_t written, written_delta;
130 static uint32_t reused, reused_delta;
135 static struct commit **indexed_commits;
136 static unsigned int indexed_commits_nr;
137 static unsigned int indexed_commits_alloc;
139 static void index_commit_for_bitmap(struct commit *commit)
141 if (indexed_commits_nr >= indexed_commits_alloc) {
142 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
143 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
146 indexed_commits[indexed_commits_nr++] = commit;
149 static void *get_delta(struct object_entry *entry)
151 unsigned long size, base_size, delta_size;
152 void *buf, *base_buf, *delta_buf;
153 enum object_type type;
155 buf = read_object_file(&entry->idx.oid, &type, &size);
157 die(_("unable to read %s"), oid_to_hex(&entry->idx.oid));
158 base_buf = read_object_file(&DELTA(entry)->idx.oid, &type,
161 die("unable to read %s",
162 oid_to_hex(&DELTA(entry)->idx.oid));
163 delta_buf = diff_delta(base_buf, base_size,
164 buf, size, &delta_size, 0);
166 * We successfully computed this delta once but dropped it for
167 * memory reasons. Something is very wrong if this time we
168 * recompute and create a different delta.
170 if (!delta_buf || delta_size != DELTA_SIZE(entry))
171 BUG("delta size changed");
177 static unsigned long do_compress(void **pptr, unsigned long size)
181 unsigned long maxsize;
183 git_deflate_init(&stream, pack_compression_level);
184 maxsize = git_deflate_bound(&stream, size);
187 out = xmalloc(maxsize);
191 stream.avail_in = size;
192 stream.next_out = out;
193 stream.avail_out = maxsize;
194 while (git_deflate(&stream, Z_FINISH) == Z_OK)
196 git_deflate_end(&stream);
199 return stream.total_out;
202 static unsigned long write_large_blob_data(struct git_istream *st, struct hashfile *f,
203 const struct object_id *oid)
206 unsigned char ibuf[1024 * 16];
207 unsigned char obuf[1024 * 16];
208 unsigned long olen = 0;
210 git_deflate_init(&stream, pack_compression_level);
215 readlen = read_istream(st, ibuf, sizeof(ibuf));
217 die(_("unable to read %s"), oid_to_hex(oid));
219 stream.next_in = ibuf;
220 stream.avail_in = readlen;
221 while ((stream.avail_in || readlen == 0) &&
222 (zret == Z_OK || zret == Z_BUF_ERROR)) {
223 stream.next_out = obuf;
224 stream.avail_out = sizeof(obuf);
225 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
226 hashwrite(f, obuf, stream.next_out - obuf);
227 olen += stream.next_out - obuf;
230 die(_("deflate error (%d)"), zret);
232 if (zret != Z_STREAM_END)
233 die(_("deflate error (%d)"), zret);
237 git_deflate_end(&stream);
242 * we are going to reuse the existing object data as is. make
243 * sure it is not corrupt.
245 static int check_pack_inflate(struct packed_git *p,
246 struct pack_window **w_curs,
249 unsigned long expect)
252 unsigned char fakebuf[4096], *in;
255 memset(&stream, 0, sizeof(stream));
256 git_inflate_init(&stream);
258 in = use_pack(p, w_curs, offset, &stream.avail_in);
260 stream.next_out = fakebuf;
261 stream.avail_out = sizeof(fakebuf);
262 st = git_inflate(&stream, Z_FINISH);
263 offset += stream.next_in - in;
264 } while (st == Z_OK || st == Z_BUF_ERROR);
265 git_inflate_end(&stream);
266 return (st == Z_STREAM_END &&
267 stream.total_out == expect &&
268 stream.total_in == len) ? 0 : -1;
271 static void copy_pack_data(struct hashfile *f,
272 struct packed_git *p,
273 struct pack_window **w_curs,
281 in = use_pack(p, w_curs, offset, &avail);
283 avail = (unsigned long)len;
284 hashwrite(f, in, avail);
290 /* Return 0 if we will bust the pack-size limit */
291 static unsigned long write_no_reuse_object(struct hashfile *f, struct object_entry *entry,
292 unsigned long limit, int usable_delta)
294 unsigned long size, datalen;
295 unsigned char header[MAX_PACK_OBJECT_HEADER],
296 dheader[MAX_PACK_OBJECT_HEADER];
298 enum object_type type;
300 struct git_istream *st = NULL;
301 const unsigned hashsz = the_hash_algo->rawsz;
304 if (oe_type(entry) == OBJ_BLOB &&
305 oe_size_greater_than(&to_pack, entry, big_file_threshold) &&
306 (st = open_istream(&entry->idx.oid, &type, &size, NULL)) != NULL)
309 buf = read_object_file(&entry->idx.oid, &type, &size);
311 die(_("unable to read %s"),
312 oid_to_hex(&entry->idx.oid));
315 * make sure no cached delta data remains from a
316 * previous attempt before a pack split occurred.
318 FREE_AND_NULL(entry->delta_data);
319 entry->z_delta_size = 0;
320 } else if (entry->delta_data) {
321 size = DELTA_SIZE(entry);
322 buf = entry->delta_data;
323 entry->delta_data = NULL;
324 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
325 OBJ_OFS_DELTA : OBJ_REF_DELTA;
327 buf = get_delta(entry);
328 size = DELTA_SIZE(entry);
329 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
330 OBJ_OFS_DELTA : OBJ_REF_DELTA;
333 if (st) /* large blob case, just assume we don't compress well */
335 else if (entry->z_delta_size)
336 datalen = entry->z_delta_size;
338 datalen = do_compress(&buf, size);
341 * The object header is a byte of 'type' followed by zero or
342 * more bytes of length.
344 hdrlen = encode_in_pack_object_header(header, sizeof(header),
347 if (type == OBJ_OFS_DELTA) {
349 * Deltas with relative base contain an additional
350 * encoding of the relative offset for the delta
351 * base from this object's position in the pack.
353 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
354 unsigned pos = sizeof(dheader) - 1;
355 dheader[pos] = ofs & 127;
357 dheader[--pos] = 128 | (--ofs & 127);
358 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) {
364 hashwrite(f, header, hdrlen);
365 hashwrite(f, dheader + pos, sizeof(dheader) - pos);
366 hdrlen += sizeof(dheader) - pos;
367 } else if (type == OBJ_REF_DELTA) {
369 * Deltas with a base reference contain
370 * additional bytes for the base object ID.
372 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) {
378 hashwrite(f, header, hdrlen);
379 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz);
382 if (limit && hdrlen + datalen + hashsz >= limit) {
388 hashwrite(f, header, hdrlen);
391 datalen = write_large_blob_data(st, f, &entry->idx.oid);
394 hashwrite(f, buf, datalen);
398 return hdrlen + datalen;
401 /* Return 0 if we will bust the pack-size limit */
402 static off_t write_reuse_object(struct hashfile *f, struct object_entry *entry,
403 unsigned long limit, int usable_delta)
405 struct packed_git *p = IN_PACK(entry);
406 struct pack_window *w_curs = NULL;
407 struct revindex_entry *revidx;
409 enum object_type type = oe_type(entry);
411 unsigned char header[MAX_PACK_OBJECT_HEADER],
412 dheader[MAX_PACK_OBJECT_HEADER];
414 const unsigned hashsz = the_hash_algo->rawsz;
415 unsigned long entry_size = SIZE(entry);
418 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
419 OBJ_OFS_DELTA : OBJ_REF_DELTA;
420 hdrlen = encode_in_pack_object_header(header, sizeof(header),
423 offset = entry->in_pack_offset;
424 revidx = find_pack_revindex(p, offset);
425 datalen = revidx[1].offset - offset;
426 if (!pack_to_stdout && p->index_version > 1 &&
427 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
428 error(_("bad packed object CRC for %s"),
429 oid_to_hex(&entry->idx.oid));
431 return write_no_reuse_object(f, entry, limit, usable_delta);
434 offset += entry->in_pack_header_size;
435 datalen -= entry->in_pack_header_size;
437 if (!pack_to_stdout && p->index_version == 1 &&
438 check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) {
439 error(_("corrupt packed object for %s"),
440 oid_to_hex(&entry->idx.oid));
442 return write_no_reuse_object(f, entry, limit, usable_delta);
445 if (type == OBJ_OFS_DELTA) {
446 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
447 unsigned pos = sizeof(dheader) - 1;
448 dheader[pos] = ofs & 127;
450 dheader[--pos] = 128 | (--ofs & 127);
451 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) {
455 hashwrite(f, header, hdrlen);
456 hashwrite(f, dheader + pos, sizeof(dheader) - pos);
457 hdrlen += sizeof(dheader) - pos;
459 } else if (type == OBJ_REF_DELTA) {
460 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) {
464 hashwrite(f, header, hdrlen);
465 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz);
469 if (limit && hdrlen + datalen + hashsz >= limit) {
473 hashwrite(f, header, hdrlen);
475 copy_pack_data(f, p, &w_curs, offset, datalen);
478 return hdrlen + datalen;
481 /* Return 0 if we will bust the pack-size limit */
482 static off_t write_object(struct hashfile *f,
483 struct object_entry *entry,
488 int usable_delta, to_reuse;
493 /* apply size limit if limited packsize and not first object */
494 if (!pack_size_limit || !nr_written)
496 else if (pack_size_limit <= write_offset)
498 * the earlier object did not fit the limit; avoid
499 * mistaking this with unlimited (i.e. limit = 0).
503 limit = pack_size_limit - write_offset;
506 usable_delta = 0; /* no delta */
507 else if (!pack_size_limit)
508 usable_delta = 1; /* unlimited packfile */
509 else if (DELTA(entry)->idx.offset == (off_t)-1)
510 usable_delta = 0; /* base was written to another pack */
511 else if (DELTA(entry)->idx.offset)
512 usable_delta = 1; /* base already exists in this pack */
514 usable_delta = 0; /* base could end up in another pack */
517 to_reuse = 0; /* explicit */
518 else if (!IN_PACK(entry))
519 to_reuse = 0; /* can't reuse what we don't have */
520 else if (oe_type(entry) == OBJ_REF_DELTA ||
521 oe_type(entry) == OBJ_OFS_DELTA)
522 /* check_object() decided it for us ... */
523 to_reuse = usable_delta;
524 /* ... but pack split may override that */
525 else if (oe_type(entry) != entry->in_pack_type)
526 to_reuse = 0; /* pack has delta which is unusable */
527 else if (DELTA(entry))
528 to_reuse = 0; /* we want to pack afresh */
530 to_reuse = 1; /* we have it in-pack undeltified,
531 * and we do not need to deltify it.
535 len = write_no_reuse_object(f, entry, limit, usable_delta);
537 len = write_reuse_object(f, entry, limit, usable_delta);
545 entry->idx.crc32 = crc32_end(f);
549 enum write_one_status {
550 WRITE_ONE_SKIP = -1, /* already written */
551 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
552 WRITE_ONE_WRITTEN = 1, /* normal */
553 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
556 static enum write_one_status write_one(struct hashfile *f,
557 struct object_entry *e,
564 * we set offset to 1 (which is an impossible value) to mark
565 * the fact that this object is involved in "write its base
566 * first before writing a deltified object" recursion.
