4 #include "environment.h"
7 #include "parse-options.h"
9 #include "run-command.h"
10 #include "server-info.h"
12 #include "string-list.h"
16 #include "prune-packed.h"
17 #include "object-store-ll.h"
18 #include "promisor-remote.h"
21 #include "pack-bitmap.h"
23 #include "list-objects-filter-options.h"
25 #define ALL_INTO_ONE 1
26 #define LOOSEN_UNREACHABLE 2
32 static int pack_everything
;
33 static int delta_base_offset
= 1;
34 static int pack_kept_objects
= -1;
35 static int write_bitmaps
= -1;
36 static int use_delta_islands
;
37 static int run_update_server_info
= 1;
38 static char *packdir
, *packtmp_name
, *packtmp
;
40 static const char *const git_repack_usage
[] = {
41 N_("git repack [<options>]"),
45 static const char incremental_bitmap_conflict_error
[] = N_(
46 "Incremental repacks are incompatible with bitmap indexes. Use\n"
47 "--no-write-bitmap-index or disable the pack.writeBitmaps configuration."
50 struct pack_objects_args
{
52 const char *window_memory
;
55 unsigned long max_pack_size
;
60 struct list_objects_filter_options filter_options
;
63 static int repack_config(const char *var
, const char *value
,
64 const struct config_context
*ctx
, void *cb
)
66 struct pack_objects_args
*cruft_po_args
= cb
;
67 if (!strcmp(var
, "repack.usedeltabaseoffset")) {
68 delta_base_offset
= git_config_bool(var
, value
);
71 if (!strcmp(var
, "repack.packkeptobjects")) {
72 pack_kept_objects
= git_config_bool(var
, value
);
75 if (!strcmp(var
, "repack.writebitmaps") ||
76 !strcmp(var
, "pack.writebitmaps")) {
77 write_bitmaps
= git_config_bool(var
, value
);
80 if (!strcmp(var
, "repack.usedeltaislands")) {
81 use_delta_islands
= git_config_bool(var
, value
);
84 if (strcmp(var
, "repack.updateserverinfo") == 0) {
85 run_update_server_info
= git_config_bool(var
, value
);
88 if (!strcmp(var
, "repack.cruftwindow"))
89 return git_config_string(&cruft_po_args
->window
, var
, value
);
90 if (!strcmp(var
, "repack.cruftwindowmemory"))
91 return git_config_string(&cruft_po_args
->window_memory
, var
, value
);
92 if (!strcmp(var
, "repack.cruftdepth"))
93 return git_config_string(&cruft_po_args
->depth
, var
, value
);
94 if (!strcmp(var
, "repack.cruftthreads"))
95 return git_config_string(&cruft_po_args
->threads
, var
, value
);
96 return git_default_config(var
, value
, ctx
, cb
);
99 struct existing_packs
{
100 struct string_list kept_packs
;
101 struct string_list non_kept_packs
;
102 struct string_list cruft_packs
;
105 #define EXISTING_PACKS_INIT { \
106 .kept_packs = STRING_LIST_INIT_DUP, \
107 .non_kept_packs = STRING_LIST_INIT_DUP, \
108 .cruft_packs = STRING_LIST_INIT_DUP, \
111 static int has_existing_non_kept_packs(const struct existing_packs
*existing
)
113 return existing
->non_kept_packs
.nr
|| existing
->cruft_packs
.nr
;
116 static void pack_mark_for_deletion(struct string_list_item
*item
)
118 item
->util
= (void*)((uintptr_t)item
->util
| DELETE_PACK
);
121 static void pack_unmark_for_deletion(struct string_list_item
*item
)
123 item
->util
= (void*)((uintptr_t)item
->util
& ~DELETE_PACK
);
126 static int pack_is_marked_for_deletion(struct string_list_item
*item
)
128 return (uintptr_t)item
->util
& DELETE_PACK
;
131 static void pack_mark_retained(struct string_list_item
*item
)
133 item
->util
= (void*)((uintptr_t)item
->util
| RETAIN_PACK
);
136 static int pack_is_retained(struct string_list_item
*item
)
138 return (uintptr_t)item
->util
& RETAIN_PACK
;
141 static void mark_packs_for_deletion_1(struct string_list
*names
,
142 struct string_list
*list
)
144 struct string_list_item
*item
;
145 const int hexsz
= the_hash_algo
->hexsz
;
147 for_each_string_list_item(item
, list
) {
149 size_t len
= strlen(item
->string
);
152 sha1
= item
->string
+ len
- hexsz
;
154 if (pack_is_retained(item
)) {
155 pack_unmark_for_deletion(item
);
156 } else if (!string_list_has_string(names
, sha1
)) {
158 * Mark this pack for deletion, which ensures
159 * that this pack won't be included in a MIDX
160 * (if `--write-midx` was given) and that we
161 * will actually delete this pack (if `-d` was
164 pack_mark_for_deletion(item
);
169 static void retain_cruft_pack(struct existing_packs
*existing
,
170 struct packed_git
*cruft
)
172 struct strbuf buf
= STRBUF_INIT
;
173 struct string_list_item
*item
;
175 strbuf_addstr(&buf
, pack_basename(cruft
));
176 strbuf_strip_suffix(&buf
, ".pack");
178 item
= string_list_lookup(&existing
->cruft_packs
, buf
.buf
);
180 BUG("could not find cruft pack '%s'", pack_basename(cruft
));
182 pack_mark_retained(item
);
183 strbuf_release(&buf
);
186 static void mark_packs_for_deletion(struct existing_packs
*existing
,
187 struct string_list
*names
)
190 mark_packs_for_deletion_1(names
, &existing
->non_kept_packs
);
191 mark_packs_for_deletion_1(names
, &existing
->cruft_packs
);
194 static void remove_redundant_pack(const char *dir_name
, const char *base_name
)
196 struct strbuf buf
= STRBUF_INIT
;
197 struct multi_pack_index
*m
= get_local_multi_pack_index(the_repository
);
198 strbuf_addf(&buf
, "%s.pack", base_name
);
199 if (m
&& midx_contains_pack(m
, buf
.buf
))
200 clear_midx_file(the_repository
);
201 strbuf_insertf(&buf
, 0, "%s/", dir_name
);
202 unlink_pack_path(buf
.buf
, 1);
203 strbuf_release(&buf
);
206 static void remove_redundant_packs_1(struct string_list
*packs
)
208 struct string_list_item
*item
;
209 for_each_string_list_item(item
, packs
) {
210 if (!pack_is_marked_for_deletion(item
))
212 remove_redundant_pack(packdir
, item
->string
);
216 static void remove_redundant_existing_packs(struct existing_packs
*existing
)
218 remove_redundant_packs_1(&existing
->non_kept_packs
);
219 remove_redundant_packs_1(&existing
->cruft_packs
);
222 static void existing_packs_release(struct existing_packs
*existing
)
224 string_list_clear(&existing
->kept_packs
, 0);
225 string_list_clear(&existing
->non_kept_packs
, 0);
226 string_list_clear(&existing
->cruft_packs
, 0);
230 * Adds all packs hex strings (pack-$HASH) to either packs->non_kept
231 * or packs->kept based on whether each pack has a corresponding
232 * .keep file or not. Packs without a .keep file are not to be kept
233 * if we are going to pack everything into one file.
