On Tue, Nov 17, 2020 at 06:07:07PM -0500, Eric Sunshine wrote:
> On Tue, Nov 17, 2020 at 5:55 PM Jeff King <peff@xxxxxxxx> wrote:
> > I actually wonder if it is time to drop HAVE_VARIADIC_MACROS completely.
> > They are in C99, and we have been introducing many other C99-isms. It
> > would be a minor cleanup to a few bits of code, which perhaps isn't
> > worth the risk. But I also have a vague memory of not being able to
> > implement some interfaces because we couldn't count on them.
> >
> > Poking around, 3689539127 (add helpers for allocating flex-array
> > structs, 2016-02-22) points out one such case. I think discussion of
> > BUG_ON() got blocked by that, too. Looks like we also discussed them in
> > the big "C99 weather balloon" thread:
>
> There was also an idea[1] to use variadic macros to prevent `errno`
> from being clobbered in a case like this:
>
> die_errno(_("blah %s"), foobar());
>
> In which, if foobar() changes `errno`, then die_errno() reports the wrong value.
>
> [1]: https://lore.kernel.org/git/CAPig+cQKMxwadf9aGyC5ESa-vxDy9PzrYo+m+JaVQ3S=12PyQQ@xxxxxxxxxxxxxx/
Thanks, that's another good example. This _could_ be done conditionally
on HAVE_VARIADIC_MACROS, but it would mean on the non-variadic platform
that they'd see the wrong errno. I'm willing to give them worse BUG()
messages, but that may be going too far. :)
I still suspect there may be zero such platforms, which would mean
nobody is affected. But if that is the case, then we should just say so
and get rid of the HAVE_VARIADIC_MACROS flag entirely.
-Peff