568 recursing = (e->idx.offset == 1);
570 warning(_("recursive delta detected for object %s"),
571 oid_to_hex(&e->idx.oid));
572 return WRITE_ONE_RECURSIVE;
573 } else if (e->idx.offset || e->preferred_base) {
574 /* offset is non zero if object is written already. */
575 return WRITE_ONE_SKIP;
578 /* if we are deltified, write out base object first. */
580 e->idx.offset = 1; /* now recurse */
581 switch (write_one(f, DELTA(e), offset)) {
582 case WRITE_ONE_RECURSIVE:
583 /* we cannot depend on this one */
588 case WRITE_ONE_BREAK:
589 e->idx.offset = recursing;
590 return WRITE_ONE_BREAK;
594 e->idx.offset = *offset;
595 size = write_object(f, e, *offset);
597 e->idx.offset = recursing;
598 return WRITE_ONE_BREAK;
600 written_list[nr_written++] = &e->idx;
602 /* make sure off_t is sufficiently large not to wrap */
603 if (signed_add_overflows(*offset, size))
604 die(_("pack too large for current definition of off_t"));
606 return WRITE_ONE_WRITTEN;
609 static int mark_tagged(const char *path, const struct object_id *oid, int flag,
612 struct object_id peeled;
613 struct object_entry *entry = packlist_find(&to_pack, oid);
617 if (!peel_ref(path, &peeled)) {
618 entry = packlist_find(&to_pack, &peeled);
625 static inline void add_to_write_order(struct object_entry **wo,
627 struct object_entry *e)
629 if (e->filled || oe_layer(&to_pack, e) != write_layer)
635 static void add_descendants_to_write_order(struct object_entry **wo,
637 struct object_entry *e)
639 int add_to_order = 1;
642 struct object_entry *s;
643 /* add this node... */
644 add_to_write_order(wo, endp, e);
645 /* all its siblings... */
646 for (s = DELTA_SIBLING(e); s; s = DELTA_SIBLING(s)) {
647 add_to_write_order(wo, endp, s);
650 /* drop down a level to add left subtree nodes if possible */
651 if (DELTA_CHILD(e)) {
656 /* our sibling might have some children, it is next */
657 if (DELTA_SIBLING(e)) {
658 e = DELTA_SIBLING(e);
661 /* go back to our parent node */
663 while (e && !DELTA_SIBLING(e)) {
664 /* we're on the right side of a subtree, keep
665 * going up until we can go right again */
669 /* done- we hit our original root node */
672 /* pass it off to sibling at this level */
673 e = DELTA_SIBLING(e);
678 static void add_family_to_write_order(struct object_entry **wo,
680 struct object_entry *e)
682 struct object_entry *root;
684 for (root = e; DELTA(root); root = DELTA(root))
686 add_descendants_to_write_order(wo, endp, root);
689 static void compute_layer_order(struct object_entry **wo, unsigned int *wo_end)
691 unsigned int i, last_untagged;
692 struct object_entry *objects = to_pack.objects;
694 for (i = 0; i < to_pack.nr_objects; i++) {
695 if (objects[i].tagged)
697 add_to_write_order(wo, wo_end, &objects[i]);
702 * Then fill all the tagged tips.
704 for (; i < to_pack.nr_objects; i++) {
705 if (objects[i].tagged)
706 add_to_write_order(wo, wo_end, &objects[i]);
710 * And then all remaining commits and tags.
712 for (i = last_untagged; i < to_pack.nr_objects; i++) {
713 if (oe_type(&objects[i]) != OBJ_COMMIT &&
714 oe_type(&objects[i]) != OBJ_TAG)
716 add_to_write_order(wo, wo_end, &objects[i]);
720 * And then all the trees.
722 for (i = last_untagged; i < to_pack.nr_objects; i++) {
723 if (oe_type(&objects[i]) != OBJ_TREE)
725 add_to_write_order(wo, wo_end, &objects[i]);
729 * Finally all the rest in really tight order
731 for (i = last_untagged; i < to_pack.nr_objects; i++) {
732 if (!objects[i].filled && oe_layer(&to_pack, &objects[i]) == write_layer)
733 add_family_to_write_order(wo, wo_end, &objects[i]);
737 static struct object_entry **compute_write_order(void)
739 uint32_t max_layers = 1;
740 unsigned int i, wo_end;
742 struct object_entry **wo;
743 struct object_entry *objects = to_pack.objects;
745 for (i = 0; i < to_pack.nr_objects; i++) {
746 objects[i].tagged = 0;
747 objects[i].filled = 0;
748 SET_DELTA_CHILD(&objects[i], NULL);
749 SET_DELTA_SIBLING(&objects[i], NULL);
753 * Fully connect delta_child/delta_sibling network.
754 * Make sure delta_sibling is sorted in the original
757 for (i = to_pack.nr_objects; i > 0;) {
758 struct object_entry *e = &objects[--i];
761 /* Mark me as the first child */
762 e->delta_sibling_idx = DELTA(e)->delta_child_idx;
763 SET_DELTA_CHILD(DELTA(e), e);
767 * Mark objects that are at the tip of tags.
769 for_each_tag_ref(mark_tagged, NULL);
771 if (use_delta_islands)
772 max_layers = compute_pack_layers(&to_pack);
774 ALLOC_ARRAY(wo, to_pack.nr_objects);
777 for (; write_layer < max_layers; ++write_layer)
778 compute_layer_order(wo, &wo_end);
780 if (wo_end != to_pack.nr_objects)
781 die(_("ordered %u objects, expected %"PRIu32),
782 wo_end, to_pack.nr_objects);
787 static off_t write_reused_pack(struct hashfile *f)
789 unsigned char buffer[8192];
790 off_t to_write, total;
793 if (!is_pack_valid(reuse_packfile))
794 die(_("packfile is invalid: %s"), reuse_packfile->pack_name);
796 fd = git_open(reuse_packfile->pack_name);
798 die_errno(_("unable to open packfile for reuse: %s"),
799 reuse_packfile->pack_name);
801 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
802 die_errno(_("unable to seek in reused packfile"));
804 if (reuse_packfile_offset < 0)
805 reuse_packfile_offset = reuse_packfile->pack_size - the_hash_algo->rawsz;
807 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
810 int read_pack = xread(fd, buffer, sizeof(buffer));
813 die_errno(_("unable to read from reused packfile"));
815 if (read_pack > to_write)
816 read_pack = to_write;
818 hashwrite(f, buffer, read_pack);
819 to_write -= read_pack;
822 * We don't know the actual number of objects written,
823 * only how many bytes written, how many bytes total, and
824 * how many objects total. So we can fake it by pretending all
825 * objects we are writing are the same size. This gives us a
826 * smooth progress meter, and at the end it matches the true
829 written = reuse_packfile_objects *
830 (((double)(total - to_write)) / total);
831 display_progress(progress_state, written);
835 written = reuse_packfile_objects;
836 display_progress(progress_state, written);
837 return reuse_packfile_offset - sizeof(struct pack_header);
840 static const char no_split_warning[] = N_(
841 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
844 static void write_pack_file(void)
849 uint32_t nr_remaining = nr_result;
850 time_t last_mtime = 0;
851 struct object_entry **write_order;
853 if (progress > pack_to_stdout)
854 progress_state = start_progress(_("Writing objects"), nr_result);
855 ALLOC_ARRAY(written_list, to_pack.nr_objects);
856 write_order = compute_write_order();
859 struct object_id oid;
860 char *pack_tmp_name = NULL;
863 f = hashfd_throughput(1, "<stdout>", progress_state);
865 f = create_tmp_packfile(&pack_tmp_name);
867 offset = write_pack_header(f, nr_remaining);
869 if (reuse_packfile) {
871 assert(pack_to_stdout);
873 packfile_size = write_reused_pack(f);
874 offset += packfile_size;
878 for (; i < to_pack.nr_objects; i++) {
879 struct object_entry *e = write_order[i];
880 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
882 display_progress(progress_state, written);
886 * Did we write the wrong # entries in the header?
887 * If so, rewrite it like in fast-import
889 if (pack_to_stdout) {
890 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE);
891 } else if (nr_written == nr_remaining) {
892 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE);
894 int fd = finalize_hashfile(f, oid.hash, 0);
895 fixup_pack_header_footer(fd, oid.hash, pack_tmp_name,
896 nr_written, oid.hash, offset);
898 if (write_bitmap_index) {
899 if (write_bitmap_index != WRITE_BITMAP_QUIET)
900 warning(_(no_split_warning));
901 write_bitmap_index = 0;
905 if (!pack_to_stdout) {
907 struct strbuf tmpname = STRBUF_INIT;
910 * Packs are runtime accessed in their mtime
911 * order since newer packs are more likely to contain
912 * younger objects. So if we are creating multiple
913 * packs then we should modify the mtime of later ones
914 * to preserve this property.
916 if (stat(pack_tmp_name, &st) < 0) {
917 warning_errno(_("failed to stat %s"), pack_tmp_name);
918 } else if (!last_mtime) {
919 last_mtime = st.st_mtime;
922 utb.actime = st.st_atime;
923 utb.modtime = --last_mtime;
924 if (utime(pack_tmp_name, &utb) < 0)
925 warning_errno(_("failed utime() on %s"), pack_tmp_name);
928 strbuf_addf(&tmpname, "%s-", base_name);
930 if (write_bitmap_index) {
931 bitmap_writer_set_checksum(oid.hash);
932 bitmap_writer_build_type_index(
933 &to_pack, written_list, nr_written);
936 finish_tmp_packfile(&tmpname, pack_tmp_name,
937 written_list, nr_written,
938 &pack_idx_opts, oid.hash);
940 if (write_bitmap_index) {
941 strbuf_addf(&tmpname, "%s.bitmap", oid_to_hex(&oid));
943 stop_progress(&progress_state);
945 bitmap_writer_show_progress(progress);
946 bitmap_writer_reuse_bitmaps(&to_pack);
947 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
948 bitmap_writer_build(&to_pack);
949 bitmap_writer_finish(written_list, nr_written,
950 tmpname.buf, write_bitmap_options);
951 write_bitmap_index = 0;
954 strbuf_release(&tmpname);
956 puts(oid_to_hex(&oid));
959 /* mark written objects as written to previous pack */
960 for (j = 0; j < nr_written; j++) {
961 written_list[j]->offset = (off_t)-1;
963 nr_remaining -= nr_written;
964 } while (nr_remaining && i < to_pack.nr_objects);
968 stop_progress(&progress_state);
969 if (written != nr_result)
970 die(_("wrote %"PRIu32" objects while expecting %"PRIu32),
972 trace2_data_intmax("pack-objects", the_repository,
973 "write_pack_file/wrote", nr_result);
976 static int no_try_delta(const char *path)
978 static struct attr_check *check;
981 check = attr_check_initl("delta", NULL);
982 git_check_attr(the_repository->index, path, check);
983 if (ATTR_FALSE(check->items[0].value))
989 * When adding an object, check whether we have already added it
990 * to our packing list. If so, we can skip. However, if we are
991 * being asked to excludei t, but the previous mention was to include
992 * it, make sure to adjust its flags and tweak our numbers accordingly.