235 static void collect_pack_filenames(struct existing_packs
*existing
,
236 const struct string_list
*extra_keep
)
238 struct packed_git
*p
;
239 struct strbuf buf
= STRBUF_INIT
;
241 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
248 base
= pack_basename(p
);
250 for (i
= 0; i
< extra_keep
->nr
; i
++)
251 if (!fspathcmp(base
, extra_keep
->items
[i
].string
))
255 strbuf_addstr(&buf
, base
);
256 strbuf_strip_suffix(&buf
, ".pack");
258 if ((extra_keep
->nr
> 0 && i
< extra_keep
->nr
) || p
->pack_keep
)
259 string_list_append(&existing
->kept_packs
, buf
.buf
);
260 else if (p
->is_cruft
)
261 string_list_append(&existing
->cruft_packs
, buf
.buf
);
263 string_list_append(&existing
->non_kept_packs
, buf
.buf
);
266 string_list_sort(&existing
->kept_packs
);
267 string_list_sort(&existing
->non_kept_packs
);
268 string_list_sort(&existing
->cruft_packs
);
269 strbuf_release(&buf
);
272 static void prepare_pack_objects(struct child_process
*cmd
,
273 const struct pack_objects_args
*args
,
276 strvec_push(&cmd
->args
, "pack-objects");
278 strvec_pushf(&cmd
->args
, "--window=%s", args
->window
);
279 if (args
->window_memory
)
280 strvec_pushf(&cmd
->args
, "--window-memory=%s", args
->window_memory
);
282 strvec_pushf(&cmd
->args
, "--depth=%s", args
->depth
);
284 strvec_pushf(&cmd
->args
, "--threads=%s", args
->threads
);
285 if (args
->max_pack_size
)
286 strvec_pushf(&cmd
->args
, "--max-pack-size=%lu", args
->max_pack_size
);
287 if (args
->no_reuse_delta
)
288 strvec_pushf(&cmd
->args
, "--no-reuse-delta");
289 if (args
->no_reuse_object
)
290 strvec_pushf(&cmd
->args
, "--no-reuse-object");
292 strvec_push(&cmd
->args
, "--local");
294 strvec_push(&cmd
->args
, "--quiet");
295 if (delta_base_offset
)
296 strvec_push(&cmd
->args
, "--delta-base-offset");
297 strvec_push(&cmd
->args
, out
);
303 * Write oid to the given struct child_process's stdin, starting it first if
306 static int write_oid(const struct object_id
*oid
,
307 struct packed_git
*pack UNUSED
,
308 uint32_t pos UNUSED
, void *data
)
310 struct child_process
*cmd
= data
;
313 if (start_command(cmd
))
314 die(_("could not start pack-objects to repack promisor objects"));
317 xwrite(cmd
->in
, oid_to_hex(oid
), the_hash_algo
->hexsz
);
318 xwrite(cmd
->in
, "\n", 1);
334 struct generated_pack_data
{
335 struct tempfile
*tempfiles
[ARRAY_SIZE(exts
)];
338 static struct generated_pack_data
*populate_pack_exts(const char *name
)
341 struct strbuf path
= STRBUF_INIT
;
342 struct generated_pack_data
*data
= xcalloc(1, sizeof(*data
));
345 for (i
= 0; i
< ARRAY_SIZE(exts
); i
++) {
347 strbuf_addf(&path
, "%s-%s%s", packtmp
, name
, exts
[i
].name
);
349 if (stat(path
.buf
, &statbuf
))
352 data
->tempfiles
[i
] = register_tempfile(path
.buf
);
355 strbuf_release(&path
);
359 static int has_pack_ext(const struct generated_pack_data
*data
,
363 for (i
= 0; i
< ARRAY_SIZE(exts
); i
++) {
364 if (strcmp(exts
[i
].name
, ext
))
366 return !!data
->tempfiles
[i
];
368 BUG("unknown pack extension: '%s'", ext
);
371 static void repack_promisor_objects(const struct pack_objects_args
*args
,
372 struct string_list
*names
)
374 struct child_process cmd
= CHILD_PROCESS_INIT
;
376 struct strbuf line
= STRBUF_INIT
;
378 prepare_pack_objects(&cmd
, args
, packtmp
);
382 * NEEDSWORK: Giving pack-objects only the OIDs without any ordering
383 * hints may result in suboptimal deltas in the resulting pack. See if
384 * the OIDs can be sent with fake paths such that pack-objects can use a
385 * {type -> existing pack order} ordering when computing deltas instead
386 * of a {type -> size} ordering, which may produce better deltas.
388 for_each_packed_object(write_oid
, &cmd
,
389 FOR_EACH_OBJECT_PROMISOR_ONLY
);
392 /* No packed objects; cmd was never started */
393 child_process_clear(&cmd
);
399 out
= xfdopen(cmd
.out
, "r");
400 while (strbuf_getline_lf(&line
, out
) != EOF
) {
401 struct string_list_item
*item
;
404 if (line
.len
!= the_hash_algo
->hexsz
)
405 die(_("repack: Expecting full hex object ID lines only from pack-objects."));
406 item
= string_list_append(names
, line
.buf
);
409 * pack-objects creates the .pack and .idx files, but not the
410 * .promisor file. Create the .promisor file, which is empty.
