Re: [PATCH v4 3/4] object-file: emit corruption errors when detected

[Jeff King <peff@peff.net>]

On Mon, Dec 12, 2022 at 02:52:12PM -0800, Jonathan Tan wrote:

> > > Junio made a point that there could, for example, be no path when the
> > > odb list is empty (maybe in the future) so I don't think this would be
> > > sufficient. But there is already a comment there pointing to a comment
> > > in another function that states "path ... (if any)" so this is something
> > > that callers should already take care of. In my changes, I'll initialize
> > > it to NULL and whenever I use it, I'll check for non-NULL first.
> > 
> > If we return a non-negative value, then we opened something, so by
> > definition, don't we have a path of the thing we opened?
> 
> Hmm...are you saying "path is guaranteed when there is no error; when
> there is an error, the caller must check"? If yes, I think we are in
> agreement. In any case, to make things more concrete, I've just sent a
> new version [1].

Almost. I'm saying "path is guaranteed when there is no error; when
there is an error, the value of path is meaningless and should not be
looked at".

If you want to enforce that open_loose_object() sets "path" to NULL on
error, and then say "the caller must check", that would be valid. But
without that, even checking it for NULL is pointless (because you may
see a path which got ENOENT, even though we got an interesting errno
from an earlier path).

-Peff