On Wed, Apr 28, 2021 at 6:40 PM Ævar Arnfjörð Bjarmason
[off-list ref] wrote:

quoted

quoted

I.e. per [1] once if and when we have GIT_TEST_REFTABLE passing surely
one of the best way to garner feedback on the rest is to discover those
parts (using "make gcov", after running with/without
GIT_TEST_REFTABLE=[true|false]) where we still have outstanding blind
spots between the tests and code.

Getting GIT_TEST_REFTABLE=1 passing is the hard part, because it means
having  to understand exactly how the current code is supposed to
work.  Once I get to that point (with knowledge being complete and
tests passing), it will be easy to document what is happening and why.

I was hoping that by posting these series with known test failures,
and questions marked "XXX" in reftable-backend.c, I would get feedback
from the other people who know exactly how this part of the code
works.  But from your mail, I get the sense that nobody understands
how the whole picture fits together?

Almost definitely not. I don't know about you but when I'm looking at
code I wrote 6 months ago handling some special case it takes me a while
to get up to speed on just knowing what I knew then, and when we're
talking about something like refs.c ...

:-(

on the bright side, once this is in, the expectations will be much
more explicit, because there are two backends that have to work in
roughly the same way.

On the topic of the way forward: I for one would very much be for a plan
where step 0 is to just a series import the reftable code you have
as-is. I.e. we'd include it as an imported external library, maybe have
some light test-tool integration and compile it / run its own tests, but
not have/advertise the "git init" etc. integration yet.

I would really like that too, and I support this way forward. The bulk
of the work and problems are in the refs/reftable-backend.c and
assorted incompatibilities across the code base. The library itself
seems pretty solid at this point.

I'm sure there's some things that'll need to change as we start the
test/integration work, e.g. the reflog topic that's been discussed. But
that's surely better done as some patches on top of the already-landed
library import at this point v.s. trying to get the library perfect
before getting it in-tree.

Maybe Junio disagrees, just my 0.02....

-- 
Han-Wen Nienhuys - Google Munich
I work 80%. Don't expect answers from me on Fridays.
--

Google Germany GmbH, Erika-Mann-Strasse 33, 80636 Munich

Registergericht und -nummer: Hamburg, HRB 86891

Sitz der Gesellschaft: Hamburg

Geschäftsführer: Paul Manicle, Halimah DeLaine Prado