412 * NEEDSWORK: fetch-pack sometimes generates non-empty
413 * .promisor files containing the ref names and associated
414 * hashes at the point of generation of the corresponding
415 * packfile, but this would not preserve their contents. Maybe
416 * concatenate the contents of all .promisor files instead of
417 * just creating a new empty file.
419 promisor_name
= mkpathdup("%s-%s.promisor", packtmp
,
421 write_promisor_file(promisor_name
, NULL
, 0);
423 item
->util
= populate_pack_exts(item
->string
);
428 if (finish_command(&cmd
))
429 die(_("could not finish pack-objects to repack promisor objects"));
432 struct pack_geometry
{
433 struct packed_git
**pack
;
434 uint32_t pack_nr
, pack_alloc
;
440 static uint32_t geometry_pack_weight(struct packed_git
*p
)
442 if (open_pack_index(p
))
443 die(_("cannot open index for %s"), p
->pack_name
);
444 return p
->num_objects
;
447 static int geometry_cmp(const void *va
, const void *vb
)
449 uint32_t aw
= geometry_pack_weight(*(struct packed_git
**)va
),
450 bw
= geometry_pack_weight(*(struct packed_git
**)vb
);
459 static void init_pack_geometry(struct pack_geometry
*geometry
,
460 struct existing_packs
*existing
,
461 const struct pack_objects_args
*args
)
463 struct packed_git
*p
;
464 struct strbuf buf
= STRBUF_INIT
;
466 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
467 if (args
->local
&& !p
->pack_local
)
469 * When asked to only repack local packfiles we skip
470 * over any packfiles that are borrowed from alternate
471 * object directories.
475 if (!pack_kept_objects
) {
477 * Any pack that has its pack_keep bit set will
478 * appear in existing->kept_packs below, but
479 * this saves us from doing a more expensive
486 * The pack may be kept via the --keep-pack
487 * option; check 'existing->kept_packs' to
488 * determine whether to ignore it.
491 strbuf_addstr(&buf
, pack_basename(p
));
492 strbuf_strip_suffix(&buf
, ".pack");
494 if (string_list_has_string(&existing
->kept_packs
, buf
.buf
))
500 ALLOC_GROW(geometry
->pack
,
501 geometry
->pack_nr
+ 1,
502 geometry
->pack_alloc
);
504 geometry
->pack
[geometry
->pack_nr
] = p
;
508 QSORT(geometry
->pack
, geometry
->pack_nr
, geometry_cmp
);
509 strbuf_release(&buf
);
512 static void split_pack_geometry(struct pack_geometry
*geometry
)
516 off_t total_size
= 0;
518 if (!geometry
->pack_nr
) {
519 geometry
->split
= geometry
->pack_nr
;
524 * First, count the number of packs (in descending order of size) which
525 * already form a geometric progression.
527 for (i
= geometry
->pack_nr
- 1; i
> 0; i
--) {
528 struct packed_git
*ours
= geometry
->pack
[i
];
529 struct packed_git
*prev
= geometry
->pack
[i
- 1];
531 if (unsigned_mult_overflows(geometry
->split_factor
,
532 geometry_pack_weight(prev
)))
533 die(_("pack %s too large to consider in geometric "
537 if (geometry_pack_weight(ours
) <
538 geometry
->split_factor
* geometry_pack_weight(prev
))
546 * Move the split one to the right, since the top element in the
547 * last-compared pair can't be in the progression. Only do this
548 * when we split in the middle of the array (otherwise if we got
549 * to the end, then the split is in the right place).
555 * Then, anything to the left of 'split' must be in a new pack. But,
556 * creating that new pack may cause packs in the heavy half to no longer
557 * form a geometric progression.
559 * Compute an expected size of the new pack, and then determine how many
560 * packs in the heavy half need to be joined into it (if any) to restore
561 * the geometric progression.
563 for (i
= 0; i
< split
; i
++) {
564 struct packed_git
*p
= geometry
->pack
[i
];
566 if (unsigned_add_overflows(total_size
, geometry_pack_weight(p
)))
567 die(_("pack %s too large to roll up"), p
->pack_name
);
568 total_size
+= geometry_pack_weight(p
);
570 for (i
= split
; i
< geometry
->pack_nr
; i
++) {
571 struct packed_git
*ours
= geometry
->pack
[i
];
573 if (unsigned_mult_overflows(geometry
->split_factor
,
575 die(_("pack %s too large to roll up"), ours
->pack_name
);
577 if (geometry_pack_weight(ours
) <
578 geometry
->split_factor
* total_size
) {
579 if (unsigned_add_overflows(total_size
,
580 geometry_pack_weight(ours
)))
581 die(_("pack %s too large to roll up"),
585 total_size
+= geometry_pack_weight(ours
);
590 geometry
->split
= split
;
593 static struct packed_git
*get_preferred_pack(struct pack_geometry
*geometry
)
599 * No geometry means either an all-into-one repack (in which
600 * case there is only one pack left and it is the largest) or an
603 * If repacking incrementally, then we could check the size of
604 * all packs to determine which should be preferred, but leave
609 if (geometry
->split
== geometry
->pack_nr
)
613 * The preferred pack is the largest pack above the split line. In
614 * other words, it is the largest pack that does not get rolled up in
615 * the geometric repack.
617 for (i
= geometry
->pack_nr
; i
> geometry
->split
; i
--)
619 * A pack that is not local would never be included in a
620 * multi-pack index. We thus skip over any non-local packs.
