[PATCH] math: Fix errno setting for narrowing sqrt functions on underflow [BZ #33393]

[Sam James]

Osama Abdelkader <osama.abdelkader@gmail.com> writes:

> The CHECK_NARROW_SQRT macro in math-narrow.h was not setting errno to
> ERANGE when an underflow exception occurred during narrowing square root
> operations. While the FE_UNDERFLOW exception flag was correctly set, errno
> remained unchanged, violating IEEE 754 and POSIX requirements.
>
> This fix adds a check for the FE_UNDERFLOW exception after computing the
> result. When underflow is detected via fetestexcept(FE_UNDERFLOW), errno
> is now properly set to ERANGE.
>
> The issue affects all narrowing square root functions that use this macro:
> - fsqrt (double to float)
> - fsqrtl (long double to float)
> - dsqrtl (long double to double)
>
> Tested on x86_64-linux-gnu. All tests pass including the new test case
> that verifies errno is set correctly on underflow conditions.

What test case?

>
> Signed-off-by: Osama Abdelkader <osama.abdelkader@gmail.com>
> ---
>  math/math-narrow.h | 2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/math/math-narrow.h b/math/math-narrow.h
> index 2f27b736fd..89bbcc26f1 100644
> --- a/math/math-narrow.h
> +++ b/math/math-narrow.h
> @@ -307,6 +307,8 @@
>  	}					\
>        else if ((RET) == 0 && (X) != 0)		\
>  	__set_errno (ERANGE);			\
> +      else if (fetestexcept (FE_UNDERFLOW))	\
> +	__set_errno (ERANGE);			\
>      }						\
>    while (0)