Re: [PATCH v2 1/4] odb: decouple source path comparisons from `the_repository`

[Patrick Steinhardt <ps@pks.im>]

On Mon, Aug 17, 2026 at 03:36:21AM -0400, Jeff King wrote:
> On Mon, Aug 17, 2026 at 09:16:53AM +0200, Patrick Steinhardt wrote:
> 
> > > So if we want to go down this path I'm inclined to just unconditionally
> > > use case sensitive matching and not introduce any secondary machinery.
> > 
> > Thinking about this a bit more: I'd suggest that we leave this out of
> > this patch and instead document this as a NEEDSWORK area for now. I
> > _think_ that this proposed refactoring should be generally fine, and I
> > quite like the simplification that results from it. But the risk for
> > regression is quite a bit higher compared to the origanal patch that
> > I've proposed.
> 
> OK. The inline lookup of core.ignoreCase feels quite gross to me, but
> it's _probably_ OK.
> 
> There are all kinds of weird timing issues lurking with config lookup,
> though. In particular you cache the result in o->source_paths_icase. But
> would we ever load odb source paths before the repo is fully loaded into
> memory (or in the case of clone, even fully formed on disk)? In that
> case we'd cache the wrong value forever.

Good callout, there's one gotcha here that I was already fixing in a
subsequent patch series. Namely, we call `create_object_directory()`
before we set "core.sharedRepository" in `init_db()`. But in all the
other cases we should be fine.

I'll cherry-pick that patch into this series.

> I think we have repo_ignore_case() now, since e6a79c9eb8 (config: use
> repo_ignore_case() to access core.ignorecase, 2026-06-19). That's in
> 'master', so it might be worth building on that instead. And then if
> there's any cache invalidation to do, it would eventually happen there.

We can't use that one though, as it uses `repo_config_values()`, and
that function only works with `the_repository`. So that'd break with
submodule repositories.

Patrick