622 if (geometry
->pack
[i
- 1]->pack_local
)
623 return geometry
->pack
[i
- 1];
628 static void geometry_remove_redundant_packs(struct pack_geometry
*geometry
,
629 struct string_list
*names
,
630 struct existing_packs
*existing
)
632 struct strbuf buf
= STRBUF_INIT
;
635 for (i
= 0; i
< geometry
->split
; i
++) {
636 struct packed_git
*p
= geometry
->pack
[i
];
637 if (string_list_has_string(names
, hash_to_hex(p
->hash
)))
641 strbuf_addstr(&buf
, pack_basename(p
));
642 strbuf_strip_suffix(&buf
, ".pack");
644 if ((p
->pack_keep
) ||
645 (string_list_has_string(&existing
->kept_packs
, buf
.buf
)))
648 remove_redundant_pack(packdir
, buf
.buf
);
651 strbuf_release(&buf
);
654 static void free_pack_geometry(struct pack_geometry
*geometry
)
659 free(geometry
->pack
);
662 struct midx_snapshot_ref_data
{
668 static int midx_snapshot_ref_one(const char *refname UNUSED
,
669 const struct object_id
*oid
,
670 int flag UNUSED
, void *_data
)
672 struct midx_snapshot_ref_data
*data
= _data
;
673 struct object_id peeled
;
675 if (!peel_iterated_oid(oid
, &peeled
))
678 if (oidset_insert(&data
->seen
, oid
))
679 return 0; /* already seen */
681 if (oid_object_info(the_repository
, oid
, NULL
) != OBJ_COMMIT
)
684 fprintf(data
->f
->fp
, "%s%s\n", data
->preferred
? "+" : "",
690 static void midx_snapshot_refs(struct tempfile
*f
)
692 struct midx_snapshot_ref_data data
;
693 const struct string_list
*preferred
= bitmap_preferred_tips(the_repository
);
697 oidset_init(&data
.seen
, 0);
699 if (!fdopen_tempfile(f
, "w"))
700 die(_("could not open tempfile %s for writing"),
701 get_tempfile_path(f
));
704 struct string_list_item
*item
;
707 for_each_string_list_item(item
, preferred
)
708 for_each_ref_in(item
->string
, midx_snapshot_ref_one
, &data
);
712 for_each_ref(midx_snapshot_ref_one
, &data
);
714 if (close_tempfile_gently(f
)) {
715 int save_errno
= errno
;
718 die_errno(_("could not close refs snapshot tempfile"));
721 oidset_clear(&data
.seen
);
724 static void midx_included_packs(struct string_list
*include
,
725 struct existing_packs
*existing
,
726 struct string_list
*names
,
727 struct pack_geometry
*geometry
)
729 struct string_list_item
*item
;
731 for_each_string_list_item(item
, &existing
->kept_packs
)
732 string_list_insert(include
, xstrfmt("%s.idx", item
->string
));
733 for_each_string_list_item(item
, names
)
734 string_list_insert(include
, xstrfmt("pack-%s.idx", item
->string
));
735 if (geometry
->split_factor
) {
736 struct strbuf buf
= STRBUF_INIT
;
738 for (i
= geometry
->split
; i
< geometry
->pack_nr
; i
++) {
739 struct packed_git
*p
= geometry
->pack
[i
];
742 * The multi-pack index never refers to packfiles part
743 * of an alternate object database, so we skip these.
744 * While git-multi-pack-index(1) would silently ignore
745 * them anyway, this allows us to skip executing the
746 * command completely when we have only non-local
752 strbuf_addstr(&buf
, pack_basename(p
));
753 strbuf_strip_suffix(&buf
, ".pack");
754 strbuf_addstr(&buf
, ".idx");
756 string_list_insert(include
, strbuf_detach(&buf
, NULL
));
759 for_each_string_list_item(item
, &existing
->non_kept_packs
) {
760 if (pack_is_marked_for_deletion(item
))
762 string_list_insert(include
, xstrfmt("%s.idx", item
->string
));
766 for_each_string_list_item(item
, &existing
->cruft_packs
) {
768 * When doing a --geometric repack, there is no need to check
769 * for deleted packs, since we're by definition not doing an
770 * ALL_INTO_ONE repack (hence no packs will be deleted).
771 * Otherwise we must check for and exclude any packs which are
772 * enqueued for deletion.
774 * So we could omit the conditional below in the --geometric
775 * case, but doing so is unnecessary since no packs are marked
776 * as pending deletion (since we only call
777 * `mark_packs_for_deletion()` when doing an all-into-one
780 if (pack_is_marked_for_deletion(item
))
782 string_list_insert(include
, xstrfmt("%s.idx", item
->string
));
786 static int write_midx_included_packs(struct string_list
*include
,
787 struct pack_geometry
*geometry
,
788 struct string_list
*names
,
789 const char *refs_snapshot
,
790 int show_progress
, int write_bitmaps
)
792 struct child_process cmd
= CHILD_PROCESS_INIT
;
793 struct string_list_item
*item
;
794 struct packed_git
*preferred
= get_preferred_pack(geometry
);
804 strvec_push(&cmd
.args
, "multi-pack-index");
805 strvec_pushl(&cmd
.args
, "write", "--stdin-packs", NULL
);
808 strvec_push(&cmd
.args
, "--progress");
810 strvec_push(&cmd
.args
, "--no-progress");
813 strvec_push(&cmd
.args
, "--bitmap");
816 strvec_pushf(&cmd
.args
, "--preferred-pack=%s",
817 pack_basename(preferred
));
818 else if (names
->nr
) {
819 /* The largest pack was repacked, meaning that either
820 * one or two packs exist depending on whether the
821 * repository has a cruft pack or not.
823 * Select the non-cruft one as preferred to encourage
824 * pack-reuse among packs containing reachable objects
825 * over unreachable ones.
827 * (Note we could write multiple packs here if
828 * `--max-pack-size` was given, but any one of them
829 * will suffice, so pick the first one.)
831 for_each_string_list_item(item
, names
) {
832 struct generated_pack_data
*data
= item
->util
;
833 if (has_pack_ext(data
, ".mtimes"))
836 strvec_pushf(&cmd
.args
, "--preferred-pack=pack-%s.pack",
842 * No packs were kept, and no packs were written. The
843 * only thing remaining are .keep packs (unless
844 * --pack-kept-objects was given).
846 * Set the `--preferred-pack` arbitrarily here.
852 strvec_pushf(&cmd
.args
, "--refs-snapshot=%s", refs_snapshot
);
854 ret
= start_command(&cmd
);
858 in
= xfdopen(cmd
.in
, "w");
859 for_each_string_list_item(item
, include
)
860 fprintf(in
, "%s\n", item
->string
);
863 return finish_command(&cmd
);
866 static void remove_redundant_bitmaps(struct string_list
*include
,
869 struct strbuf path
= STRBUF_INIT
;
870 struct string_list_item
*item
;
873 strbuf_addstr(&path
, packdir
);
874 strbuf_addch(&path
, '/');
875 packdir_len
= path
.len
;
878 * Remove any pack bitmaps corresponding to packs which are now
879 * included in the MIDX.
