infer "of" narrowing

Let’s face it: it’s totally weird that this is well-typed:

alias Bool => \Itrue|\Ifalse;
void fun(Bool bool) {}
Boolean t = true;
fun(t of Bool);

But this is not:

alias Bool => \Itrue|\Ifalse;
void fun(Bool bool) {}
Boolean t = true;
fun(t);

In such cases, instead of producing an error, we could, completely safely and reasonably, just infer the of Bool bit. I’m conflicted about that because while it works perfectly nicely here, I guess it would create the expectation in the user’s head that \Itrue|\Ifalse and Boolean are the same type, which is an expectation that is violated as soon as these types are used as type arguments. (A List<Boolean> is not a List<\Itrue|\Ifalse> in Ceylon!

[gavinking]

Even if the compiler didn’t infer the of Bool, it could still generate an error message which suggests it as a possible fix!

[jvasileff]

Ah, that’s the one. It would be nice to dial it back a bit for refinement:

class A() {
    shared \Itrue | \Ifalse foo() => true;
}
class B() extends A() {
    shared Boolean foo() => true;
    // type of member must be assignable to type of refined member 'foo' in 'A':
    // the assigned type 'Boolean' is covered by, but not assignable to, the type
    // 'true|false' (explicitly narrow assigned expression using 'of true|false')
}