repack: support combining '--geometric' with '--cruft'
Teach 'git repack' to accept '--geometric' and '--cruft' together. When both are given, the geometric repack rolls up non-cruft packs as usual, and a separate cruft pack is written to collect unreachable objects. Previously, '--cruft' implied `ALL_INTO_ONE`, which is fundamentally incompatible with geometric repacking. Relax this so that '--cruft' only implies `ALL_INTO_ONE` when '--geometric' is not also given. When combining the two modes: - Use the new '--stdin-packs=follow-reachable' mode so that only reachable objects from the rolled-up packs (and any reachable loose objects) appear in the geometric pack. Unreachable objects are left for the cruft writer to collect. - Plumb our `pack_geometry` into `write_cruft_pack()`, so that the latter can tell 'pack-objects' which non-kept packs are below the split (excluded, so their unreachable objects are candidates for the cruft pack) versus above the split (included, so they are treated as reachable). - Handle promisor packs in the cruft writer's geometry path, since promisor packs have their own split point. - Use the refs snapshot (when available) so that pack-objects and the MIDX bitmap writer see the same set of reference tips. Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Taylor Blau committed
Jun 26, 2026 at 15:02 UTC
a53c3b6193d783f1dd2bd283782b1c3033ef0b3e
5 files changed
+300
-11
Documentation/git-repack.adoc
+11
@@ -70,6 +70,11 @@ to the new separate pack will be written.
70
are packed into a separate cruft pack. Unreachable objects can
71
be pruned using the normal expiry rules with the next `git gc`
72
invocation (see linkgit:git-gc[1]). Incompatible with `-k`.
73
++
74
+When combined with `--geometric`, `--cruft` does not imply `-a`. Instead,
75
+the geometric repack rolls up packs as usual, and a separate cruft pack is
76
+written to collect unreachable objects. Only reachable objects from the
77
+rolled-up packs are included in the resulting geometric pack.
78
79
--cruft-expiration=<approxidate>::
80
Expire unreachable objects older than `<approxidate>`
@@ -245,6 +250,12 @@ progression.
250
Loose objects are implicitly included in this "roll-up", without respect to
251
their reachability. This is subject to change in the future.
252
+
253
+When combined with `--cruft`, only reachable objects from rolled-up packs
254
+are included in the geometric pack, along with any reachable loose objects.
255
+Unreachable objects (both from rolled-up packs and loose) are collected
256
+into a separate cruft pack. Existing cruft packs are retained. See
257
+`--cruft` above for details.
258
++
259
When writing a multi-pack bitmap, `git repack` selects the largest resulting
260
pack as the preferred pack for object selection by the MIDX (see
261
linkgit:git-multi-pack-index[1]).
builtin/repack.c
+16
-7
@@ -260,7 +260,7 @@ int cmd_repack(int argc,
260
keep_unreachable, "-k/--keep-unreachable",
261
pack_everything & PACK_CRUFT, "--cruft");
262
263
- if (pack_everything & PACK_CRUFT)
263
+ if (pack_everything & PACK_CRUFT && !geometry.split_factor)
264
pack_everything |= ALL_INTO_ONE;
265
266
if (write_bitmaps < 0) {
@@ -296,7 +296,8 @@ int cmd_repack(int argc,
296
die(_("invalid value for %s: %d"), "--midx-new-layer-threshold",
297
config_ctx.midx_new_layer_threshold);
298
299
- if (write_midx != REPACK_WRITE_MIDX_NONE && write_bitmaps) {
299
+ if ((write_midx != REPACK_WRITE_MIDX_NONE && write_bitmaps) ||
300
+ (geometry.split_factor && (pack_everything & PACK_CRUFT))) {
301
struct strbuf path = STRBUF_INIT;
302
303
strbuf_addf(&path, "%s/%s_XXXXXX",
@@ -317,7 +318,7 @@ int cmd_repack(int argc,
318
existing_packs_collect(&existing, &keep_pack_list);
319
320
if (geometry.split_factor) {
320
- if (pack_everything)
321
+ if (pack_everything & ~PACK_CRUFT)
322
die(_("options '%s' and '%s' cannot be used together"), "--geometric", "-A/-a");
323
if (write_midx == REPACK_WRITE_MIDX_INCREMENTAL) {