881 for_each_string_list_item(item
, include
) {
882 strbuf_addstr(&path
, item
->string
);
883 strbuf_strip_suffix(&path
, ".idx");
884 strbuf_addstr(&path
, ".bitmap");
886 if (unlink(path
.buf
) && errno
!= ENOENT
)
887 warning_errno(_("could not remove stale bitmap: %s"),
890 strbuf_setlen(&path
, packdir_len
);
892 strbuf_release(&path
);
895 static int finish_pack_objects_cmd(struct child_process
*cmd
,
896 struct string_list
*names
,
900 struct strbuf line
= STRBUF_INIT
;
902 out
= xfdopen(cmd
->out
, "r");
903 while (strbuf_getline_lf(&line
, out
) != EOF
) {
904 struct string_list_item
*item
;
906 if (line
.len
!= the_hash_algo
->hexsz
)
907 die(_("repack: Expecting full hex object ID lines only "
908 "from pack-objects."));
910 * Avoid putting packs written outside of the repository in the
914 item
= string_list_append(names
, line
.buf
);
915 item
->util
= populate_pack_exts(line
.buf
);
920 strbuf_release(&line
);
922 return finish_command(cmd
);
925 static int write_filtered_pack(const struct pack_objects_args
*args
,
926 const char *destination
,
927 const char *pack_prefix
,
928 struct existing_packs
*existing
,
929 struct string_list
*names
)
931 struct child_process cmd
= CHILD_PROCESS_INIT
;
932 struct string_list_item
*item
;
937 int local
= skip_prefix(destination
, packdir
, &scratch
);
939 prepare_pack_objects(&cmd
, args
, destination
);
941 strvec_push(&cmd
.args
, "--stdin-packs");
943 if (!pack_kept_objects
)
944 strvec_push(&cmd
.args
, "--honor-pack-keep");
945 for_each_string_list_item(item
, &existing
->kept_packs
)
946 strvec_pushf(&cmd
.args
, "--keep-pack=%s", item
->string
);
950 ret
= start_command(&cmd
);
955 * Here 'names' contains only the pack(s) that were just
956 * written, which is exactly the packs we want to keep. Also
957 * 'existing_kept_packs' already contains the packs in
960 in
= xfdopen(cmd
.in
, "w");
961 for_each_string_list_item(item
, names
)
962 fprintf(in
, "^%s-%s.pack\n", pack_prefix
, item
->string
);
963 for_each_string_list_item(item
, &existing
->non_kept_packs
)
964 fprintf(in
, "%s.pack\n", item
->string
);
965 for_each_string_list_item(item
, &existing
->cruft_packs
)
966 fprintf(in
, "%s.pack\n", item
->string
);
967 caret
= pack_kept_objects
? "" : "^";
968 for_each_string_list_item(item
, &existing
->kept_packs
)
969 fprintf(in
, "%s%s.pack\n", caret
, item
->string
);
972 return finish_pack_objects_cmd(&cmd
, names
, local
);
975 static int existing_cruft_pack_cmp(const void *va
, const void *vb
)
977 struct packed_git
*a
= *(struct packed_git
**)va
;
978 struct packed_git
*b
= *(struct packed_git
**)vb
;
980 if (a
->pack_size
< b
->pack_size
)
982 if (a
->pack_size
> b
->pack_size
)
987 static void collapse_small_cruft_packs(FILE *in
, size_t max_size
,
988 struct existing_packs
*existing
)
990 struct packed_git
**existing_cruft
, *p
;
991 struct strbuf buf
= STRBUF_INIT
;
992 size_t total_size
= 0;
993 size_t existing_cruft_nr
= 0;
996 ALLOC_ARRAY(existing_cruft
, existing
->cruft_packs
.nr
);
998 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
999 if (!(p
->is_cruft
&& p
->pack_local
))
1003 strbuf_addstr(&buf
, pack_basename(p
));
1004 strbuf_strip_suffix(&buf
, ".pack");
1006 if (!string_list_has_string(&existing
->cruft_packs
, buf
.buf
))
1009 if (existing_cruft_nr
>= existing
->cruft_packs
.nr
)
1010 BUG("too many cruft packs (found %"PRIuMAX
", but knew "
1012 (uintmax_t)existing_cruft_nr
+ 1,
1013 (uintmax_t)existing
->cruft_packs
.nr
);
1014 existing_cruft
[existing_cruft_nr
++] = p
;
1017 QSORT(existing_cruft
, existing_cruft_nr
, existing_cruft_pack_cmp
);
1019 for (i
= 0; i
< existing_cruft_nr
; i
++) {
1022 p
= existing_cruft
[i
];
1023 proposed
= st_add(total_size
, p
->pack_size
);
1025 if (proposed
<= max_size
) {
1026 total_size
= proposed
;
1027 fprintf(in
, "-%s\n", pack_basename(p
));
1029 retain_cruft_pack(existing
, p
);
1030 fprintf(in
, "%s\n", pack_basename(p
));
1034 for (i
= 0; i
< existing
->non_kept_packs
.nr
; i
++)
1035 fprintf(in
, "-%s.pack\n",
1036 existing
->non_kept_packs
.items
[i
].string
);
1038 strbuf_release(&buf
);
1039 free(existing_cruft
);
1042 static int write_cruft_pack(const struct pack_objects_args
*args
,
1043 const char *destination
,
1044 const char *pack_prefix
,
1045 const char *cruft_expiration
,
1046 struct string_list
*names
,
1047 struct existing_packs
*existing
)
1049 struct child_process cmd
= CHILD_PROCESS_INIT
;
1050 struct string_list_item
*item
;
1053 const char *scratch
;
1054 int local
= skip_prefix(destination
, packdir
, &scratch
);
1056 prepare_pack_objects(&cmd
, args
, destination
);
1058 strvec_push(&cmd
.args
, "--cruft");
1059 if (cruft_expiration
)
1060 strvec_pushf(&cmd
.args
, "--cruft-expiration=%s",
1063 strvec_push(&cmd
.args
, "--honor-pack-keep");
1064 strvec_push(&cmd
.args
, "--non-empty");
1068 ret
= start_command(&cmd
);
1073 * names has a confusing double use: it both provides the list
1074 * of just-written new packs, and accepts the name of the cruft
1075 * pack we are writing.
1077 * By the time it is read here, it contains only the pack(s)
1078 * that were just written, which is exactly the set of packs we
1079 * want to consider kept.
