Re: [PATCH v2 04/18] make: merge reftable lib into libgit.a

From: Ezekiel Newren <ezekielnewren@gmail.com>
Date: Sep 19, 2025, 23:02 UTC
Message-ID: <CAH=ZcbD8zyw882D3HNSVNhovVCMO18fsJy6k5UfXncXv0u05dg@mail.gmail.com>
In-Reply-To: <xmqq1po242uo.fsf@gitster.g>

On Fri, Sep 19, 2025 at 2:14 PM Junio C Hamano <gitster@pobox.com> wrote:

> Instead of rolling everything into LIB_OBJS directly, wouldn't it
> have been much easier to work with if reftable-related ones are left
> in REFTABLE_OBJS and then RERFTABLE_OBJS gets added to LIB_OBJS?
> Wouldn't it have been less prone to mismerges to do it that way?

I didn't do it that way because I didn't think of that. I prefer your way because it's much cleaner.

Show 7 quoted lines

> As we can always extend things more, getting something close to the
> minimally viable set with some tests for sanity checking would be a
> good first goal.  If you pare down way too much, however, we may end
> up to be pretty close to what Patrick sent out originally with the
> varint conversion, so let's make sure we do not drop below the
> minimum that still demonstrates that we have Rust integration that
> is viable going forward.

I'm torn between A and B from [1]: I'm thinking of choosing A because my approach and Patrick's approach are incompatible. In order to use multiple crates and ensure that Rust is built and tested the same way with Make and Meson I'd have to rip out lots of things from Patrick's patch series.

But maybe I should go with B because there's a lot of prep work that should be done with Git before I can start doing what I'd like to do in Rust. So let Patrick's stuff merge while I work on cleaning up xdiff and updating Makefile's libgit.a build process and then deal with ugly merging later.

I can't decide between the two.

[1] https://lore.kernel.org/git/CABPp-BHJUkSERQon6xx=sHrhN7i=6ekv+Hz1+P+2mh0=Xw15Mg@mail.gmail.com/