| Message ID | 285695deafa5a4a49f774dc484782dd8e4fd4997.1681384405.git.ps@pks.im (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | repack: fix geometric repacking with gitalternates | expand |
On 4/13/2023 7:16 AM, Patrick Steinhardt wrote: > When the user passes `-l` to git-repack(1), then they essentially ask us > to only repack objects part of the local object database while ignoring > any packfiles part of an alternate object database. And we in fact honor > this bit when doing a geometric repack as the resulting packfile will > only ever contain local objects. > + # Verify that our assumptions actually hold: both generated packfiles > + # should have three objects and should be non-equal. > + packed_objects shared/.git/objects/pack/pack-*.idx >packed-objects && > + test_line_count = 3 packed-objects && > + packed_objects member/.git/objects/pack/pack-*.idx >packed-objetcs && Typo: s/packed-objetcs/packed-objects/ > + test_line_count = 3 packed-objects && > + test "$(basename member/.git/objects/pack/pack-*.pack)" != "$(basename shared/.git/objects/pack/pack-*.pack)" && nit: could we do this where we store the output of the previous two commands into different files and then use "! test_cmp"? packed_objects shared/.git/objects/pack/pack-*.idx >shared-objects && packed_objects member/.git/objects/pack/pack-*.idx >member-objects && test_line_count = 3 shared-objects && test_line_count = 3 member-objects && ! test_cmp shared-objects member-objects && Thanks, -Stolee
On Thu, Apr 13, 2023 at 09:59:02AM -0400, Derrick Stolee wrote: > On 4/13/2023 7:16 AM, Patrick Steinhardt wrote: > > When the user passes `-l` to git-repack(1), then they essentially ask us > > to only repack objects part of the local object database while ignoring > > any packfiles part of an alternate object database. And we in fact honor > > this bit when doing a geometric repack as the resulting packfile will > > only ever contain local objects. > > > + # Verify that our assumptions actually hold: both generated packfiles > > + # should have three objects and should be non-equal. > > + packed_objects shared/.git/objects/pack/pack-*.idx >packed-objects && > > + test_line_count = 3 packed-objects && > > + packed_objects member/.git/objects/pack/pack-*.idx >packed-objetcs && > > Typo: s/packed-objetcs/packed-objects/ > > > + test_line_count = 3 packed-objects && > > + test "$(basename member/.git/objects/pack/pack-*.pack)" != "$(basename shared/.git/objects/pack/pack-*.pack)" && > > nit: could we do this where we store the output of the previous two > commands into different files and then use "! test_cmp"? > > packed_objects shared/.git/objects/pack/pack-*.idx >shared-objects && > packed_objects member/.git/objects/pack/pack-*.idx >member-objects && > test_line_count = 3 shared-objects && > test_line_count = 3 member-objects && > ! test_cmp shared-objects member-objects && Yeah, this is definitely easier to read. Patrick
Derrick Stolee <derrickstolee@github.com> writes: >> + test_line_count = 3 packed-objects && >> + packed_objects member/.git/objects/pack/pack-*.idx >packed-objetcs && > > Typo: s/packed-objetcs/packed-objects/ > >> + test_line_count = 3 packed-objects && That's a good one. Because the result file created by the previous step happens to also have three lines, the typo does not cause the test to fail, but it is not testing what it designed to test. Here is "git diff @{1}" after my local "rebase -i" ("diff" of the result is easier to read than "range-diff" in this case, as I did not touch any log message). Thanks for a very good set of eyes. diff --git c/t/t5331-pack-objects-stdin.sh w/t/t5331-pack-objects-stdin.sh index 45e24fa94a..acab31667a 100755 --- c/t/t5331-pack-objects-stdin.sh +++ w/t/t5331-pack-objects-stdin.sh @@ -7,7 +7,7 @@ export GIT_TEST_DEFAULT_INITIAL_BRANCH_NAME TEST_PASSES_SANITIZE_LEAK=true . ./test-lib.sh -packed_objects() { +packed_objects () { git show-index <"$1" >tmp-object-list && cut -d' ' -f2 tmp-object-list | sort && rm tmp-object-list diff --git c/t/t7703-repack-geometric.sh w/t/t7703-repack-geometric.sh index 33d7977fca..57796f3437 100755 --- c/t/t7703-repack-geometric.sh +++ w/t/t7703-repack-geometric.sh @@ -10,7 +10,7 @@ objdir=.git/objects packdir=$objdir/pack midx=$objdir/pack/multi-pack-index -packed_objects() { +packed_objects () { git show-index <"$1" >tmp-object-list && cut -d' ' -f2 tmp-object-list && rm tmp-object-list @@ -380,7 +380,7 @@ test_expect_success '--geometric -l with non-intact geometric sequence across OD # should have three objects and should be non-equal. packed_objects shared/.git/objects/pack/pack-*.idx >packed-objects && test_line_count = 3 packed-objects && - packed_objects member/.git/objects/pack/pack-*.idx >packed-objetcs && + packed_objects member/.git/objects/pack/pack-*.idx >packed-objects && test_line_count = 3 packed-objects && test "$(basename member/.git/objects/pack/pack-*.pack)" != "$(basename shared/.git/objects/pack/pack-*.pack)" &&