1081 * If `--expire-to` is given, the double-use served by `names`
1082 * ensures that the pack written to `--expire-to` excludes any
1083 * objects contained in the cruft pack.
1085 in
= xfdopen(cmd
.in
, "w");
1086 for_each_string_list_item(item
, names
)
1087 fprintf(in
, "%s-%s.pack\n", pack_prefix
, item
->string
);
1088 if (args
->max_pack_size
&& !cruft_expiration
) {
1089 collapse_small_cruft_packs(in
, args
->max_pack_size
, existing
);
1091 for_each_string_list_item(item
, &existing
->non_kept_packs
)
1092 fprintf(in
, "-%s.pack\n", item
->string
);
1093 for_each_string_list_item(item
, &existing
->cruft_packs
)
1094 fprintf(in
, "-%s.pack\n", item
->string
);
1096 for_each_string_list_item(item
, &existing
->kept_packs
)
1097 fprintf(in
, "%s.pack\n", item
->string
);
1100 return finish_pack_objects_cmd(&cmd
, names
, local
);
1103 static const char *find_pack_prefix(const char *packdir
, const char *packtmp
)
1105 const char *pack_prefix
;
1106 if (!skip_prefix(packtmp
, packdir
, &pack_prefix
))
1107 die(_("pack prefix %s does not begin with objdir %s"),
1109 if (*pack_prefix
== '/')
1114 int cmd_repack(int argc
, const char **argv
, const char *prefix
)
1116 struct child_process cmd
= CHILD_PROCESS_INIT
;
1117 struct string_list_item
*item
;
1118 struct string_list names
= STRING_LIST_INIT_DUP
;
1119 struct existing_packs existing
= EXISTING_PACKS_INIT
;
1120 struct pack_geometry geometry
= { 0 };
1121 struct tempfile
*refs_snapshot
= NULL
;
1125 /* variables to be filled by option parsing */
1126 int delete_redundant
= 0;
1127 const char *unpack_unreachable
= NULL
;
1128 int keep_unreachable
= 0;
1129 struct string_list keep_pack_list
= STRING_LIST_INIT_NODUP
;
1130 struct pack_objects_args po_args
= {NULL
};
1131 struct pack_objects_args cruft_po_args
= {NULL
};
1133 const char *cruft_expiration
= NULL
;
1134 const char *expire_to
= NULL
;
1135 const char *filter_to
= NULL
;
1137 struct option builtin_repack_options
[] = {
1138 OPT_BIT('a', NULL
, &pack_everything
,
1139 N_("pack everything in a single pack"), ALL_INTO_ONE
),
1140 OPT_BIT('A', NULL
, &pack_everything
,
1141 N_("same as -a, and turn unreachable objects loose"),
1142 LOOSEN_UNREACHABLE
| ALL_INTO_ONE
),
1143 OPT_BIT(0, "cruft", &pack_everything
,
1144 N_("same as -a, pack unreachable cruft objects separately"),
1146 OPT_STRING(0, "cruft-expiration", &cruft_expiration
, N_("approxidate"),
1147 N_("with --cruft, expire objects older than this")),
1148 OPT_MAGNITUDE(0, "max-cruft-size", &cruft_po_args
.max_pack_size
,
1149 N_("with --cruft, limit the size of new cruft packs")),
1150 OPT_BOOL('d', NULL
, &delete_redundant
,
1151 N_("remove redundant packs, and run git-prune-packed")),
1152 OPT_BOOL('f', NULL
, &po_args
.no_reuse_delta
,
1153 N_("pass --no-reuse-delta to git-pack-objects")),
1154 OPT_BOOL('F', NULL
, &po_args
.no_reuse_object
,
1155 N_("pass --no-reuse-object to git-pack-objects")),
1156 OPT_NEGBIT('n', NULL
, &run_update_server_info
,
1157 N_("do not run git-update-server-info"), 1),
1158 OPT__QUIET(&po_args
.quiet
, N_("be quiet")),
1159 OPT_BOOL('l', "local", &po_args
.local
,
1160 N_("pass --local to git-pack-objects")),
1161 OPT_BOOL('b', "write-bitmap-index", &write_bitmaps
,
1162 N_("write bitmap index")),
1163 OPT_BOOL('i', "delta-islands", &use_delta_islands
,
1164 N_("pass --delta-islands to git-pack-objects")),
1165 OPT_STRING(0, "unpack-unreachable", &unpack_unreachable
, N_("approxidate"),
1166 N_("with -A, do not loosen objects older than this")),
1167 OPT_BOOL('k', "keep-unreachable", &keep_unreachable
,
1168 N_("with -a, repack unreachable objects")),
1169 OPT_STRING(0, "window", &po_args
.window
, N_("n"),
1170 N_("size of the window used for delta compression")),
1171 OPT_STRING(0, "window-memory", &po_args
.window_memory
, N_("bytes"),
1172 N_("same as the above, but limit memory size instead of entries count")),
1173 OPT_STRING(0, "depth", &po_args
.depth
, N_("n"),
1174 N_("limits the maximum delta depth")),
1175 OPT_STRING(0, "threads", &po_args
.threads
, N_("n"),
1176 N_("limits the maximum number of threads")),
1177 OPT_MAGNITUDE(0, "max-pack-size", &po_args
.max_pack_size
,
1178 N_("maximum size of each packfile")),
1179 OPT_PARSE_LIST_OBJECTS_FILTER(&po_args
.filter_options
),
1180 OPT_BOOL(0, "pack-kept-objects", &pack_kept_objects
,
1181 N_("repack objects in packs marked with .keep")),
1182 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list
, N_("name"),
1183 N_("do not repack this pack")),
1184 OPT_INTEGER('g', "geometric", &geometry
.split_factor
,
1185 N_("find a geometric progression with factor <N>")),
1186 OPT_BOOL('m', "write-midx", &write_midx
,
1187 N_("write a multi-pack index of the resulting packs")),
1188 OPT_STRING(0, "expire-to", &expire_to
, N_("dir"),
1189 N_("pack prefix to store a pack containing pruned objects")),
1190 OPT_STRING(0, "filter-to", &filter_to
, N_("dir"),
1191 N_("pack prefix to store a pack containing filtered out objects")),
1195 list_objects_filter_init(&po_args
.filter_options
);
1197 git_config(repack_config
, &cruft_po_args
);
1199 argc
= parse_options(argc
, argv
, prefix
, builtin_repack_options
,
1200 git_repack_usage
, 0);
1202 if (delete_redundant
&& repository_format_precious_objects
)
1203 die(_("cannot delete packs in a precious-objects repo"));
1205 die_for_incompatible_opt3(unpack_unreachable
|| (pack_everything
& LOOSEN_UNREACHABLE
), "-A",
1206 keep_unreachable
, "-k/--keep-unreachable",
1207 pack_everything
& PACK_CRUFT
, "--cruft");
1209 if (pack_everything
& PACK_CRUFT
)
1210 pack_everything
|= ALL_INTO_ONE
;
1212 if (write_bitmaps
< 0) {
1214 (!(pack_everything
& ALL_INTO_ONE
) || !is_bare_repository()))
1216 } else if (write_bitmaps
&&
1217 git_env_bool(GIT_TEST_MULTI_PACK_INDEX
, 0) &&
1218 git_env_bool(GIT_TEST_MULTI_PACK_INDEX_WRITE_BITMAP
, 0)) {
1221 if (pack_kept_objects
< 0)
1222 pack_kept_objects
= write_bitmaps
> 0 && !write_midx
;
1224 if (write_bitmaps
&& !(pack_everything
& ALL_INTO_ONE
) && !write_midx
)
1225 die(_(incremental_bitmap_conflict_error
));
1227 if (write_bitmaps
&& po_args
.local
&& has_alt_odb(the_repository
)) {
1229 * When asked to do a local repack, but we have
1230 * packfiles that are inherited from an alternate, then
1231 * we cannot guarantee that the multi-pack-index would
1232 * have full coverage of all objects. We thus disable
1233 * writing bitmaps in that case.
