Template function inside template inhibits inferring arguments when called

[TurkeyMan]

We all know about this, but I really reckon it's a bug and should be fixed.

template fun(T, U)
{
    void fun(T x, U y) {}
}

template fun2(T)
{
    void fun2(U)(T x, U y) {}
}

template fun3(T)
{
    alias fun3(U) = fun!(T, U);
}


void test()
{
    fun(1, 2); // arguments are inferred
    fun!int(1, 2); // T is given, but U is still inferred

//    fun2(1, 2); // ERROR: T is NOT inferred
    fun2!int(1, 2); // T is given, and U is inferred (inference works 'through' an outer template here)

//    fun3!int(1, 2); // ERROR: T is given, but U is NOT inferred?
    // why is this different than the `fun!int` case?
    // if you say the inference didn't work 'through' the outer template, then how do we explain the `fun2` case?
}

I feel like all the cases shown above should work, including the commented out ones.
Especially the fun3 case; once the outer template is applied, the situation should be the same as fun!int...

Basically, as I see it now; the first case with fun feels like a hack in the language, where the hack was applied to a specific edge case, rather than this concept being implemented seamlessly.

[pbackus]

The fun3 case is just Bugzilla 1807 again.

[Herringway]

The fun3 case is just Bugzilla 1807 again.

(#17667)