Re: [PATCH 2/7] t0410: enable tests with extensions with non-default repo format

From: Junio C Hamano <gitster@pobox.com>
Date: Feb 15, 2024, 17:18 UTC
Message-ID: <xmqqr0hdsl25.fsf@gitster.g>
In-Reply-To: <Zc3EbCXFdG8E7_v2@tanuki>

Patrick Steinhardt <ps@pks.im> writes:

Show 28 quoted lines

> On Wed, Feb 14, 2024 at 02:57:55PM -0800, Junio C Hamano wrote:
>> Patrick Steinhardt <ps@pks.im> writes:
>> 
>> > In t0410 we have two tests which exercise how partial clones behave in
>> > the context of a repository with extensions. These tests are marked to
>> > require a default repository using SHA1 and the "files" backend because
>> > we explicitly set the repository format version to 0.
>> >
>> > Changing the repository format version to 0 is not needed though. The
>> > "noop" extension is ignored as expected regardless of what the version
>> > is set to, same as the "nonsense" extension leads to failure regardless
>> > of the version.
>> 
>> Isn't the reason why 11664196 kept the forcing of the format version
>> because it wanted to see noop ignored and nonsense failed even if
>> the format version is 0 to ensure the regression it fixed will stay
>> fixed?  IOW, we force version 0 not because we do not want to test
>> with anything but SHA1 and REFFILES; we pretty much assume that with
>> the default version, noop and nonsense will be handled sensibly, and
>> we want to make sure they will be with version 0 as well.
>> 
>> And once we force to version 0, we have trouble running with
>> anything other than SHA1 and REFFILES, hence these prerequisites.
>> 
>> So, I dunno.
>
> Hum, I guess that's fair. Let me adapt the test case to instead use the
> DEFAULT_REPO_FORMAT prerequisite then.

If we expect that "git init" by default will create version 0 repository in the foreseeable future (and when the default gets fliped to create version 1, we will add CI task that forces version 0 somehow, just like we have special CI task to run test with ref backend set to reftable), then I think the patch as posted may actually be a good idea. As long as somebody is checking these with a repository in version 0 format, the original motivation behind the tests is satisified, and if we in addition make sure noop and nonsense are handled sensibly with higher versions, that would also be good, too.

So perhaps we need some in-code comment next to these tests to remind us about that, while removing these prerequisites?

Thanks.