1235 warning(_("disabling bitmap writing, as some objects are not being packed"));
1239 if (write_midx
&& write_bitmaps
) {
1240 struct strbuf path
= STRBUF_INIT
;
1242 strbuf_addf(&path
, "%s/%s_XXXXXX", get_object_directory(),
1245 refs_snapshot
= xmks_tempfile(path
.buf
);
1246 midx_snapshot_refs(refs_snapshot
);
1248 strbuf_release(&path
);
1251 packdir
= mkpathdup("%s/pack", get_object_directory());
1252 packtmp_name
= xstrfmt(".tmp-%d-pack", (int)getpid());
1253 packtmp
= mkpathdup("%s/%s", packdir
, packtmp_name
);
1255 collect_pack_filenames(&existing
, &keep_pack_list
);
1257 if (geometry
.split_factor
) {
1258 if (pack_everything
)
1259 die(_("options '%s' and '%s' cannot be used together"), "--geometric", "-A/-a");
1260 init_pack_geometry(&geometry
, &existing
, &po_args
);
1261 split_pack_geometry(&geometry
);
1264 prepare_pack_objects(&cmd
, &po_args
, packtmp
);
1266 show_progress
= !po_args
.quiet
&& isatty(2);
1268 strvec_push(&cmd
.args
, "--keep-true-parents");
1269 if (!pack_kept_objects
)
1270 strvec_push(&cmd
.args
, "--honor-pack-keep");
1271 for (i
= 0; i
< keep_pack_list
.nr
; i
++)
1272 strvec_pushf(&cmd
.args
, "--keep-pack=%s",
1273 keep_pack_list
.items
[i
].string
);
1274 strvec_push(&cmd
.args
, "--non-empty");
1275 if (!geometry
.split_factor
) {
1277 * We need to grab all reachable objects, including those that
1278 * are reachable from reflogs and the index.
1280 * When repacking into a geometric progression of packs,
1281 * however, we ask 'git pack-objects --stdin-packs', and it is
1282 * not about packing objects based on reachability but about
1283 * repacking all the objects in specified packs and loose ones
1284 * (indeed, --stdin-packs is incompatible with these options).
1286 strvec_push(&cmd
.args
, "--all");
1287 strvec_push(&cmd
.args
, "--reflog");
1288 strvec_push(&cmd
.args
, "--indexed-objects");
1290 if (repo_has_promisor_remote(the_repository
))
1291 strvec_push(&cmd
.args
, "--exclude-promisor-objects");
1293 if (write_bitmaps
> 0)
1294 strvec_push(&cmd
.args
, "--write-bitmap-index");
1295 else if (write_bitmaps
< 0)
1296 strvec_push(&cmd
.args
, "--write-bitmap-index-quiet");
1298 if (use_delta_islands
)
1299 strvec_push(&cmd
.args
, "--delta-islands");
1301 if (pack_everything
& ALL_INTO_ONE
) {
1302 repack_promisor_objects(&po_args
, &names
);
1304 if (has_existing_non_kept_packs(&existing
) &&
1306 !(pack_everything
& PACK_CRUFT
)) {
1307 for_each_string_list_item(item
, &names
) {
1308 strvec_pushf(&cmd
.args
, "--keep-pack=%s-%s.pack",
1309 packtmp_name
, item
->string
);
1311 if (unpack_unreachable
) {
1312 strvec_pushf(&cmd
.args
,
1313 "--unpack-unreachable=%s",
1314 unpack_unreachable
);
1315 } else if (pack_everything
& LOOSEN_UNREACHABLE
) {
1316 strvec_push(&cmd
.args
,
1317 "--unpack-unreachable");
1318 } else if (keep_unreachable
) {
1319 strvec_push(&cmd
.args
, "--keep-unreachable");
1320 strvec_push(&cmd
.args
, "--pack-loose-unreachable");
1323 } else if (geometry
.split_factor
) {
1324 strvec_push(&cmd
.args
, "--stdin-packs");
1325 strvec_push(&cmd
.args
, "--unpacked");
1327 strvec_push(&cmd
.args
, "--unpacked");
1328 strvec_push(&cmd
.args
, "--incremental");
1331 if (po_args
.filter_options
.choice
)
1332 strvec_pushf(&cmd
.args
, "--filter=%s",
1333 expand_list_objects_filter_spec(&po_args
.filter_options
));
1335 die(_("option '%s' can only be used along with '%s'"), "--filter-to", "--filter");
1337 if (geometry
.split_factor
)
1342 ret
= start_command(&cmd
);
1346 if (geometry
.split_factor
) {
1347 FILE *in
= xfdopen(cmd
.in
, "w");
1349 * The resulting pack should contain all objects in packs that
1350 * are going to be rolled up, but exclude objects in packs which
1351 * are being left alone.