324
geometry.midx_layer_threshold = config_ctx.midx_new_layer_threshold;
@@ -393,10 +394,16 @@ int cmd_repack(int argc,
394
pack_geometry_repack_promisors(repo, &po_args, &geometry,
395
&names, packtmp);
396
396
- if (midx_must_contain_cruft)
397
+ if (pack_everything & PACK_CRUFT) {
398
+ strvec_push(&cmd.args, "--stdin-packs=follow-reachable");
399
+ if (refs_snapshot)
400
+ strvec_pushf(&cmd.args, "--refs-snapshot=%s",
401
+ get_tempfile_path(refs_snapshot));
402
+ } else if (midx_must_contain_cruft)
403
strvec_push(&cmd.args, "--stdin-packs");
404
else
405
strvec_push(&cmd.args, "--stdin-packs=follow");
406
+
407
strvec_push(&cmd.args, "--unpacked");
408
} else {
409
strvec_push(&cmd.args, "--unpacked");
@@ -431,7 +438,8 @@ int cmd_repack(int argc,
438
const char *basename = pack_basename(geometry.pack[i]);
439
char marker = '^';
440
434
- if (!midx_must_contain_cruft &&
441
+ if ((pack_everything & PACK_CRUFT ||
442
+ !midx_must_contain_cruft) &&
443
!string_list_has_string(&existing.midx_packs,
444
basename)) {
445
/*
@@ -505,7 +513,8 @@ int cmd_repack(int argc,
513
514
ret = write_cruft_pack(&opts, cruft_expiration,
515
combine_cruft_below_size, &names,
508
- &existing);
516
+ &existing,
517
+ geometry.split_factor ? &geometry : NULL);
518
if (ret)
519
goto cleanup;
520
@@ -540,7 +549,7 @@ int cmd_repack(int argc,
549
*/
550
opts.destination = expire_to;
551
ret = write_cruft_pack(&opts, NULL, 0ul, &names,
543
- &existing);
552
+ &existing, NULL);
553
if (ret)
554
goto cleanup;
555
}
repack-cruft.c
+20
-3
@@ -36,7 +36,8 @@ int write_cruft_pack(const struct write_pack_opts *opts,
36
const char *cruft_expiration,
37
unsigned long combine_cruft_below_size,
38
struct string_list *names,
39
- struct existing_packs *existing)
39
+ struct existing_packs *existing,
40
+ struct pack_geometry *geometry)
41
{
42
struct child_process cmd = CHILD_PROCESS_INIT;
43
struct string_list_item *item;
@@ -81,8 +82,24 @@ int write_cruft_pack(const struct write_pack_opts *opts,
82
else
83
for_each_string_list_item(item, &existing->cruft_packs)
84
fprintf(in, "-%s.pack\n", item->string);
84
- for_each_string_list_item(item, &existing->non_kept_packs)
85
- fprintf(in, "-%s.pack\n", item->string);
85
+ if (geometry) {
86
+ uint32_t j;
87
+ for (j = 0; j < geometry->split; j++)
88
+ fprintf(in, "-%s\n",
89
+ pack_basename(geometry->pack[j]));
90
+ for (; j < geometry->pack_nr; j++)
91
+ fprintf(in, "%s\n",
92
+ pack_basename(geometry->pack[j]));
93
+ for (j = 0; j < geometry->promisor_split; j++)
94
+ fprintf(in, "-%s\n",
95
+ pack_basename(geometry->promisor_pack[j]));
96
+ for (; j < geometry->promisor_pack_nr; j++)
97
+ fprintf(in, "%s\n",
98
+ pack_basename(geometry->promisor_pack[j]));
99
+ } else {
100
+ for_each_string_list_item(item, &existing->non_kept_packs)
101
+ fprintf(in, "-%s.pack\n", item->string);
102
+ }
103
for_each_string_list_item(item, &existing->kept_packs)
104
fprintf(in, "%s.pack\n", item->string);
105
fclose(in);
repack.h
+2
-1
@@ -169,6 +169,7 @@ int write_cruft_pack(const struct write_pack_opts *opts,
169
const char *cruft_expiration,
170
unsigned long combine_cruft_below_size,
171
struct string_list *names,
172
- struct existing_packs *existing);
172
+ struct existing_packs *existing,
173
+ struct pack_geometry *geometry);
174
175
#endif /* REPACK_H */
t/t7704-repack-cruft.sh
+251
@@ -891,4 +891,255 @@ test_expect_success 'repack rescues once-cruft objects above geometric split' '
891
git repack --geometric=2 -d --write-midx --write-bitmap-index
892
'
893
894
+test_expect_success 'repack --geometric --cruft combines packs and writes cruft' '
895
+ git init geometric-cruft-basic &&
896
+ (
897
+ cd geometric-cruft-basic &&
898
+
899
+ test_commit A &&
900
+ test_commit B &&
901
+
902
+ B="$(git rev-parse B)" &&
903
+
904
+ git reset --hard $B^ &&
905
+ git tag -d B &&
906
+ git reflog expire --all --expire=all &&
907
+
908
+ # Initial state: one non-cruft pack, one cruft pack.
