Re: [PATCH] math: Fix unsupported check in test-narrowing-trap

On Thu, Jul 16, 2026 at 12:13:58PM -0300, Adhemerval Zanella Netto wrote:
> 
> 
> On 16/07/26 12:03, Yury Khrustalev wrote:
> > On Thu, Jul 16, 2026 at 12:00:16PM -0300, Adhemerval Zanella Netto wrote:
> >>
> >> On 16/07/26 11:01, Yury Khrustalev wrote:
> >>> ---
> >>>
> >>> Base commit: af51ed63d0
> >>>
> >>> Test change only. Passes regression on aarch64-linux-gnu,
> >>> x86_64-linux-gnu, i686-linux-gnu and build-tested for
> >>> several targets with build-many-glibcs.py.
> >>>
> >>> Andreas, I suppose we could merge this test fix now? It's OK if not.
> >>
> >> Is this fix for loongarch? If so I think you can check with cfarm401
> >> machine [1] (for some reason cfarm400 is not accepting my keys).
> >>
> >> [1] https://portal.cfarm.net/machines/list/
> > 
> > The fix is for aarch64 actually. Some hosts support these tests and
> > in this case the 'math/test-narrowing-trap' test was incorrectly
> > reported as UNSUPPORTED.
> Right, so maybe add some context on the commit message. Another option
> would to rewrite EXCEPTION_ENABLE_SUPPORTED on aarch64/arm to actually
> test whether exceptions are support (since on ARM seems to have the
> concept of being 'optional').

The fix is not aarch64-specific though. All other tests that use this
functionality use return value of feenableexcept before they decide if
the test is supported or not. In the test-narrowing-trap test, however,
the order of checks was wrong. As a side-effect of this I see spurious
UNSUPPORTED math test on some aarch64 hosts. I've asked the authors of
the these tests to fix it, but they haven't, hence this patch.