1353 for (i
= 0; i
< geometry
.split
; i
++)
1354 fprintf(in
, "%s\n", pack_basename(geometry
.pack
[i
]));
1355 for (i
= geometry
.split
; i
< geometry
.pack_nr
; i
++)
1356 fprintf(in
, "^%s\n", pack_basename(geometry
.pack
[i
]));
1360 ret
= finish_pack_objects_cmd(&cmd
, &names
, 1);
1364 if (!names
.nr
&& !po_args
.quiet
)
1365 printf_ln(_("Nothing new to pack."));
1367 if (pack_everything
& PACK_CRUFT
) {
1368 const char *pack_prefix
= find_pack_prefix(packdir
, packtmp
);
1370 if (!cruft_po_args
.window
)
1371 cruft_po_args
.window
= po_args
.window
;
1372 if (!cruft_po_args
.window_memory
)
1373 cruft_po_args
.window_memory
= po_args
.window_memory
;
1374 if (!cruft_po_args
.depth
)
1375 cruft_po_args
.depth
= po_args
.depth
;
1376 if (!cruft_po_args
.threads
)
1377 cruft_po_args
.threads
= po_args
.threads
;
1378 if (!cruft_po_args
.max_pack_size
)
1379 cruft_po_args
.max_pack_size
= po_args
.max_pack_size
;
1381 cruft_po_args
.local
= po_args
.local
;
1382 cruft_po_args
.quiet
= po_args
.quiet
;
1384 ret
= write_cruft_pack(&cruft_po_args
, packtmp
, pack_prefix
,
1385 cruft_expiration
, &names
,
1390 if (delete_redundant
&& expire_to
) {
1392 * If `--expire-to` is given with `-d`, it's possible
1393 * that we're about to prune some objects. With cruft
1394 * packs, pruning is implicit: any objects from existing
1395 * packs that weren't picked up by new packs are removed
1396 * when their packs are deleted.
1398 * Generate an additional cruft pack, with one twist:
1399 * `names` now includes the name of the cruft pack
1400 * written in the previous step. So the contents of
1401 * _this_ cruft pack exclude everything contained in the
1402 * existing cruft pack (that is, all of the unreachable
1403 * objects which are no older than
1404 * `--cruft-expiration`).
1406 * To make this work, cruft_expiration must become NULL
1407 * so that this cruft pack doesn't actually prune any
1408 * objects. If it were non-NULL, this call would always
1409 * generate an empty pack (since every object not in the
1410 * cruft pack generated above will have an mtime older
1411 * than the expiration).
1413 ret
= write_cruft_pack(&cruft_po_args
, expire_to
,
1423 if (po_args
.filter_options
.choice
) {
1425 filter_to
= packtmp
;
1427 ret
= write_filtered_pack(&po_args
,
1429 find_pack_prefix(packdir
, packtmp
),
1436 string_list_sort(&names
);
1438 close_object_store(the_repository
->objects
);
1441 * Ok we have prepared all new packfiles.
1443 for_each_string_list_item(item
, &names
) {
1444 struct generated_pack_data
*data
= item
->util
;
1446 for (ext
= 0; ext
< ARRAY_SIZE(exts
); ext
++) {
1449 fname
= mkpathdup("%s/pack-%s%s",
1450 packdir
, item
->string
, exts
[ext
].name
);
1452 if (data
->tempfiles
[ext
]) {
1453 const char *fname_old
= get_tempfile_path(data
->tempfiles
[ext
]);
1454 struct stat statbuffer
;
1456 if (!stat(fname_old
, &statbuffer
)) {
1457 statbuffer
.st_mode
&= ~(S_IWUSR
| S_IWGRP
| S_IWOTH
);
1458 chmod(fname_old
, statbuffer
.st_mode
);
1461 if (rename_tempfile(&data
->tempfiles
[ext
], fname
))
1462 die_errno(_("renaming pack to '%s' failed"), fname
);
1463 } else if (!exts
[ext
].optional
)
1464 die(_("pack-objects did not write a '%s' file for pack %s-%s"),
1465 exts
[ext
].name
, packtmp
, item
->string
);
1466 else if (unlink(fname
) < 0 && errno
!= ENOENT
)
1467 die_errno(_("could not unlink: %s"), fname
);
1472 /* End of pack replacement. */
1474 if (delete_redundant
&& pack_everything
& ALL_INTO_ONE
)
1475 mark_packs_for_deletion(&existing
, &names
);
1478 struct string_list include
= STRING_LIST_INIT_NODUP
;
1479 midx_included_packs(&include
, &existing
, &names
, &geometry
);
1481 ret
= write_midx_included_packs(&include
, &geometry
, &names
,
1482 refs_snapshot
? get_tempfile_path(refs_snapshot
) : NULL
,
1483 show_progress
, write_bitmaps
> 0);
1485 if (!ret
&& write_bitmaps
)
1486 remove_redundant_bitmaps(&include
, packdir
);
1488 string_list_clear(&include
, 0);
1494 reprepare_packed_git(the_repository
);
1496 if (delete_redundant
) {
1498 remove_redundant_existing_packs(&existing
);
1500 if (geometry
.split_factor
)
1501 geometry_remove_redundant_packs(&geometry
, &names
,
1504 opts
|= PRUNE_PACKED_VERBOSE
;
1505 prune_packed_objects(opts
);
1507 if (!keep_unreachable
&&
1508 (!(pack_everything
& LOOSEN_UNREACHABLE
) ||
1509 unpack_unreachable
) &&
1510 is_repository_shallow(the_repository
))
1511 prune_shallow(PRUNE_QUICK
);
1514 if (run_update_server_info
)
1515 update_server_info(0);
1517 if (git_env_bool(GIT_TEST_MULTI_PACK_INDEX
, 0)) {
1519 if (git_env_bool(GIT_TEST_MULTI_PACK_INDEX_WRITE_BITMAP
, 0))
1520 flags
|= MIDX_WRITE_BITMAP
| MIDX_WRITE_REV_INDEX
;
1521 write_midx_file(get_object_directory(), NULL
, NULL
, flags
);
1525 string_list_clear(&names
, 1);
1526 existing_packs_release(&existing
);
1527 free_pack_geometry(&geometry
);
1528 list_objects_filter_release(&po_args
.filter_options
);