On Fri, Nov 22, 2024 at 10:41:16AM -0500, Jeff King wrote:
> > A possible alternative "fix" for this typo could be to unify these
> > {type}_objects members into a single .non_commit_objects member in
> > the rev_info structure; given that we (as far as I remember) never
> > utilized the "feature" that, say, we can ask only for trees but not
> > blobs for the past ~20 years, nobody knows if the apparent "support"
> > for that feature is subtly broken in multiple ways (and one of them
> > you just fixed with this patch) and the "feature" may not be worth
> > keeping.
>
> I have been tempted to do that, too. But FWIW, I do remember
> implementing something that set some but not all of the fields in a
> series. Digging in my old branches, I think it may have been for a
> reachability check for remote git-archive, where we want to dig only as
> deep as the object we are looking for (so if we are finding out if a
> tree is reachable, we do not need to ask about blobs, and looking for a
> tag needs neither trees nor blobs).
>
> If no such code made it into the project in those 20 years, it may not
> worth worrying about too much. But I wanted to point it out as a
> plausible use case.
There are also two curious cases added by f18b512bbb (bundle: create
filtered bundles, 2022-03-09) that set tree/blob, but not "tag". Not
sure if that is a bug or not.
-Peff
