This bug has grown worse. A lot worse. Good news: this now compiles:

struct foo {
    int n;
    void baz( ) {
        void qux(T)() {
            n = 3;
            assert(n == 3); // Passes
        }
        qux!int();
    }
}

void main( ) {
    foo b;
    assert(b.n == 0); // Passes
    b.baz();
    assert(b.n == 3); // Fails
}

As we can see, the change in n is not properly propagated. What's even worse, is this:

struct foo {
    int n;
    void baz( ) {
        void qux(T)(void* a) { // Added parameter
            n = 3;
            assert(n == 3); // Passes
            assert(a == &this); // Added.
        }
        qux!int(&this);
    }
}

void main( ) {
    foo b;
    assert(b.n == 0); // Passes
    b.baz();
    assert(b.n == 3); // No longer fails.
}

You can then remove the assert on line 7 (a == &this) to get the old behavior back, the additional parameter changes nothing on its own.