[PATCH v7] assert: Support assert as variadic macro for C++26 [PR27276]

[Tomasz Kaminski]

On Tue, Mar 3, 2026 at 1:39 PM Adhemerval Zanella Netto <
adhemerval.zanella@linaro.org> wrote:

>
>
> On 03/03/26 07:56, Yury Khrustalev wrote:
> > On Wed, Feb 25, 2026 at 03:01:23PM +0100, Tomasz Kamiński wrote:
> >> From: Jonathan Wakely <jwakely@redhat.com>
> >
> > This patch has a new test that fails to compile with GCC 14.2 (although
> > it works with GCC 14.3) and causes make check to fail e.g. on Debian 13.
> >
> >> ...
> >>
> >> diff --git a/assert/test-assert-c++-variadic.cc
> b/assert/test-assert-c++-variadic.cc
> >> new file mode 100644
> >> index 0000000000..1bb8eeb525
> >> --- /dev/null
> >>
> >> ...
> >>
> >> +template <typename Ts>
> >> +constexpr bool
> >> +assert_works ()
> >> +{
> >> +  return requires (Ts ts) {
> >> +    assert (ts);
> >> +  };
> >> +}
> >> +
> >> +enum OE { oe };
> >> +enum TE : int { te };
> >> +enum class SE : int { se };
> >> +
> >> +static_assert ( assert_works <OE> ());
> >> +static_assert ( assert_works <TE> ());
> >
> > These static asserts fail:
> >
> >   test-assert-c++-variadic.cc:81:35: error: static assertion failed
> >      81 | static_assert ( assert_works <OE> ());
> >       |                 ~~~~~~~~~~~~~~~~~~^~
> >   test-assert-c++-variadic.cc:82:35: error: static assertion failed
> >      82 | static_assert ( assert_works <TE> ())
> >
> > Also:
> >
> >   in ‘constexpr’ expansion of ‘assert_works<OE>()’
> >   error: ‘constexpr’ call flows off the end of the function
> >
> > Looks like
> >
> >   #if defined __cplusplus && __cplusplus > 202302L
> >
> > is not enough to check if this test is supported or not. XFAIL will not
> > help here because the test doesn't compile which causes make check to
> > fail.
>
> It does seems to me this is a compiler bug, I checked with gcc version
> 14.3.1 20250814
> and gcc version 15.2.1 20260112 without issues.
>
> Tomasz, is this a compiler bug or something has changes for gcc 14? If so
> I think we might need __GNUC_PREREQ for the test.
>
> Yes, this looks like compiler bug that was fixed in 14.3 and 15, but
wasn't able to identify
the specific change yet. I agree that we need a __GNUC_PREREQ for the
snippet
in question, so around these lines:

+#if __GNUC_PREREQ (14, 2)
template <typename Ts>
constexpr bool
assert_works ()
{
  return requires (Ts ts) {
    assert (ts);
  };
}

enum OE { oe };
enum TE : int { te };
enum class SE : int { se };

static_assert ( assert_works <OE> ());
static_assert ( assert_works <TE> ());
static_assert (!assert_works <SE> ());
+#endif

Let me know if I should prepare a patch with above change?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://sourceware.org/pipermail/libc-alpha/attachments/20260303/84a150bd/attachment.htm>