994 * As an optimization, we pass out the index position where we would have
995 * found the item, since that saves us from having to look it up again a
996 * few lines later when we want to add the new entry.
998 static int have_duplicate_entry(const struct object_id *oid,
1001 struct object_entry *entry;
1003 entry = packlist_find(&to_pack, oid);
1008 if (!entry->preferred_base)
1010 entry->preferred_base = 1;
1016 static int want_found_object(int exclude, struct packed_git *p)
1024 * When asked to do --local (do not include an object that appears in a
1025 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
1026 * an object that appears in a pack marked with .keep), finding a pack
1027 * that matches the criteria is sufficient for us to decide to omit it.
1028 * However, even if this pack does not satisfy the criteria, we need to
1029 * make sure no copy of this object appears in _any_ pack that makes us
1030 * to omit the object, so we need to check all the packs.
1032 * We can however first check whether these options can possible matter;
1033 * if they do not matter we know we want the object in generated pack.
1034 * Otherwise, we signal "-1" at the end to tell the caller that we do
1035 * not know either way, and it needs to check more packs.
1037 if (!ignore_packed_keep_on_disk &&
1038 !ignore_packed_keep_in_core &&
1039 (!local || !have_non_local_packs))
1042 if (local && !p->pack_local)
1044 if (p->pack_local &&
1045 ((ignore_packed_keep_on_disk && p->pack_keep) ||
1046 (ignore_packed_keep_in_core && p->pack_keep_in_core)))
1049 /* we don't know yet; keep looking for more packs */
1054 * Check whether we want the object in the pack (e.g., we do not want
1055 * objects found in non-local stores if the "--local" option was used).
1057 * If the caller already knows an existing pack it wants to take the object
1058 * from, that is passed in *found_pack and *found_offset; otherwise this
1059 * function finds if there is any pack that has the object and returns the pack
1060 * and its offset in these variables.
1062 static int want_object_in_pack(const struct object_id *oid,
1064 struct packed_git **found_pack,
1065 off_t *found_offset)
1068 struct list_head *pos;
1069 struct multi_pack_index *m;
1071 if (!exclude && local && has_loose_object_nonlocal(oid))
1075 * If we already know the pack object lives in, start checks from that
1076 * pack - in the usual case when neither --local was given nor .keep files
1077 * are present we will determine the answer right now.
1080 want = want_found_object(exclude, *found_pack);
1085 for (m = get_multi_pack_index(the_repository); m; m = m->next) {
1086 struct pack_entry e;
1087 if (fill_midx_entry(the_repository, oid, &e, m)) {
1088 struct packed_git *p = e.p;
1091 if (p == *found_pack)
1092 offset = *found_offset;
1094 offset = find_pack_entry_one(oid->hash, p);
1098 if (!is_pack_valid(p))
1100 *found_offset = offset;
1103 want = want_found_object(exclude, p);
1110 list_for_each(pos, get_packed_git_mru(the_repository)) {
1111 struct packed_git *p = list_entry(pos, struct packed_git, mru);
1114 if (p == *found_pack)
1115 offset = *found_offset;
1117 offset = find_pack_entry_one(oid->hash, p);
1121 if (!is_pack_valid(p))
1123 *found_offset = offset;
1126 want = want_found_object(exclude, p);
1127 if (!exclude && want > 0)
1129 get_packed_git_mru(the_repository));
1138 static void create_object_entry(const struct object_id *oid,
1139 enum object_type type,
1143 struct packed_git *found_pack,
1146 struct object_entry *entry;
1148 entry = packlist_alloc(&to_pack, oid);
1150 oe_set_type(entry, type);
1152 entry->preferred_base = 1;
1156 oe_set_in_pack(&to_pack, entry, found_pack);
1157 entry->in_pack_offset = found_offset;
1160 entry->no_try_delta = no_try_delta;
1163 static const char no_closure_warning[] = N_(
1164 "disabling bitmap writing, as some objects are not being packed"
1167 static int add_object_entry(const struct object_id *oid, enum object_type type,
1168 const char *name, int exclude)
1170 struct packed_git *found_pack = NULL;
1171 off_t found_offset = 0;
1173 display_progress(progress_state, ++nr_seen);
1175 if (have_duplicate_entry(oid, exclude))
1178 if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {
1179 /* The pack is missing an object, so it will not have closure */
1180 if (write_bitmap_index) {
1181 if (write_bitmap_index != WRITE_BITMAP_QUIET)
1182 warning(_(no_closure_warning));
1183 write_bitmap_index = 0;
1188 create_object_entry(oid, type, pack_name_hash(name),
1189 exclude, name && no_try_delta(name),
1190 found_pack, found_offset);
1194 static int add_object_entry_from_bitmap(const struct object_id *oid,
1195 enum object_type type,
1196 int flags, uint32_t name_hash,
1197 struct packed_git *pack, off_t offset)
1199 display_progress(progress_state, ++nr_seen);
1201 if (have_duplicate_entry(oid, 0))
1204 if (!want_object_in_pack(oid, 0, &pack, &offset))
1207 create_object_entry(oid, type, name_hash, 0, 0, pack, offset);
1211 struct pbase_tree_cache {
1212 struct object_id oid;
1216 unsigned long tree_size;
1219 static struct pbase_tree_cache *(pbase_tree_cache[256]);
1220 static int pbase_tree_cache_ix(const struct object_id *oid)
1222 return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);
1224 static int pbase_tree_cache_ix_incr(int ix)
1226 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1229 static struct pbase_tree {
1230 struct pbase_tree *next;
1231 /* This is a phony "cache" entry; we are not
1232 * going to evict it or find it through _get()
1233 * mechanism -- this is for the toplevel node that
1234 * would almost always change with any commit.
1236 struct pbase_tree_cache pcache;
1239 static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)
1241 struct pbase_tree_cache *ent, *nent;
1244 enum object_type type;
1246 int my_ix = pbase_tree_cache_ix(oid);
1247 int available_ix = -1;
1249 /* pbase-tree-cache acts as a limited hashtable.
1250 * your object will be found at your index or within a few
1251 * slots after that slot if it is cached.
1253 for (neigh = 0; neigh < 8; neigh++) {
1254 ent = pbase_tree_cache[my_ix];
1255 if (ent && oideq(&ent->oid, oid)) {
1259 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1260 ((0 <= available_ix) &&
1261 (!ent && pbase_tree_cache[available_ix])))
1262 available_ix = my_ix;
1265 my_ix = pbase_tree_cache_ix_incr(my_ix);
1268 /* Did not find one. Either we got a bogus request or
1269 * we need to read and perhaps cache.
1271 data = read_object_file(oid, &type, &size);
1274 if (type != OBJ_TREE) {
1279 /* We need to either cache or return a throwaway copy */
1281 if (available_ix < 0)
1284 ent = pbase_tree_cache[available_ix];
1285 my_ix = available_ix;
1289 nent = xmalloc(sizeof(*nent));
1290 nent->temporary = (available_ix < 0);
1293 /* evict and reuse */
1294 free(ent->tree_data);
1297 oidcpy(&nent->oid, oid);
1298 nent->tree_data = data;
1299 nent->tree_size = size;
1301 if (!nent->temporary)
1302 pbase_tree_cache[my_ix] = nent;
1306 static void pbase_tree_put(struct pbase_tree_cache *cache)
1308 if (!cache->temporary) {
1312 free(cache->tree_data);
1316 static int name_cmp_len(const char *name)
1319 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1324 static void add_pbase_object(struct tree_desc *tree,
1327 const char *fullname)
1329 struct name_entry entry;
1332 while (tree_entry(tree,&entry)) {
1333 if (S_ISGITLINK(entry.mode))
1335 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1336 memcmp(name, entry.path, cmplen);
1341 if (name[cmplen] != '/') {
1342 add_object_entry(&entry.oid,
1343 object_type(entry.mode),
1347 if (S_ISDIR(entry.mode)) {
1348 struct tree_desc sub;
1349 struct pbase_tree_cache *tree;
1350 const char *down = name+cmplen+1;
1351 int downlen = name_cmp_len(down);
1353 tree = pbase_tree_get(&entry.oid);
1356 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1358 add_pbase_object(&sub, down, downlen, fullname);
1359 pbase_tree_put(tree);
1364 static unsigned *done_pbase_paths;
1365 static int done_pbase_paths_num;
1366 static int done_pbase_paths_alloc;
1367 static int done_pbase_path_pos(unsigned hash)
1370 int hi = done_pbase_paths_num;
1372 int mi = lo + (hi - lo) / 2;
1373 if (done_pbase_paths[mi] == hash)
1375 if (done_pbase_paths[mi] < hash)
1383 static int check_pbase_path(unsigned hash)
1385 int pos = done_pbase_path_pos(hash);
1389 ALLOC_GROW(done_pbase_paths,
1390 done_pbase_paths_num + 1,
1391 done_pbase_paths_alloc);
1392 done_pbase_paths_num++;
1393 if (pos < done_pbase_paths_num)
1394 MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,
1395 done_pbase_paths_num - pos - 1);
1396 done_pbase_paths[pos] = hash;
1400 static void add_preferred_base_object(const char *name)
1402 struct pbase_tree *it;
1404 unsigned hash = pack_name_hash(name);
1406 if (!num_preferred_base || check_pbase_path(hash))
1409 cmplen = name_cmp_len(name);
1410 for (it = pbase_tree; it; it = it->next) {
1412 add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);
1415 struct tree_desc tree;
1416 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1417 add_pbase_object(&tree, name, cmplen, name);
1422 static void add_preferred_base(struct object_id *oid)
1424 struct pbase_tree *it;
1427 struct object_id tree_oid;
1429 if (window <= num_preferred_base++)
1432 data = read_object_with_reference(the_repository, oid,
1433 tree_type, &size, &tree_oid);
1437 for (it = pbase_tree; it; it = it->next) {
1438 if (oideq(&it->pcache.oid, &tree_oid)) {
1444 it = xcalloc(1, sizeof(*it));
1445 it->next = pbase_tree;
1448 oidcpy(&it->pcache.oid, &tree_oid);
1449 it->pcache.tree_data = data;
1450 it->pcache.tree_size = size;
1453 static void cleanup_preferred_base(void)
1455 struct pbase_tree *it;
1461 struct pbase_tree *tmp = it;
1463 free(tmp->pcache.tree_data);
1467 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1468 if (!pbase_tree_cache[i])
1470 free(pbase_tree_cache[i]->tree_data);
1471 FREE_AND_NULL(pbase_tree_cache[i]);
1474 FREE_AND_NULL(done_pbase_paths);
1475 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1479 * Return 1 iff the object specified by "delta" can be sent
1480 * literally as a delta against the base in "base_sha1". If
1481 * so, then *base_out will point to the entry in our packing
1482 * list, or NULL if we must use the external-base list.
1484 * Depth value does not matter - find_deltas() will
1485 * never consider reused delta as the base object to
1486 * deltify other objects against, in order to avoid
1489 static int can_reuse_delta(const unsigned char *base_sha1,
1490 struct object_entry *delta,
1491 struct object_entry **base_out)
1493 struct object_entry *base;
1494 struct object_id base_oid;
1499 oidread(&base_oid, base_sha1);
1502 * First see if we're already sending the base (or it's explicitly in
1503 * our "excluded" list).
