Re: [PATCH 02/12] remove REFFILES prerequisite

From: Patrick Steinhardt <ps@pks.im>
Date: Jan 18, 2024, 11:21 UTC
Message-ID: <ZakJz_Kj5-tlKL_U@tanuki>
In-Reply-To: <xmqqedeftqn7.fsf@gitster.g>

On Wed, Jan 17, 2024 at 04:46:20PM -0800, Junio C Hamano wrote:

Show 19 quoted lines

> "John Cai via GitGitGadget" <gitgitgadget@gmail.com> writes:
> 
> > From: John Cai <johncai86@gmail.com>
> >
> > These tests are compatible with the reftable backend and thus do not
> > need the REFFILES prerequisite.
> 
> May want to give a bit more backstory here?  After all, 53af25e4
> (t1405: mark test that checks existence as REFFILES, 2022-01-31) and
> 53af25e4 (t1405: mark test that checks existence as REFFILES,
> 2022-01-31) marked these tests to require REFFILES and they explain
> the reason for doing so was exactly because the reftable backend did
> not have the notion of "the reflog for this ref exists" that is
> independent from "the reflog for this ref exists and has one or more
> reflog records".  If your work on the reftable backend during the
> past few years added support for "already exists, but there is no
> entry yet" state for reflogs, that would be great, but it would make
> sense to explain why they suddenly have become "compatible with the
> reftable backend".

I don't know a lot about the history any why we initially didn't think it would be compatible, mostly because there is no history of how the reftable backend itself evolved over time. I can only say that when I took over the effort that this indeed worked as expected by writing "existence" markers into the reflog, where this existence marker is a simple entry where both old and new object ID are set to the null OID.

Patrick