909
+ git repack -d --cruft &&
910
+
911
+ ls $packdir/pack-*.mtimes >cruft.before &&
912
+ test_line_count = 1 cruft.before &&
913
+
914
+ test_commit C &&
915
+ git repack &&
916
+
917
+ # At this point we have three packs:
918
+ # - the non-cruft pack from A
919
+ # - the cruft pack from B
920
+ # - a new non-cruft pack from C
921
+ #
922
+ # The two non-cruft packs are not in a geometric
923
+ # progression, so they should be rolled up.
924
+ git repack -d --geometric=2 --cruft &&
925
+
926
+ # The old cruft pack for B is retained, since the
927
+ # geometric repack does not touch cruft packs.
928
+ ls $packdir/pack-*.mtimes >cruft.after &&
929
+ test_line_count = 1 cruft.after &&
930
+
931
+ # Ensure that all reachable objects are present.
932
+ git fsck
933
+ )
934
+'
935
+
936
+test_expect_success 'repack --geometric --cruft writes new cruft for loose unreachable' '
937
+ git init geometric-cruft-new-cruft &&
938
+ (
939
+ cd geometric-cruft-new-cruft &&
940
+
941
+ git config set maintenance.auto false &&
942
+
943
+ test_commit A &&
944
+ git repack &&
945
+
946
+ test_commit B &&
947
+ git repack &&
948
+
949
+ # Create an unreachable commit whose objects are
950
+ # still loose (never packed).
951
+ test_commit C &&
952
+ C="$(git rev-parse C)" &&
953
+ git reset --hard $C^ &&
954
+ git tag -d C &&
955
+ git reflog expire --all --expire=all &&
956
+
957
+ # At this point we have two non-cruft packs of
958
+ # similar size that are not in geometric progression,
959
+ # and loose unreachable objects from commit C.
960
+ ls $packdir/pack-*.idx >packs.before &&
961
+ test_line_count = 2 packs.before &&
962
+
963
+ # Geometric+cruft repack should roll up the two
964
+ # non-cruft packs and write a new cruft pack for C
965
+ # (whose objects are loose and unreachable).
966
+ git repack -d --geometric=2 --cruft &&
967
+
968
+ ls $packdir/pack-*.mtimes >cruft.after &&
969
+ test_line_count = 1 cruft.after &&
970
+
971
+ git fsck
972
+ )
973
+'
974
+
975
+test_expect_success 'repack --geometric --cruft -d deletes rolled-up packs' '
976
+ git init geometric-cruft-delete &&
977
+ (
978
+ cd geometric-cruft-delete &&
979
+
980
+ test_commit A &&
981
+ git repack -d &&
982
+
983
+ test_commit B &&
984
+ git repack -d &&
985
+
986
+ ls $packdir/pack-*.idx >before &&
987
+
988
+ git repack -d --geometric=2 --cruft &&
989
+
990
+ # Two packs should have been rolled into one. No cruft
991
+ # pack is written because there are no unreachable objects.
992
+ ls $packdir/pack-*.idx >after &&
993
+ test_line_count = 1 after &&
994
+
995
+ # The rolled-up packs should be gone.
996
+ ! test_cmp before after
997
+ )
998
+'
999
+
1000
+test_expect_success 'repack --geometric --cruft collects loose unreachable objects' '
1001
+ git init geometric-cruft-loose &&
1002
+ (
1003
+ cd geometric-cruft-loose &&
1004
+
1005
+ test_commit A &&
1006
+ git repack -d &&
1007
+
1008
+ test_commit B &&
1009
+ git repack &&
1010
+
1011
+ # Create a loose unreachable object by making it
1012
+ # orphaned (not in any pack).
1013
+ loose="$(echo "cruft object" | git hash-object -w --stdin)" &&
1014
+
1015
+ # We have two non-cruft packs and a loose unreachable
1016
+ # object. The geometric+cruft repack should roll up
1017
+ # the packs AND write a cruft pack for the loose
1018
+ # unreachable object.
1019
+ git repack -d --geometric=2 --cruft &&
1020
+
1021
+ ls $packdir/pack-*.mtimes >cruft.packs &&
1022
+ test_line_count = 1 cruft.packs &&
1023
+
1024
+ git fsck
1025
+ )
1026
+'
1027
+
1028
+test_expect_success 'repack --geometric --cruft accumulates cruft packs' '
1029
+ git init geometric-cruft-accumulate &&
1030
+ (
1031
+ cd geometric-cruft-accumulate &&
1032
+
1033
+ git config set maintenance.auto false &&
1034
+
1035
+ test_commit A &&
1036
+ git repack &&
1037
+
1038
+ # First round: create unreachable objects and do a
1039
+ # geometric+cruft repack.
