From: Jeff King <peff@peff.net>
Date: Wed, 07 Jan 2026 07:40:00 GMT
Subject: Re: [PATCH v2 1/3] t/unit-tests: update clar to 39f11fe
Message-ID: <20260107074000.GA102520@coredump.intra.peff.net>
In-Reply-To: <aVzvDGVEI2qVJv2F@pks.im>

On Tue, Jan 06, 2026 at 12:16:28PM +0100, Patrick Steinhardt wrote:

> The problem is that we cannot do so easily. Varargs require at least one
> argument to be present, so we cannot make this `cl_fail(desc, ...)`
> without breaking the case where there are no variable arguments:
> 
>   In file included from ../t/unit-tests/clar/clar.c:1053:
>   ../t/unit-tests/clar/clar/fs.h:460:3: error: expected expression
>     460 |                 cl_fail("Cannot copy; cannot stat destination");
>         |                 ^
>   ../t/unit-tests/clar/clar.h:152:132: note: expanded from macro 'cl_fail'
>     152 | #define cl_fail(desc,...) clar__failf(CLAR_CURRENT_FILE, CLAR_CURRENT_FUNC, CLAR_CURRENT_LINE, 1, "Test failed.", desc, __VA_ARGS__)
>         |                                                                                                                                    ^
> 
> The alternative would be to make this `cl_fail(...)` instead, but to the
> best of my knowledge this isn't even a valid construct.

It is valid; that's how we define BUG(), for example.

There are other options like GNU's "## __VA_ARGS__", but I think eating
the format argument with the "..." is the only portable way, at least
for standard versions we'd support. I think C23 added __VA_OPT__.

-Peff