How to write a concept for structured bindings?

While compiling the following (reduced) code:

#include <tuple>
#include <stdlib.h>

template<size_t N> struct tying;
template<> struct tying<1> {static auto function(auto& o) -> decltype(auto) {auto& [p1]    = o;return std::tie(p1);}};
template<> struct tying<2> {static auto function(auto& o) -> decltype(auto) {auto& [p1,p2] = o;return std::tie(p1,p2);}};

template<typename T, size_t N> concept bool n_components =
requires(T& object) {
    { tying<N>::function(object) };
};

typedef struct
{
    int   a;
    float b;
} test_t;

int main(int argc, char* argv[])
{
    constexpr size_t n = 1;
    constexpr bool   t = n_components<test_t, n>;

    printf("n_components<test_t, %d>: %s\n", n, t ? "yes" : "nope");
    return 0;
}

with gcc specific options -std=c++1z -fconcepts the gcc (version 7.3.0, and x86-64 gcc-trunk at godbolt also) fails with an error:

error: only 1 name provided for structured binding
note: while 'test_t' decomposes into 2 elements

I was very surprised by this, as the error "raises" inside the requires-expression, and to my understanding, this should result in n_components constraint to get evaluated to false, as the requirement is not satisfied.

Note: replacing constexpr size_t n = 1; with other values "fixes" the error, so I presume the n_components constraint is not to blame:

• replacing with n = 2 the n_components constraint evaluates to "true" as expected.
• replacing with n = 3 the n_components constraint evaluates to "false" as expected - due to referring to unspecialized tying<3> struct.

It seems there are no other compilers supporting the c++ concepts and structured bindings available yet to compare with.

PS. I was playing around with "poor man's reflection" a.k.a. magic_get and hoped to replace the unreliable is_braces_constructible trait with something more robust...