diff --git a/builtin/repack.c b/builtin/repack.c index 80fc860613..5768aeb1f3 100644 --- a/builtin/repack.c +++ b/builtin/repack.c @@ -323,7 +323,8 @@ static int geometry_cmp(const void *va, const void *vb) } static void init_pack_geometry(struct pack_geometry **geometry_p, - struct string_list *existing_kept_packs) + struct string_list *existing_kept_packs, + const struct pack_objects_args *args) { struct packed_git *p; struct pack_geometry *geometry; @@ -333,6 +334,14 @@ static void init_pack_geometry(struct pack_geometry **geometry_p, geometry = *geometry_p; for (p = get_all_packs(the_repository); p; p = p->next) { + if (args->local && !p->pack_local) + /* + * When asked to only repack local packfiles we skip + * over any packfiles that are borrowed from alternate + * object directories. + */ + continue; + if (!pack_kept_objects) { /* * Any pack that has its pack_keep bit set will appear @@ -900,7 +909,7 @@ int cmd_repack(int argc, const char **argv, const char *prefix) if (geometric_factor) { if (pack_everything) die(_("options '%s' and '%s' cannot be used together"), "--geometric", "-A/-a"); - init_pack_geometry(&geometry, &existing_kept_packs); + init_pack_geometry(&geometry, &existing_kept_packs, &po_args); split_pack_geometry(geometry, geometric_factor); } diff --git a/t/t7703-repack-geometric.sh b/t/t7703-repack-geometric.sh index d0823f2eb2..c440956ad5 100755 --- a/t/t7703-repack-geometric.sh +++ b/t/t7703-repack-geometric.sh @@ -10,6 +10,12 @@ objdir=.git/objects packdir=$objdir/pack midx=$objdir/pack/multi-pack-index +packed_objects() { + git show-index <"$1" >tmp-object-list && + cut -d' ' -f2 tmp-object-list && + rm tmp-object-list + } + test_expect_success '--geometric with no packs' ' git init geometric && test_when_finished "rm -fr geometric" && @@ -361,4 +367,56 @@ test_expect_success '--geometric with same pack in main and alternate ODB' ' test_cmp expected-files actual-files ' +test_expect_success '--geometric -l with non-intact geometric sequence across ODBs' ' + test_when_finished "rm -fr shared member" && + + git init shared && + test_commit_bulk -C shared --start=1 1 && + + git clone --shared shared member && + test_commit_bulk -C member --start=2 1 && + + # Verify that our assumptions actually hold: both generated packfiles + # should have three objects and should be non-equal. + packed_objects shared/.git/objects/pack/pack-*.idx >packed-objects && + test_line_count = 3 packed-objects && + packed_objects member/.git/objects/pack/pack-*.idx >packed-objetcs && + test_line_count = 3 packed-objects && + test "$(basename member/.git/objects/pack/pack-*.pack)" != "$(basename shared/.git/objects/pack/pack-*.pack)" && + + # Perform the geometric repack. With `-l`, we should only see the local + # packfile and thus arrive at the conclusion that the geometric + # sequence is intact. We thus expect no changes. + # + # Note that we are tweaking mtimes of the packfiles so that we can + # verify they did not change. This is done in order to detect the case + # where we do repack objects, but the resulting packfile is the same. + test-tool chmtime --verbose =0 member/.git/objects/pack/* >expected-member-packs && + git -C member repack --geometric=2 -l -d && + test-tool chmtime --verbose member/.git/objects/pack/* >actual-member-packs && + test_cmp expected-member-packs actual-member-packs && + + { + packed_objects shared/.git/objects/pack/pack-*.idx && + packed_objects member/.git/objects/pack/pack-*.idx + } | sort >expected-objects && + + # On the other hand, when doing a non-local geometric repack we should + # see both packfiles and thus repack them. We expect that the shared + # object database was not changed. + test-tool chmtime --verbose =0 shared/.git/objects/pack/* >expected-shared-packs && + git -C member repack --geometric=2 -d && + test-tool chmtime --verbose shared/.git/objects/pack/* >actual-shared-packs && + test_cmp expected-shared-packs actual-shared-packs && + + # Furthermore, we expect that the member repository now has a single + # packfile that contains the combined shared and non-shared objects. + ls member/.git/objects/pack/pack-*.idx >actual && + test_line_count = 1 actual && + packed_objects member/.git/objects/pack/pack-*.idx >actual-objects && + test_line_count = 6 actual-objects && + sort <actual-objects >actual-objects.sorted && + test_cmp expected-objects actual-objects.sorted +' + test_done
When the user passes `-l` to git-repack(1), then they essentially ask us to only repack objects part of the local object database while ignoring any packfiles part of an alternate object database. And we in fact honor this bit when doing a geometric repack as the resulting packfile will only ever contain local objects. What we're missing though is that we don't take locality of packfiles into account when computing whether the geometric sequence is intact or not. So even though we would only ever roll up local packfiles anyway, we could end up trying to repack because of non-local packfiles. This does not make much sense, and in the worst case it can cause us to try and do the geometric repack over and over again because we're never able to restore the geometric sequence. Fix this bug by honoring whether the user has passed `-l`. If so, we skip adding any non-local packfiles to the pack geometry. Signed-off-by: Patrick Steinhardt <ps@pks.im> --- builtin/repack.c | 13 +++++++-- t/t7703-repack-geometric.sh | 58 +++++++++++++++++++++++++++++++++++++ 2 files changed, 69 insertions(+), 2 deletions(-)