I think that's just a bug in GCC. I get the expected SFINAE behavior here:
https://godbolt.org/z/PKE64rvMz

On Wed, Sep 16, 2026 at 12:35 PM Bingzhi <bingzhi2025_at_[hidden]> wrote:

> Thanks. So I modified a bit:
> static constexpr auto result = evaluate();
> template <class T> concept C = requires { typename [: X<T>::result :]; };
>
> At 2026-09-17 00:27:53, "Brian Bi" <bbi5291_at_[hidden]> wrote:
>
> There's never SFINAE for invalid alias declarations.
>
> *Brian Bi*
>
> On Wed, Sep 16, 2026, 12:20 PM Bingzhi via Std-Discussion <
> std-discussion_at_[hidden]> wrote:
>
>> template <class T>
>> struct X {
>> static consteval std::meta::info evaluate()
>> { if (std::meta::is_same_type(^^T, ^^int)) return { }; else ... }
>>
>> using type = [: evaluate() :];
>> };
>> template <class T> concept C = requires { typename X<T>::type; };
>> static_assert(!C<int>); // *error: *cannot splice a null reflection
>>
>>
>> I don't know why it is. Can somebody explain?
>> --
>> Std-Discussion mailing list
>> Std-Discussion_at_[hidden]
>> https://lists.isocpp.org/mailman/listinfo.cgi/std-discussion
>>
>

-- 
*Brian Bi*