1505 base = packlist_find(&to_pack, &base_oid);
1507 if (!in_same_island(&delta->idx.oid, &base->idx.oid))
1514 * Otherwise, reachability bitmaps may tell us if the receiver has it,
1515 * even if it was buried too deep in history to make it into the
1518 if (thin && bitmap_has_oid_in_uninteresting(bitmap_git, &base_oid)) {
1519 if (use_delta_islands) {
1520 if (!in_same_island(&delta->idx.oid, &base_oid))
1530 static void check_object(struct object_entry *entry)
1532 unsigned long canonical_size;
1534 if (IN_PACK(entry)) {
1535 struct packed_git *p = IN_PACK(entry);
1536 struct pack_window *w_curs = NULL;
1537 const unsigned char *base_ref = NULL;
1538 struct object_entry *base_entry;
1539 unsigned long used, used_0;
1540 unsigned long avail;
1542 unsigned char *buf, c;
1543 enum object_type type;
1544 unsigned long in_pack_size;
1546 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1549 * We want in_pack_type even if we do not reuse delta
1550 * since non-delta representations could still be reused.
1552 used = unpack_object_header_buffer(buf, avail,
1559 BUG("invalid type %d", type);
1560 entry->in_pack_type = type;
1563 * Determine if this is a delta and if so whether we can
1564 * reuse it or not. Otherwise let's find out as cheaply as
1565 * possible what the actual type and size for this object is.
1567 switch (entry->in_pack_type) {
1569 /* Not a delta hence we've already got all we need. */
1570 oe_set_type(entry, entry->in_pack_type);
1571 SET_SIZE(entry, in_pack_size);
1572 entry->in_pack_header_size = used;
1573 if (oe_type(entry) < OBJ_COMMIT || oe_type(entry) > OBJ_BLOB)
1575 unuse_pack(&w_curs);
1578 if (reuse_delta && !entry->preferred_base)
1579 base_ref = use_pack(p, &w_curs,
1580 entry->in_pack_offset + used, NULL);
1581 entry->in_pack_header_size = used + the_hash_algo->rawsz;
1584 buf = use_pack(p, &w_curs,
1585 entry->in_pack_offset + used, NULL);
1591 if (!ofs || MSB(ofs, 7)) {
1592 error(_("delta base offset overflow in pack for %s"),
1593 oid_to_hex(&entry->idx.oid));
1597 ofs = (ofs << 7) + (c & 127);
1599 ofs = entry->in_pack_offset - ofs;
1600 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1601 error(_("delta base offset out of bound for %s"),
1602 oid_to_hex(&entry->idx.oid));
1605 if (reuse_delta && !entry->preferred_base) {
1606 struct revindex_entry *revidx;
1607 revidx = find_pack_revindex(p, ofs);
1610 base_ref = nth_packed_object_sha1(p, revidx->nr);
1612 entry->in_pack_header_size = used + used_0;
1616 if (can_reuse_delta(base_ref, entry, &base_entry)) {
1617 oe_set_type(entry, entry->in_pack_type);
1618 SET_SIZE(entry, in_pack_size); /* delta size */
1619 SET_DELTA_SIZE(entry, in_pack_size);
1622 SET_DELTA(entry, base_entry);
1623 entry->delta_sibling_idx = base_entry->delta_child_idx;
1624 SET_DELTA_CHILD(base_entry, entry);
1626 SET_DELTA_EXT(entry, base_ref);
1629 unuse_pack(&w_curs);
1633 if (oe_type(entry)) {
1637 * This must be a delta and we already know what the
1638 * final object type is. Let's extract the actual
1639 * object size from the delta header.
1641 delta_pos = entry->in_pack_offset + entry->in_pack_header_size;
1642 canonical_size = get_size_from_delta(p, &w_curs, delta_pos);
1643 if (canonical_size == 0)
1645 SET_SIZE(entry, canonical_size);
1646 unuse_pack(&w_curs);
1651 * No choice but to fall back to the recursive delta walk
1652 * with oid_object_info() to find about the object type
1656 unuse_pack(&w_curs);
1660 oid_object_info(the_repository, &entry->idx.oid, &canonical_size));
1661 if (entry->type_valid) {
1662 SET_SIZE(entry, canonical_size);
1665 * Bad object type is checked in prepare_pack(). This is
1666 * to permit a missing preferred base object to be ignored
1667 * as a preferred base. Doing so can result in a larger
1668 * pack file, but the transfer will still take place.
1673 static int pack_offset_sort(const void *_a, const void *_b)
1675 const struct object_entry *a = *(struct object_entry **)_a;
1676 const struct object_entry *b = *(struct object_entry **)_b;
1677 const struct packed_git *a_in_pack = IN_PACK(a);
1678 const struct packed_git *b_in_pack = IN_PACK(b);
1680 /* avoid filesystem trashing with loose objects */
1681 if (!a_in_pack && !b_in_pack)
1682 return oidcmp(&a->idx.oid, &b->idx.oid);
1684 if (a_in_pack < b_in_pack)
1686 if (a_in_pack > b_in_pack)
1688 return a->in_pack_offset < b->in_pack_offset ? -1 :
1689 (a->in_pack_offset > b->in_pack_offset);
1693 * Drop an on-disk delta we were planning to reuse. Naively, this would
1694 * just involve blanking out the "delta" field, but we have to deal
1695 * with some extra book-keeping:
1697 * 1. Removing ourselves from the delta_sibling linked list.
1699 * 2. Updating our size/type to the non-delta representation. These were
1700 * either not recorded initially (size) or overwritten with the delta type
1701 * (type) when check_object() decided to reuse the delta.
1703 * 3. Resetting our delta depth, as we are now a base object.
1705 static void drop_reused_delta(struct object_entry *entry)
1707 unsigned *idx = &to_pack.objects[entry->delta_idx - 1].delta_child_idx;
1708 struct object_info oi = OBJECT_INFO_INIT;
1709 enum object_type type;
1713 struct object_entry *oe = &to_pack.objects[*idx - 1];
1716 *idx = oe->delta_sibling_idx;
1718 idx = &oe->delta_sibling_idx;
1720 SET_DELTA(entry, NULL);
1725 if (packed_object_info(the_repository, IN_PACK(entry), entry->in_pack_offset, &oi) < 0) {
1727 * We failed to get the info from this pack for some reason;
1728 * fall back to oid_object_info, which may find another copy.
1729 * And if that fails, the error will be recorded in oe_type(entry)
1730 * and dealt with in prepare_pack().
1733 oid_object_info(the_repository, &entry->idx.oid, &size));
1735 oe_set_type(entry, type);
1737 SET_SIZE(entry, size);
1741 * Follow the chain of deltas from this entry onward, throwing away any links
1742 * that cause us to hit a cycle (as determined by the DFS state flags in
1745 * We also detect too-long reused chains that would violate our --depth
1748 static void break_delta_chains(struct object_entry *entry)
1751 * The actual depth of each object we will write is stored as an int,
1752 * as it cannot exceed our int "depth" limit. But before we break
1753 * changes based no that limit, we may potentially go as deep as the
1754 * number of objects, which is elsewhere bounded to a uint32_t.
1756 uint32_t total_depth;
1757 struct object_entry *cur, *next;
1759 for (cur = entry, total_depth = 0;
1761 cur = DELTA(cur), total_depth++) {
1762 if (cur->dfs_state == DFS_DONE) {
1764 * We've already seen this object and know it isn't
1765 * part of a cycle. We do need to append its depth
1768 total_depth += cur->depth;
1773 * We break cycles before looping, so an ACTIVE state (or any
1774 * other cruft which made its way into the state variable)
1777 if (cur->dfs_state != DFS_NONE)
1778 BUG("confusing delta dfs state in first pass: %d",
1782 * Now we know this is the first time we've seen the object. If
1783 * it's not a delta, we're done traversing, but we'll mark it
1784 * done to save time on future traversals.
1787 cur->dfs_state = DFS_DONE;
1792 * Mark ourselves as active and see if the next step causes
1793 * us to cycle to another active object. It's important to do
1794 * this _before_ we loop, because it impacts where we make the
1795 * cut, and thus how our total_depth counter works.
1796 * E.g., We may see a partial loop like:
1798 * A -> B -> C -> D -> B
1800 * Cutting B->C breaks the cycle. But now the depth of A is
1801 * only 1, and our total_depth counter is at 3. The size of the
1802 * error is always one less than the size of the cycle we
1803 * broke. Commits C and D were "lost" from A's chain.
1805 * If we instead cut D->B, then the depth of A is correct at 3.
1806 * We keep all commits in the chain that we examined.
1808 cur->dfs_state = DFS_ACTIVE;
1809 if (DELTA(cur)->dfs_state == DFS_ACTIVE) {
1810 drop_reused_delta(cur);
1811 cur->dfs_state = DFS_DONE;
1817 * And now that we've gone all the way to the bottom of the chain, we
1818 * need to clear the active flags and set the depth fields as
1819 * appropriate. Unlike the loop above, which can quit when it drops a
1820 * delta, we need to keep going to look for more depth cuts. So we need
1821 * an extra "next" pointer to keep going after we reset cur->delta.
1823 for (cur = entry; cur; cur = next) {
1827 * We should have a chain of zero or more ACTIVE states down to
1828 * a final DONE. We can quit after the DONE, because either it
1829 * has no bases, or we've already handled them in a previous
1832 if (cur->dfs_state == DFS_DONE)
1834 else if (cur->dfs_state != DFS_ACTIVE)
1835 BUG("confusing delta dfs state in second pass: %d",
1839 * If the total_depth is more than depth, then we need to snip
1840 * the chain into two or more smaller chains that don't exceed
1841 * the maximum depth. Most of the resulting chains will contain
1842 * (depth + 1) entries (i.e., depth deltas plus one base), and
1843 * the last chain (i.e., the one containing entry) will contain
1844 * whatever entries are left over, namely
1845 * (total_depth % (depth + 1)) of them.
1847 * Since we are iterating towards decreasing depth, we need to
1848 * decrement total_depth as we go, and we need to write to the
1849 * entry what its final depth will be after all of the
1850 * snipping. Since we're snipping into chains of length (depth
1851 * + 1) entries, the final depth of an entry will be its
1852 * original depth modulo (depth + 1). Any time we encounter an
1853 * entry whose final depth is supposed to be zero, we snip it
1854 * from its delta base, thereby making it so.
1856 cur->depth = (total_depth--) % (depth + 1);
1858 drop_reused_delta(cur);
1860 cur->dfs_state = DFS_DONE;
1864 static void get_object_details(void)
1867 struct object_entry **sorted_by_offset;
1870 progress_state = start_progress(_("Counting objects"),
1871 to_pack.nr_objects);
1873 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1874 for (i = 0; i < to_pack.nr_objects; i++)
1875 sorted_by_offset[i] = to_pack.objects + i;
1876 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
1878 for (i = 0; i < to_pack.nr_objects; i++) {
1879 struct object_entry *entry = sorted_by_offset[i];
1880 check_object(entry);
1881 if (entry->type_valid &&
1882 oe_size_greater_than(&to_pack, entry, big_file_threshold))
1883 entry->no_try_delta = 1;
1884 display_progress(progress_state, i + 1);
1886 stop_progress(&progress_state);
1889 * This must happen in a second pass, since we rely on the delta
1890 * information for the whole list being completed.
