Re: [PATCH 2/7] setup: introduce object storage extension
- From
Karthik Nayak <karthik.188@gmail.com>
- Date
- Oct 20, 2026, 19:06 UTC
- Message-ID
- <CAOLa=ZQ9YH43igSivrCTy9ruDsRnTDPu0xipp3XdpNXU2zmtCA@mail.gmail.com>
- In-Reply-To
- <atcrO3y0o-2G8Dpu@pks.im>
Patrick Steinhardt <ps@pks.im> writes:
Show 75 quoted lines
> On Mon, Oct 19, 2026 at 02:29:20PM -0700, Karthik Nayak wrote:
>> Patrick Steinhardt <ps@pks.im> writes:
>> > diff --git a/odb/source-files.c b/odb/source-files.c
>> > index 455f409db0..4a7cbbfd9b 100644
>> > --- a/odb/source-files.c
>> > +++ b/odb/source-files.c
>> > @@ -1245,6 +1246,12 @@ struct odb_source_files *odb_source_files_new(struct object_database *odb,
>> > object_dir = xstrdup_or_null(getenv(DB_ENVIRONMENT));
>> > alternates = xstrdup_or_null(getenv(ALTERNATE_DB_ENVIRONMENT));
>> > }
>> > + if (!object_dir && payload) {
>> > + if (is_absolute_path(payload))
>> > + object_dir = xstrdup(payload);
>> > + else
>> > + object_dir = xstrfmt("%s/%s", odb->repo->commondir, payload);
>> > + }
>>
>> On the reference side, we use `strbuf_realpath()` to convert the
>> relative paths to absolute paths, should we be doing that here too?
>
> Overall I'd love for the "files" source to only ever use absolute paths,
> but historically that wasn't the case. We sometimes expose those paths
> to users, and changing the "files" source thus breaks a bunch of tests.
>
> So I'm retaining that behaviour for now, also with the extension. It may
> be something to explore going forward though.
>
>> > diff --git a/odb/source.c b/odb/source.c
>> > index b25ef14df8..10abe33a26 100644
>> > --- a/odb/source.c
>> > +++ b/odb/source.c
>> > @@ -1,4 +1,5 @@
>> > #include "git-compat-util.h"
>> > +#include "gettext.h"
>> > #include "object-file.h"
>> > #include "odb/source-files.h"
>> > #include "odb/source.h"
>>
>> Do we need this import? This commit seems to compile without it.
>
> Yeah, we add some translatable strings to `odb_source_new()`. The header
> is also implicitly included, but as Junio pointed out we should include
> headers explicitly regardless when we require their declarations.
>
>> > diff --git a/t/t0001-init.sh b/t/t0001-init.sh
>> > index f1d227f271..c7c464d3a2 100755
>> > --- a/t/t0001-init.sh
>> > +++ b/t/t0001-init.sh
>> > @@ -839,6 +839,36 @@ test_expect_success 'init with --ref-storage-format=garbage' '
>> > test_cmp expect err
>> > '
>> >
>> > +test_expect_success DEFAULT_REPO_FORMAT 'extensions.objectStorage is not allowed with repo version 0' '
>> > + test_when_finished "rm -rf objstorage" &&
>> > + git init objstorage &&
>> > + git -C objstorage config set extensions.objectStorage files &&
>> > + test_must_fail git -C objstorage rev-parse 2>err &&
>> > + test_grep "repo version is 0, but v1-only extension found" err
>> > +'
>> > +
>> > +test_expect_success DEFAULT_REPO_FORMAT 'extensions.objectStorage with files backend' '
>> > + test_when_finished "rm -rf objstorage" &&
>> > + git init objstorage &&
>> > + git -C objstorage config set core.repositoryformatversion 1 &&
>>
>> If we're doing this, do we still need DEFAULT_REPO_FORMAT?
>
> We could indeed get away with not having the prerequisite right now. But
> once we have a second backend we'd need this, because in that case we'd
> potentially initialize the repository with a different backend here. And
> then just switching over to the "files" backend would probably not
> result in anything good.
>
> So this is more of a protective measure for the future.
>Yeah but this also means that this doesn't run with refs reftable, no? It should be okay, I'll leave it to you.
Show 18 quoted lines
>> > + git -C objstorage config set extensions.objectStorage garbage &&
>> > + test_must_fail git -C objstorage rev-parse 2>err &&
>> > + test_grep "invalid value for ${SQ}extensions.objectstorage${SQ}: ${SQ}garbage${SQ}" err &&
>> > +
>> > + git config --file objstorage/.git/config extensions.objectStorage garbage://payload &&
>> > + test_must_fail git -C objstorage rev-parse 2>err &&
>> > + test_grep "invalid value for ${SQ}extensions.objectstorage${SQ}: ${SQ}garbage://payload${SQ}" err
>> > +'
>> > +
>>
>> We should add tests for the files ODB with {valid, invalid} payload. We
>> should probably also test absolute and relative paths in the valid flow.
>
> I add tests for both absolute and relative paths in a later commit. Here
> I'm basically just verifying whether the object storage extension even
> kicks in at all, without yet verifying its semantics properly.
>
> PatrickYeah, I think I saw those after this email.