1040
+ unreachable_1="$(echo "cruft 1" | git hash-object -w --stdin)" &&
1041
+ git repack -d --geometric=2 --cruft &&
1042
+
1043
+ ls $packdir/pack-*.mtimes >cruft.1 &&
1044
+ test_line_count = 1 cruft.1 &&
1045
+
1046
+ test_commit B &&
1047
+ git repack &&
1048
+
1049
+ # Second round: create more unreachable objects and
1050
+ # repack again. The old cruft pack should be retained
1051
+ # and a new one written.
1052
+ unreachable_2="$(echo "cruft 2" | git hash-object -w --stdin)" &&
1053
+ git repack -d --geometric=2 --cruft &&
1054
+
1055
+ ls $packdir/pack-*.mtimes >cruft.2 &&
1056
+ test_line_count = 2 cruft.2 &&
1057
+
1058
+ git fsck
1059
+ )
1060
+'
1061
+
1062
+test_expect_success 'repack --geometric --cruft --combine-cruft-below-size' '
1063
+ git init geometric-cruft-combine &&
1064
+ (
1065
+ cd geometric-cruft-combine &&
1066
+
1067
+ git config set maintenance.auto false &&
1068
+
1069
+ test_commit A &&
1070
+ git repack &&
1071
+
1072
+ # Create a small cruft pack.
1073
+ unreachable_1="$(echo "cruft 1" | git hash-object -w --stdin)" &&
1074
+ git repack -d --geometric=2 --cruft &&
1075
+
1076
+ ls $packdir/pack-*.mtimes >cruft.before &&
1077
+ test_line_count = 1 cruft.before &&
1078
+
1079
+ test_commit B &&
1080
+ git repack &&
1081
+
1082
+ # Create another small cruft pack.
1083
+ unreachable_2="$(echo "cruft 2" | git hash-object -w --stdin)" &&
1084
+ git repack -d --geometric=2 --cruft &&
1085
+
1086
+ ls $packdir/pack-*.mtimes >cruft.mid &&
1087
+ test_line_count = 2 cruft.mid &&
1088
+
1089
+ test_commit C &&
1090
+ git repack &&
1091
+
1092
+ # With --combine-cruft-below-size, the two small cruft
1093
+ # packs should be combined into one.
1094
+ unreachable_3="$(echo "cruft 3" | git hash-object -w --stdin)" &&
1095
+ git repack -d --geometric=2 --cruft \
1096
+ --combine-cruft-below-size=10M &&
1097
+
1098
+ ls $packdir/pack-*.mtimes >cruft.after &&
1099
+ test_line_count = 1 cruft.after &&
1100
+
1101
+ git fsck
1102
+ )
1103
+'
1104
+
1105
+test_expect_success 'repack --geometric --cruft --expire-to' '
1106
+ git init geometric-cruft-expire-to &&
1107
+ (
1108
+ cd geometric-cruft-expire-to &&
1109
+
1110
+ git config set maintenance.auto false &&
1111
+
1112
+ test_commit A &&
1113
+ git repack &&
1114
+
1115
+ test_commit B &&
1116
+ git repack &&
1117
+
1118
+ # Create unreachable objects and record them.
1119
+ test_commit C &&
1120
+ C="$(git rev-parse C)" &&
1121
+ git rev-list --objects --no-object-names B..C >unreachable.raw &&
1122
+ sort unreachable.raw >unreachable.want &&
1123
+
1124
+ git reset --hard $C^ &&
1125
+ git tag -d C &&
1126
+ git reflog expire --all --expire=all &&
1127
+
1128
+ git init --bare expired.git &&
1129
+ git repack -d --geometric=2 --cruft \
1130
+ --cruft-expiration=now \
1131
+ --expire-to="expired.git/objects/pack/pack" &&
1132
+
1133
+ # The expired objects should appear in the
1134
+ # expire-to location.
1135
+ expired="$(ls expired.git/objects/pack/pack-*.idx)" &&
1136
+ test_path_is_file "${expired%.idx}.mtimes" &&
1137
+ git show-index <"$expired" >expired.raw &&
1138
+ cut -d" " -f2 expired.raw | sort >expired.objects &&
1139
+ test_cmp unreachable.want expired.objects &&
1140
+
1141
+ git fsck
1142
+ )
1143
+'
1144
+
1145
test_done