1892 for (i = 0; i < to_pack.nr_objects; i++)
1893 break_delta_chains(&to_pack.objects[i]);
1895 free(sorted_by_offset);
1899 * We search for deltas in a list sorted by type, by filename hash, and then
1900 * by size, so that we see progressively smaller and smaller files.
1901 * That's because we prefer deltas to be from the bigger file
1902 * to the smaller -- deletes are potentially cheaper, but perhaps
1903 * more importantly, the bigger file is likely the more recent
1904 * one. The deepest deltas are therefore the oldest objects which are
1905 * less susceptible to be accessed often.
1907 static int type_size_sort(const void *_a, const void *_b)
1909 const struct object_entry *a = *(struct object_entry **)_a;
1910 const struct object_entry *b = *(struct object_entry **)_b;
1911 const enum object_type a_type = oe_type(a);
1912 const enum object_type b_type = oe_type(b);
1913 const unsigned long a_size = SIZE(a);
1914 const unsigned long b_size = SIZE(b);
1916 if (a_type > b_type)
1918 if (a_type < b_type)
1920 if (a->hash > b->hash)
1922 if (a->hash < b->hash)
1924 if (a->preferred_base > b->preferred_base)
1926 if (a->preferred_base < b->preferred_base)
1928 if (use_delta_islands) {
1929 const int island_cmp = island_delta_cmp(&a->idx.oid, &b->idx.oid);
1933 if (a_size > b_size)
1935 if (a_size < b_size)
1937 return a < b ? -1 : (a > b); /* newest first */
1941 struct object_entry *entry;
1943 struct delta_index *index;
1947 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1948 unsigned long delta_size)
1950 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1953 if (delta_size < cache_max_small_delta_size)
1956 /* cache delta, if objects are large enough compared to delta size */
1957 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1963 /* Protect delta_cache_size */
1964 static pthread_mutex_t cache_mutex;
1965 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1966 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1969 * Protect object list partitioning (e.g. struct thread_param) and
1972 static pthread_mutex_t progress_mutex;
1973 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1974 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1977 * Access to struct object_entry is unprotected since each thread owns
1978 * a portion of the main object list. Just don't access object entries
1979 * ahead in the list because they can be stolen and would need
1980 * progress_mutex for protection.
1984 * Return the size of the object without doing any delta
1985 * reconstruction (so non-deltas are true object sizes, but deltas
1986 * return the size of the delta data).
1988 unsigned long oe_get_size_slow(struct packing_data *pack,
1989 const struct object_entry *e)
1991 struct packed_git *p;
1992 struct pack_window *w_curs;
1994 enum object_type type;
1995 unsigned long used, avail, size;
1997 if (e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {
1998 packing_data_lock(&to_pack);
1999 if (oid_object_info(the_repository, &e->idx.oid, &size) < 0)
2000 die(_("unable to get size of %s"),
2001 oid_to_hex(&e->idx.oid));
2002 packing_data_unlock(&to_pack);
2006 p = oe_in_pack(pack, e);
2008 BUG("when e->type is a delta, it must belong to a pack");
2010 packing_data_lock(&to_pack);
2012 buf = use_pack(p, &w_curs, e->in_pack_offset, &avail);
2013 used = unpack_object_header_buffer(buf, avail, &type, &size);
2015 die(_("unable to parse object header of %s"),
2016 oid_to_hex(&e->idx.oid));
2018 unuse_pack(&w_curs);
2019 packing_data_unlock(&to_pack);
2023 static int try_delta(struct unpacked *trg, struct unpacked *src,
2024 unsigned max_depth, unsigned long *mem_usage)
2026 struct object_entry *trg_entry = trg->entry;
2027 struct object_entry *src_entry = src->entry;
2028 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
2030 enum object_type type;
2033 /* Don't bother doing diffs between different types */
2034 if (oe_type(trg_entry) != oe_type(src_entry))
2038 * We do not bother to try a delta that we discarded on an
2039 * earlier try, but only when reusing delta data. Note that
2040 * src_entry that is marked as the preferred_base should always
2041 * be considered, as even if we produce a suboptimal delta against
2042 * it, we will still save the transfer cost, as we already know
2043 * the other side has it and we won't send src_entry at all.
2045 if (reuse_delta && IN_PACK(trg_entry) &&
2046 IN_PACK(trg_entry) == IN_PACK(src_entry) &&
2047 !src_entry->preferred_base &&
2048 trg_entry->in_pack_type != OBJ_REF_DELTA &&
2049 trg_entry->in_pack_type != OBJ_OFS_DELTA)
2052 /* Let's not bust the allowed depth. */
2053 if (src->depth >= max_depth)
2056 /* Now some size filtering heuristics. */
2057 trg_size = SIZE(trg_entry);
2058 if (!DELTA(trg_entry)) {
2059 max_size = trg_size/2 - the_hash_algo->rawsz;
2062 max_size = DELTA_SIZE(trg_entry);
2063 ref_depth = trg->depth;
2065 max_size = (uint64_t)max_size * (max_depth - src->depth) /
2066 (max_depth - ref_depth + 1);
2069 src_size = SIZE(src_entry);
2070 sizediff = src_size < trg_size ? trg_size - src_size : 0;
2071 if (sizediff >= max_size)
2073 if (trg_size < src_size / 32)
2076 if (!in_same_island(&trg->entry->idx.oid, &src->entry->idx.oid))
2079 /* Load data if not already done */
2081 packing_data_lock(&to_pack);
2082 trg->data = read_object_file(&trg_entry->idx.oid, &type, &sz);
2083 packing_data_unlock(&to_pack);
2085 die(_("object %s cannot be read"),
2086 oid_to_hex(&trg_entry->idx.oid));
2088 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),
2089 oid_to_hex(&trg_entry->idx.oid), (uintmax_t)sz,
2090 (uintmax_t)trg_size);
2094 packing_data_lock(&to_pack);
2095 src->data = read_object_file(&src_entry->idx.oid, &type, &sz);
2096 packing_data_unlock(&to_pack);
2098 if (src_entry->preferred_base) {
2099 static int warned = 0;
2101 warning(_("object %s cannot be read"),
2102 oid_to_hex(&src_entry->idx.oid));
2104 * Those objects are not included in the
2105 * resulting pack. Be resilient and ignore
2106 * them if they can't be read, in case the
2107 * pack could be created nevertheless.
2111 die(_("object %s cannot be read"),
2112 oid_to_hex(&src_entry->idx.oid));
2115 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),
2116 oid_to_hex(&src_entry->idx.oid), (uintmax_t)sz,
2117 (uintmax_t)src_size);
2121 src->index = create_delta_index(src->data, src_size);
2123 static int warned = 0;
2125 warning(_("suboptimal pack - out of memory"));
2128 *mem_usage += sizeof_delta_index(src->index);
2131 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
2135 if (DELTA(trg_entry)) {
2136 /* Prefer only shallower same-sized deltas. */
2137 if (delta_size == DELTA_SIZE(trg_entry) &&
2138 src->depth + 1 >= trg->depth) {
2145 * Handle memory allocation outside of the cache
2146 * accounting lock. Compiler will optimize the strangeness
2147 * away when NO_PTHREADS is defined.
2149 free(trg_entry->delta_data);
2151 if (trg_entry->delta_data) {
2152 delta_cache_size -= DELTA_SIZE(trg_entry);
2153 trg_entry->delta_data = NULL;
2155 if (delta_cacheable(src_size, trg_size, delta_size)) {
2156 delta_cache_size += delta_size;
2158 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
2164 SET_DELTA(trg_entry, src_entry);
2165 SET_DELTA_SIZE(trg_entry, delta_size);
2166 trg->depth = src->depth + 1;
2171 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
2173 struct object_entry *child = DELTA_CHILD(me);
2176 const unsigned int c = check_delta_limit(child, n + 1);
2179 child = DELTA_SIBLING(child);
2184 static unsigned long free_unpacked(struct unpacked *n)
2186 unsigned long freed_mem = sizeof_delta_index(n->index);
2187 free_delta_index(n->index);
2190 freed_mem += SIZE(n->entry);
2191 FREE_AND_NULL(n->data);
2198 static void find_deltas(struct object_entry **list, unsigned *list_size,
2199 int window, int depth, unsigned *processed)
2201 uint32_t i, idx = 0, count = 0;
2202 struct unpacked *array;
2203 unsigned long mem_usage = 0;
2205 array = xcalloc(window, sizeof(struct unpacked));
2208 struct object_entry *entry;
2209 struct unpacked *n = array + idx;
2210 int j, max_depth, best_base = -1;
2219 if (!entry->preferred_base) {
2221 display_progress(progress_state, *processed);
2225 mem_usage -= free_unpacked(n);
2228 while (window_memory_limit &&
2229 mem_usage > window_memory_limit &&
2231 const uint32_t tail = (idx + window - count) % window;
2232 mem_usage -= free_unpacked(array + tail);
2236 /* We do not compute delta to *create* objects we are not
2239 if (entry->preferred_base)
2243 * If the current object is at pack edge, take the depth the
2244 * objects that depend on the current object into account
2245 * otherwise they would become too deep.
2248 if (DELTA_CHILD(entry)) {
2249 max_depth -= check_delta_limit(entry, 0);
2257 uint32_t other_idx = idx + j;
2259 if (other_idx >= window)
2260 other_idx -= window;
2261 m = array + other_idx;
2264 ret = try_delta(n, m, max_depth, &mem_usage);
2268 best_base = other_idx;
2272 * If we decided to cache the delta data, then it is best
2273 * to compress it right away. First because we have to do
2274 * it anyway, and doing it here while we're threaded will
2275 * save a lot of time in the non threaded write phase,
2276 * as well as allow for caching more deltas within
2277 * the same cache size limit.
2279 * But only if not writing to stdout, since in that case
2280 * the network is most likely throttling writes anyway,
2281 * and therefore it is best to go to the write phase ASAP
2282 * instead, as we can afford spending more time compressing
2283 * between writes at that moment.
2285 if (entry->delta_data && !pack_to_stdout) {
2288 size = do_compress(&entry->delta_data, DELTA_SIZE(entry));
2289 if (size < (1U << OE_Z_DELTA_BITS)) {
2290 entry->z_delta_size = size;
2292 delta_cache_size -= DELTA_SIZE(entry);
2293 delta_cache_size += entry->z_delta_size;
2296 FREE_AND_NULL(entry->delta_data);
2297 entry->z_delta_size = 0;
2301 /* if we made n a delta, and if n is already at max
2302 * depth, leaving it in the window is pointless. we
2303 * should evict it first.
2305 if (DELTA(entry) && max_depth <= n->depth)
2309 * Move the best delta base up in the window, after the
2310 * currently deltified object, to keep it longer. It will
2311 * be the first base object to be attempted next.
2314 struct unpacked swap = array[best_base];
2315 int dist = (window + idx - best_base) % window;
2316 int dst = best_base;
2318 int src = (dst + 1) % window;
2319 array[dst] = array[src];
2327 if (count + 1 < window)
2333 for (i = 0; i < window; ++i) {
2334 free_delta_index(array[i].index);
2335 free(array[i].data);
2341 * The main object list is split into smaller lists, each is handed to
2344 * The main thread waits on the condition that (at least) one of the workers
2345 * has stopped working (which is indicated in the .working member of
2346 * struct thread_params).
2348 * When a work thread has completed its work, it sets .working to 0 and
2349 * signals the main thread and waits on the condition that .data_ready
2352 * The main thread steals half of the work from the worker that has
2353 * most work left to hand it to the idle worker.
2356 struct thread_params {
2358 struct object_entry **list;
2365 pthread_mutex_t mutex;
2366 pthread_cond_t cond;
2367 unsigned *processed;
2370 static pthread_cond_t progress_cond;
2373 * Mutex and conditional variable can't be statically-initialized on Windows.
2375 static void init_threaded_search(void)
2377 pthread_mutex_init(&cache_mutex, NULL);
2378 pthread_mutex_init(&progress_mutex, NULL);
2379 pthread_cond_init(&progress_cond, NULL);
2382 static void cleanup_threaded_search(void)
2384 pthread_cond_destroy(&progress_cond);
2385 pthread_mutex_destroy(&cache_mutex);
2386 pthread_mutex_destroy(&progress_mutex);
2389 static void *threaded_find_deltas(void *arg)
2391 struct thread_params *me = arg;
2394 while (me->remaining) {
2397 find_deltas(me->list, &me->remaining,
2398 me->window, me->depth, me->processed);
2402 pthread_cond_signal(&progress_cond);
2406 * We must not set ->data_ready before we wait on the
2407 * condition because the main thread may have set it to 1
2408 * before we get here. In order to be sure that new
2409 * work is available if we see 1 in ->data_ready, it
2410 * was initialized to 0 before this thread was spawned
2411 * and we reset it to 0 right away.
2413 pthread_mutex_lock(&me->mutex);
2414 while (!me->data_ready)
2415 pthread_cond_wait(&me->cond, &me->mutex);
2417 pthread_mutex_unlock(&me->mutex);
2422 /* leave ->working 1 so that this doesn't get more work assigned */
2426 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2427 int window, int depth, unsigned *processed)
2429 struct thread_params *p;
2430 int i, ret, active_threads = 0;
2432 init_threaded_search();
2434 if (delta_search_threads <= 1) {
2435 find_deltas(list, &list_size, window, depth, processed);
2436 cleanup_threaded_search();
2439 if (progress > pack_to_stdout)
2440 fprintf_ln(stderr, _("Delta compression using up to %d threads"),
2441 delta_search_threads);
2442 p = xcalloc(delta_search_threads, sizeof(*p));
2444 /* Partition the work amongst work threads. */
2445 for (i = 0; i < delta_search_threads; i++) {
2446 unsigned sub_size = list_size / (delta_search_threads - i);
2448 /* don't use too small segments or no deltas will be found */
2449 if (sub_size < 2*window && i+1 < delta_search_threads)
2452 p[i].window = window;
2454 p[i].processed = processed;
2456 p[i].data_ready = 0;
2458 /* try to split chunks on "path" boundaries */
2459 while (sub_size && sub_size < list_size &&
2460 list[sub_size]->hash &&
2461 list[sub_size]->hash == list[sub_size-1]->hash)
2465 p[i].list_size = sub_size;
2466 p[i].remaining = sub_size;
2469 list_size -= sub_size;
2472 /* Start work threads. */
2473 for (i = 0; i < delta_search_threads; i++) {
2474 if (!p[i].list_size)
2476 pthread_mutex_init(&p[i].mutex, NULL);
2477 pthread_cond_init(&p[i].cond, NULL);
2478 ret = pthread_create(&p[i].thread, NULL,
2479 threaded_find_deltas, &p[i]);
2481 die(_("unable to create thread: %s"), strerror(ret));
2486 * Now let's wait for work completion. Each time a thread is done
2487 * with its work, we steal half of the remaining work from the
2488 * thread with the largest number of unprocessed objects and give
2489 * it to that newly idle thread. This ensure good load balancing
2490 * until the remaining object list segments are simply too short
2491 * to be worth splitting anymore.
2493 while (active_threads) {
2494 struct thread_params *target = NULL;
2495 struct thread_params *victim = NULL;
2496 unsigned sub_size = 0;
2500 for (i = 0; !target && i < delta_search_threads; i++)
2505 pthread_cond_wait(&progress_cond, &progress_mutex);
2508 for (i = 0; i < delta_search_threads; i++)
2509 if (p[i].remaining > 2*window &&
2510 (!victim || victim->remaining < p[i].remaining))
2513 sub_size = victim->remaining / 2;
2514 list = victim->list + victim->list_size - sub_size;
2515 while (sub_size && list[0]->hash &&
2516 list[0]->hash == list[-1]->hash) {
2522 * It is possible for some "paths" to have
2523 * so many objects that no hash boundary
2524 * might be found. Let's just steal the
2525 * exact half in that case.
2527 sub_size = victim->remaining / 2;
2530 target->list = list;
2531 victim->list_size -= sub_size;
2532 victim->remaining -= sub_size;
2534 target->list_size = sub_size;
2535 target->remaining = sub_size;
2536 target->working = 1;
2539 pthread_mutex_lock(&target->mutex);
2540 target->data_ready = 1;
2541 pthread_cond_signal(&target->cond);
2542 pthread_mutex_unlock(&target->mutex);
2545 pthread_join(target->thread, NULL);
2546 pthread_cond_destroy(&target->cond);
2547 pthread_mutex_destroy(&target->mutex);
2551 cleanup_threaded_search();
2555 static void add_tag_chain(const struct object_id *oid)
2560 * We catch duplicates already in add_object_entry(), but we'd
2561 * prefer to do this extra check to avoid having to parse the
2562 * tag at all if we already know that it's being packed (e.g., if
2563 * it was included via bitmaps, we would not have parsed it
2566 if (packlist_find(&to_pack, oid))
2569 tag = lookup_tag(the_repository, oid);
2571 if (!tag || parse_tag(tag) || !tag->tagged)
2572 die(_("unable to pack objects reachable from tag %s"),
2575 add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);
2577 if (tag->tagged->type != OBJ_TAG)
2580 tag = (struct tag *)tag->tagged;
2584 static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2586 struct object_id peeled;
2588 if (starts_with(path, "refs/tags/") && /* is a tag? */
2589 !peel_ref(path, &peeled) && /* peelable? */
2590 packlist_find(&to_pack, &peeled)) /* object packed? */
2595 static void prepare_pack(int window, int depth)
2597 struct object_entry **delta_list;
2598 uint32_t i, nr_deltas;
2601 if (use_delta_islands)
2602 resolve_tree_islands(the_repository, progress, &to_pack);
2604 get_object_details();
2607 * If we're locally repacking then we need to be doubly careful
2608 * from now on in order to make sure no stealth corruption gets
2609 * propagated to the new pack. Clients receiving streamed packs
2610 * should validate everything they get anyway so no need to incur
2611 * the additional cost here in that case.
2613 if (!pack_to_stdout)
2614 do_check_packed_object_crc = 1;
2616 if (!to_pack.nr_objects || !window || !depth)
2619 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2622 for (i = 0; i < to_pack.nr_objects; i++) {
2623 struct object_entry *entry = to_pack.objects + i;
2626 /* This happens if we decided to reuse existing
2627 * delta from a pack. "reuse_delta &&" is implied.
2631 if (!entry->type_valid ||
2632 oe_size_less_than(&to_pack, entry, 50))
2635 if (entry->no_try_delta)
2638 if (!entry->preferred_base) {
2640 if (oe_type(entry) < 0)
2641 die(_("unable to get type of object %s"),
2642 oid_to_hex(&entry->idx.oid));
2644 if (oe_type(entry) < 0) {
2646 * This object is not found, but we
2647 * don't have to include it anyway.
2653 delta_list[n++] = entry;
2656 if (nr_deltas && n > 1) {
2657 unsigned nr_done = 0;
2659 progress_state = start_progress(_("Compressing objects"),
2661 QSORT(delta_list, n, type_size_sort);
2662 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2663 stop_progress(&progress_state);
2664 if (nr_done != nr_deltas)
2665 die(_("inconsistency with delta count"));
2670 static int git_pack_config(const char *k, const char *v, void *cb)
2672 if (!strcmp(k, "pack.window")) {
2673 window = git_config_int(k, v);
2676 if (!strcmp(k, "pack.windowmemory")) {
2677 window_memory_limit = git_config_ulong(k, v);
2680 if (!strcmp(k, "pack.depth")) {
2681 depth = git_config_int(k, v);
2684 if (!strcmp(k, "pack.deltacachesize")) {
2685 max_delta_cache_size = git_config_int(k, v);
2688 if (!strcmp(k, "pack.deltacachelimit")) {
2689 cache_max_small_delta_size = git_config_int(k, v);
2692 if (!strcmp(k, "pack.writebitmaphashcache")) {
2693 if (git_config_bool(k, v))
2694 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2696 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2698 if (!strcmp(k, "pack.usebitmaps")) {
2699 use_bitmap_index_default = git_config_bool(k, v);
2702 if (!strcmp(k, "pack.threads")) {
2703 delta_search_threads = git_config_int(k, v);
2704 if (delta_search_threads < 0)
2705 die(_("invalid number of threads specified (%d)"),
2706 delta_search_threads);
2707 if (!HAVE_THREADS && delta_search_threads != 1) {
2708 warning(_("no threads support, ignoring %s"), k);
2709 delta_search_threads = 0;
2713 if (!strcmp(k, "pack.indexversion")) {
2714 pack_idx_opts.version = git_config_int(k, v);
2715 if (pack_idx_opts.version > 2)
2716 die(_("bad pack.indexversion=%"PRIu32),
2717 pack_idx_opts.version);
2720 return git_default_config(k, v, cb);
2723 static void read_object_list_from_stdin(void)
2725 char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];
2726 struct object_id oid;
2730 if (!fgets(line, sizeof(line), stdin)) {
2734 die("BUG: fgets returned NULL, not EOF, not error!");
2740 if (line[0] == '-') {
2741 if (get_oid_hex(line+1, &oid))
2742 die(_("expected edge object ID, got garbage:\n %s"),
2744 add_preferred_base(&oid);
2747 if (parse_oid_hex(line, &oid, &p))
2748 die(_("expected object ID, got garbage:\n %s"), line);
2750 add_preferred_base_object(p + 1);
2751 add_object_entry(&oid, OBJ_NONE, p + 1, 0);
2755 /* Remember to update object flag allocation in object.h */
2756 #define OBJECT_ADDED (1u<<20)
2758 static void show_commit(struct commit *commit, void *data)
2760 add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);
2761 commit->object.flags |= OBJECT_ADDED;
2763 if (write_bitmap_index)
2764 index_commit_for_bitmap(commit);
2766 if (use_delta_islands)
2767 propagate_island_marks(commit);
2770 static void show_object(struct object *obj, const char *name, void *data)
2772 add_preferred_base_object(name);
2773 add_object_entry(&obj->oid, obj->type, name, 0);
2774 obj->flags |= OBJECT_ADDED;
2776 if (use_delta_islands) {
2779 struct object_entry *ent;
2781 /* the empty string is a root tree, which is depth 0 */
2782 depth = *name ? 1 : 0;
2783 for (p = strchr(name, '/'); p; p = strchr(p + 1, '/'))
2786 ent = packlist_find(&to_pack, &obj->oid);
2787 if (ent && depth > oe_tree_depth(&to_pack, ent))
2788 oe_set_tree_depth(&to_pack, ent, depth);
2792 static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)
2794 assert(arg_missing_action == MA_ALLOW_ANY);
2797 * Quietly ignore ALL missing objects. This avoids problems with
2798 * staging them now and getting an odd error later.
2800 if (!has_object_file(&obj->oid))
2803 show_object(obj, name, data);
2806 static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)
2808 assert(arg_missing_action == MA_ALLOW_PROMISOR);
2811 * Quietly ignore EXPECTED missing objects. This avoids problems with
2812 * staging them now and getting an odd error later.
2814 if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))
2817 show_object(obj, name, data);
2820 static int option_parse_missing_action(const struct option *opt,
2821 const char *arg, int unset)
2826 if (!strcmp(arg, "error")) {
2827 arg_missing_action = MA_ERROR;
2828 fn_show_object = show_object;
2832 if (!strcmp(arg, "allow-any")) {
2833 arg_missing_action = MA_ALLOW_ANY;
2834 fetch_if_missing = 0;
2835 fn_show_object = show_object__ma_allow_any;
2839 if (!strcmp(arg, "allow-promisor")) {
2840 arg_missing_action = MA_ALLOW_PROMISOR;
2841 fetch_if_missing = 0;
2842 fn_show_object = show_object__ma_allow_promisor;
2846 die(_("invalid value for --missing"));
2850 static void show_edge(struct commit *commit)
2852 add_preferred_base(&commit->object.oid);
2855 struct in_pack_object {
2857 struct object *object;
2863 struct in_pack_object *array;
2866 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2868 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2869 in_pack->array[in_pack->nr].object = object;
2874 * Compare the objects in the offset order, in order to emulate the
2875 * "git rev-list --objects" output that produced the pack originally.
2877 static int ofscmp(const void *a_, const void *b_)
2879 struct in_pack_object *a = (struct in_pack_object *)a_;
2880 struct in_pack_object *b = (struct in_pack_object *)b_;
2882 if (a->offset < b->offset)
2884 else if (a->offset > b->offset)
2887 return oidcmp(&a->object->oid, &b->object->oid);
2890 static void add_objects_in_unpacked_packs(void)
2892 struct packed_git *p;
2893 struct in_pack in_pack;
2896 memset(&in_pack, 0, sizeof(in_pack));
2898 for (p = get_all_packs(the_repository); p; p = p->next) {
2899 struct object_id oid;
2902 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)
2904 if (open_pack_index(p))
2905 die(_("cannot open pack index"));
2907 ALLOC_GROW(in_pack.array,
2908 in_pack.nr + p->num_objects,
2911 for (i = 0; i < p->num_objects; i++) {
2912 nth_packed_object_oid(&oid, p, i);
2913 o = lookup_unknown_object(&oid);
2914 if (!(o->flags & OBJECT_ADDED))
2915 mark_in_pack_object(o, p, &in_pack);
2916 o->flags |= OBJECT_ADDED;
2921 QSORT(in_pack.array, in_pack.nr, ofscmp);
2922 for (i = 0; i < in_pack.nr; i++) {
2923 struct object *o = in_pack.array[i].object;
2924 add_object_entry(&o->oid, o->type, "", 0);
2927 free(in_pack.array);
2930 static int add_loose_object(const struct object_id *oid, const char *path,
2933 enum object_type type = oid_object_info(the_repository, oid, NULL);
2936 warning(_("loose object at %s could not be examined"), path);
2940 add_object_entry(oid, type, "", 0);
2945 * We actually don't even have to worry about reachability here.
2946 * add_object_entry will weed out duplicates, so we just add every
2947 * loose object we find.
2949 static void add_unreachable_loose_objects(void)
2951 for_each_loose_file_in_objdir(get_object_directory(),
2956 static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)
2958 static struct packed_git *last_found = (void *)1;
2959 struct packed_git *p;
2961 p = (last_found != (void *)1) ? last_found :
2962 get_all_packs(the_repository);
2965 if ((!p->pack_local || p->pack_keep ||
2966 p->pack_keep_in_core) &&
2967 find_pack_entry_one(oid->hash, p)) {
2971 if (p == last_found)
2972 p = get_all_packs(the_repository);
2975 if (p == last_found)
2982 * Store a list of sha1s that are should not be discarded
2983 * because they are either written too recently, or are
2984 * reachable from another object that was.
2986 * This is filled by get_object_list.
2988 static struct oid_array recent_objects;
2990 static int loosened_object_can_be_discarded(const struct object_id *oid,
2993 if (!unpack_unreachable_expiration)
2995 if (mtime > unpack_unreachable_expiration)
2997 if (oid_array_lookup(&recent_objects, oid) >= 0)
3002 static void loosen_unused_packed_objects(void)
3004 struct packed_git *p;
3006 struct object_id oid;
3008 for (p = get_all_packs(the_repository); p; p = p->next) {
3009 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)
3012 if (open_pack_index(p))
3013 die(_("cannot open pack index"));
3015 for (i = 0; i < p->num_objects; i++) {
3016 nth_packed_object_oid(&oid, p, i);
3017 if (!packlist_find(&to_pack, &oid) &&
3018 !has_sha1_pack_kept_or_nonlocal(&oid) &&
3019 !loosened_object_can_be_discarded(&oid, p->mtime))
3020 if (force_object_loose(&oid, p->mtime))
3021 die(_("unable to force loose object"));
3027 * This tracks any options which pack-reuse code expects to be on, or which a
3028 * reader of the pack might not understand, and which would therefore prevent
3029 * blind reuse of what we have on disk.
3031 static int pack_options_allow_reuse(void)
3033 return pack_to_stdout &&
3035 !ignore_packed_keep_on_disk &&
3036 !ignore_packed_keep_in_core &&
3037 (!local || !have_non_local_packs) &&
3041 static int get_object_list_from_bitmap(struct rev_info *revs)
3043 if (!(bitmap_git = prepare_bitmap_walk(revs)))
3046 if (pack_options_allow_reuse() &&
3047 !reuse_partial_packfile_from_bitmap(
3050 &reuse_packfile_objects,
3051 &reuse_packfile_offset)) {
3052 assert(reuse_packfile_objects);
3053 nr_result += reuse_packfile_objects;
3054 display_progress(progress_state, nr_result);
3057 traverse_bitmap_commit_list(bitmap_git, &add_object_entry_from_bitmap);
3061 static void record_recent_object(struct object *obj,
3065 oid_array_append(&recent_objects, &obj->oid);
3068 static void record_recent_commit(struct commit *commit, void *data)
3070 oid_array_append(&recent_objects, &commit->object.oid);
3073 static void get_object_list(int ac, const char **av)
3075 struct rev_info revs;
3076 struct setup_revision_opt s_r_opt = {
3077 .allow_exclude_promisor_objects = 1,
3083 repo_init_revisions(the_repository, &revs, NULL);
3084 save_commit_buffer = 0;
3085 setup_revisions(ac, av, &revs, &s_r_opt);
3087 /* make sure shallows are read */
3088 is_repository_shallow(the_repository);
3090 save_warning = warn_on_object_refname_ambiguity;
3091 warn_on_object_refname_ambiguity = 0;
3093 while (fgets(line, sizeof(line), stdin) != NULL) {
3094 int len = strlen(line);
3095 if (len && line[len - 1] == '\n')
3100 if (!strcmp(line, "--not")) {
3101 flags ^= UNINTERESTING;
3102 write_bitmap_index = 0;
3105 if (starts_with(line, "--shallow ")) {
3106 struct object_id oid;
3107 if (get_oid_hex(line + 10, &oid))
3108 die("not an SHA-1 '%s'", line + 10);
3109 register_shallow(the_repository, &oid);
3110 use_bitmap_index = 0;
3113 die(_("not a rev '%s'"), line);
3115 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
3116 die(_("bad revision '%s'"), line);
3119 warn_on_object_refname_ambiguity = save_warning;
3121 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
3124 if (use_delta_islands)
3125 load_delta_islands(the_repository, progress);
3127 if (prepare_revision_walk(&revs))
3128 die(_("revision walk setup failed"));
3129 mark_edges_uninteresting(&revs, show_edge, sparse);
3131 if (!fn_show_object)
3132 fn_show_object = show_object;
3133 traverse_commit_list_filtered(&filter_options, &revs,
3134 show_commit, fn_show_object, NULL,
3137 if (unpack_unreachable_expiration) {
3138 revs.ignore_missing_links = 1;
3139 if (add_unseen_recent_objects_to_traversal(&revs,
3140 unpack_unreachable_expiration))
3141 die(_("unable to add recent objects"));
3142 if (prepare_revision_walk(&revs))
3143 die(_("revision walk setup failed"));
3144 traverse_commit_list(&revs, record_recent_commit,
3145 record_recent_object, NULL);
3148 if (keep_unreachable)
3149 add_objects_in_unpacked_packs();
3150 if (pack_loose_unreachable)
3151 add_unreachable_loose_objects();
3152 if (unpack_unreachable)
3153 loosen_unused_packed_objects();
3155 oid_array_clear(&recent_objects);
3158 static void add_extra_kept_packs(const struct string_list *names)
3160 struct packed_git *p;
3165 for (p = get_all_packs(the_repository); p; p = p->next) {
3166 const char *name = basename(p->pack_name);
3172 for (i = 0; i < names->nr; i++)
3173 if (!fspathcmp(name, names->items[i].string))
3176 if (i < names->nr) {
3177 p->pack_keep_in_core = 1;
3178 ignore_packed_keep_in_core = 1;
3184 static int option_parse_index_version(const struct option *opt,
3185 const char *arg, int unset)
3188 const char *val = arg;
3190 BUG_ON_OPT_NEG(unset);
3192 pack_idx_opts.version = strtoul(val, &c, 10);
3193 if (pack_idx_opts.version > 2)
3194 die(_("unsupported index version %s"), val);
3195 if (*c == ',' && c[1])
3196 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
3197 if (*c || pack_idx_opts.off32_limit & 0x80000000)
3198 die(_("bad index version '%s'"), val);
3202 static int option_parse_unpack_unreachable(const struct option *opt,
3203 const char *arg, int unset)
3206 unpack_unreachable = 0;
3207 unpack_unreachable_expiration = 0;
3210 unpack_unreachable = 1;
3212 unpack_unreachable_expiration = approxidate(arg);
3217 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
3219 int use_internal_rev_list = 0;
3221 int all_progress_implied = 0;
3222 struct argv_array rp = ARGV_ARRAY_INIT;
3223 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
3224 int rev_list_index = 0;
3225 struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;
3226 struct option pack_objects_options[] = {
3227 OPT_SET_INT('q', "quiet", &progress,
3228 N_("do not show progress meter"), 0),
3229 OPT_SET_INT(0, "progress", &progress,
3230 N_("show progress meter"), 1),
3231 OPT_SET_INT(0, "all-progress", &progress,
3232 N_("show progress meter during object writing phase"), 2),
3233 OPT_BOOL(0, "all-progress-implied",
3234 &all_progress_implied,
3235 N_("similar to --all-progress when progress meter is shown")),
3236 { OPTION_CALLBACK, 0, "index-version", NULL, N_("<version>[,<offset>]"),
3237 N_("write the pack index file in the specified idx format version"),
3238 PARSE_OPT_NONEG, option_parse_index_version },
3239 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
3240 N_("maximum size of each output pack file")),
3241 OPT_BOOL(0, "local", &local,
3242 N_("ignore borrowed objects from alternate object store")),
3243 OPT_BOOL(0, "incremental", &incremental,
3244 N_("ignore packed objects")),
3245 OPT_INTEGER(0, "window", &window,
3246 N_("limit pack window by objects")),
3247 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
3248 N_("limit pack window by memory in addition to object limit")),
3249 OPT_INTEGER(0, "depth", &depth,
3250 N_("maximum length of delta chain allowed in the resulting pack")),
3251 OPT_BOOL(0, "reuse-delta", &reuse_delta,
3252 N_("reuse existing deltas")),
3253 OPT_BOOL(0, "reuse-object", &reuse_object,
3254 N_("reuse existing objects")),
3255 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
3256 N_("use OFS_DELTA objects")),
3257 OPT_INTEGER(0, "threads", &delta_search_threads,
3258 N_("use threads when searching for best delta matches")),
3259 OPT_BOOL(0, "non-empty", &non_empty,
3260 N_("do not create an empty pack output")),
3261 OPT_BOOL(0, "revs", &use_internal_rev_list,
3262 N_("read revision arguments from standard input")),
3263 OPT_SET_INT_F(0, "unpacked", &rev_list_unpacked,
3264 N_("limit the objects to those that are not yet packed"),
3265 1, PARSE_OPT_NONEG),
3266 OPT_SET_INT_F(0, "all", &rev_list_all,
3267 N_("include objects reachable from any reference"),
3268 1, PARSE_OPT_NONEG),
3269 OPT_SET_INT_F(0, "reflog", &rev_list_reflog,
3270 N_("include objects referred by reflog entries"),
3271 1, PARSE_OPT_NONEG),
3272 OPT_SET_INT_F(0, "indexed-objects", &rev_list_index,
3273 N_("include objects referred to by the index"),
3274 1, PARSE_OPT_NONEG),
3275 OPT_BOOL(0, "stdout", &pack_to_stdout,
3276 N_("output pack to stdout")),
3277 OPT_BOOL(0, "include-tag", &include_tag,
3278 N_("include tag objects that refer to objects to be packed")),
3279 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
3280 N_("keep unreachable objects")),
3281 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
3282 N_("pack loose unreachable objects")),
3283 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
3284 N_("unpack unreachable objects newer than <time>"),
3285 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
3286 OPT_BOOL(0, "sparse", &sparse,
3287 N_("use the sparse reachability algorithm")),
3288 OPT_BOOL(0, "thin", &thin,
3289 N_("create thin packs")),
3290 OPT_BOOL(0, "shallow", &shallow,
3291 N_("create packs suitable for shallow fetches")),
3292 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk,
3293 N_("ignore packs that have companion .keep file")),
3294 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list, N_("name"),
3295 N_("ignore this pack")),
3296 OPT_INTEGER(0, "compression", &pack_compression_level,
3297 N_("pack compression level")),
3298 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
3299 N_("do not hide commits by grafts"), 0),
3300 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
3301 N_("use a bitmap index if available to speed up counting objects")),
3302 OPT_SET_INT(0, "write-bitmap-index", &write_bitmap_index,
3303 N_("write a bitmap index together with the pack index"),
3305 OPT_SET_INT_F(0, "write-bitmap-index-quiet",
3306 &write_bitmap_index,
3307 N_("write a bitmap index if possible"),
3308 WRITE_BITMAP_QUIET, PARSE_OPT_HIDDEN),
3309 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),
3310 { OPTION_CALLBACK, 0, "missing", NULL, N_("action"),
3311 N_("handling for missing objects"), PARSE_OPT_NONEG,
3312 option_parse_missing_action },
3313 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,
3314 N_("do not pack objects in promisor packfiles")),
3315 OPT_BOOL(0, "delta-islands", &use_delta_islands,
3316 N_("respect islands during delta compression")),
3320 if (DFS_NUM_STATES > (1 << OE_DFS_STATE_BITS))
3321 BUG("too many dfs states, increase OE_DFS_STATE_BITS");
3323 read_replace_refs = 0;
3325 sparse = git_env_bool("GIT_TEST_PACK_SPARSE", 0);
3326 prepare_repo_settings(the_repository);
3327 if (!sparse && the_repository->settings.pack_use_sparse != -1)
3328 sparse = the_repository->settings.pack_use_sparse;
3330 reset_pack_idx_option(&pack_idx_opts);
3331 git_config(git_pack_config, NULL);
3333 progress = isatty(2);
3334 argc = parse_options(argc, argv, prefix, pack_objects_options,
3338 base_name = argv[0];
3341 if (pack_to_stdout != !base_name || argc)
3342 usage_with_options(pack_usage, pack_objects_options);
3344 if (depth >= (1 << OE_DEPTH_BITS)) {
3345 warning(_("delta chain depth %d is too deep, forcing %d"),
3346 depth, (1 << OE_DEPTH_BITS) - 1);
3347 depth = (1 << OE_DEPTH_BITS) - 1;
3349 if (cache_max_small_delta_size >= (1U << OE_Z_DELTA_BITS)) {
3350 warning(_("pack.deltaCacheLimit is too high, forcing %d"),
3351 (1U << OE_Z_DELTA_BITS) - 1);
3352 cache_max_small_delta_size = (1U << OE_Z_DELTA_BITS) - 1;
3355 argv_array_push(&rp, "pack-objects");
3357 use_internal_rev_list = 1;
3358 argv_array_push(&rp, shallow
3359 ? "--objects-edge-aggressive"
3360 : "--objects-edge");
3362 argv_array_push(&rp, "--objects");
3365 use_internal_rev_list = 1;
3366 argv_array_push(&rp, "--all");
3368 if (rev_list_reflog) {
3369 use_internal_rev_list = 1;
3370 argv_array_push(&rp, "--reflog");
3372 if (rev_list_index) {
3373 use_internal_rev_list = 1;
3374 argv_array_push(&rp, "--indexed-objects");
3376 if (rev_list_unpacked) {
3377 use_internal_rev_list = 1;
3378 argv_array_push(&rp, "--unpacked");
3381 if (exclude_promisor_objects) {
3382 use_internal_rev_list = 1;
3383 fetch_if_missing = 0;
3384 argv_array_push(&rp, "--exclude-promisor-objects");
3386 if (unpack_unreachable || keep_unreachable || pack_loose_unreachable)
3387 use_internal_rev_list = 1;
3391 if (pack_compression_level == -1)
3392 pack_compression_level = Z_DEFAULT_COMPRESSION;
3393 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
3394 die(_("bad pack compression level %d"), pack_compression_level);
3396 if (!delta_search_threads) /* --threads=0 means autodetect */
3397 delta_search_threads = online_cpus();
3399 if (!HAVE_THREADS && delta_search_threads != 1)
3400 warning(_("no threads support, ignoring --threads"));
3401 if (!pack_to_stdout && !pack_size_limit)
3402 pack_size_limit = pack_size_limit_cfg;
3403 if (pack_to_stdout && pack_size_limit)
3404 die(_("--max-pack-size cannot be used to build a pack for transfer"));
3405 if (pack_size_limit && pack_size_limit < 1024*1024) {
3406 warning(_("minimum pack size limit is 1 MiB"));
3407 pack_size_limit = 1024*1024;
3410 if (!pack_to_stdout && thin)
3411 die(_("--thin cannot be used to build an indexable pack"));
3413 if (keep_unreachable && unpack_unreachable)
3414 die(_("--keep-unreachable and --unpack-unreachable are incompatible"));
3415 if (!rev_list_all || !rev_list_reflog || !rev_list_index)
3416 unpack_unreachable_expiration = 0;
3418 if (filter_options.choice) {
3419 if (!pack_to_stdout)
3420 die(_("cannot use --filter without --stdout"));
3421 use_bitmap_index = 0;
3425 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3427 * - to produce good pack (with bitmap index not-yet-packed objects are
3428 * packed in suboptimal order).
3430 * - to use more robust pack-generation codepath (avoiding possible
3431 * bugs in bitmap code and possible bitmap index corruption).
3433 if (!pack_to_stdout)
3434 use_bitmap_index_default = 0;
3436 if (use_bitmap_index < 0)
3437 use_bitmap_index = use_bitmap_index_default;
3439 /* "hard" reasons not to use bitmaps; these just won't work at all */
3440 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow(the_repository))
3441 use_bitmap_index = 0;
3443 if (pack_to_stdout || !rev_list_all)
3444 write_bitmap_index = 0;
3446 if (use_delta_islands)
3447 argv_array_push(&rp, "--topo-order");
3449 if (progress && all_progress_implied)
3452 add_extra_kept_packs(&keep_pack_list);
3453 if (ignore_packed_keep_on_disk) {
3454 struct packed_git *p;
3455 for (p = get_all_packs(the_repository); p; p = p->next)
3456 if (p->pack_local && p->pack_keep)
3458 if (!p) /* no keep-able packs found */
3459 ignore_packed_keep_on_disk = 0;
3463 * unlike ignore_packed_keep_on_disk above, we do not
3464 * want to unset "local" based on looking at packs, as
3465 * it also covers non-local objects
3467 struct packed_git *p;
3468 for (p = get_all_packs(the_repository); p; p = p->next) {
3469 if (!p->pack_local) {
3470 have_non_local_packs = 1;
3476 trace2_region_enter("pack-objects", "enumerate-objects",
3478 prepare_packing_data(the_repository, &to_pack);
3481 progress_state = start_progress(_("Enumerating objects"), 0);
3482 if (!use_internal_rev_list)
3483 read_object_list_from_stdin();
3485 get_object_list(rp.argc, rp.argv);
3486 argv_array_clear(&rp);
3488 cleanup_preferred_base();
3489 if (include_tag && nr_result)
3490 for_each_ref(add_ref_tag, NULL);
3491 stop_progress(&progress_state);
3492 trace2_region_leave("pack-objects", "enumerate-objects",
3495 if (non_empty && !nr_result)
3498 trace2_region_enter("pack-objects", "prepare-pack",
3500 prepare_pack(window, depth);
3501 trace2_region_leave("pack-objects", "prepare-pack",
3505 trace2_region_enter("pack-objects", "write-pack-file", the_repository);
3507 trace2_region_leave("pack-objects", "write-pack-file", the_repository);
3511 _("Total %"PRIu32" (delta %"PRIu32"),"
3512 " reused %"PRIu32" (delta %"PRIu32")"),
3513 written, written_delta, reused, reused_delta);