Typechecker should stop this

[eholk]

tag foo = int;

fn main() {
   let x = foo(1);
   if x == -1 { }
}

This program should not compile, but it does.

[msullivan]

That seems legitimate. We do autoderef on binary operators.

[eholk]

I guess maybe the real solution is that we need proper abstract types. Still, I question whether autoderefing tags makes sense. Tags seem very different from pointers.

[marijnh]

Autoderef on newtype-tags is important in at least one case: Using such a type to break a recursive type. You'll wrap your record in a newtype in such a case, and it is hugely convenient to be able to do val.field instead of (*val).field on wrapped values.

Maybe we should just stop autodereferencing for binary ops? That's not a feature I would expect.

[eholk]

I think removing autoderef for binary ops is a good idea. As Sully was pointing out earlier today, it limits our ability to do type-inference for binary operations.

[pcwalton]

As written, this works as intended. I'd personally be in favor of removing autoderef too, but IMHO it should be a separate bug.

[kevina]

I personally think that for true pointer types Rust should auto-deference more. In 99.99% of the cases I do not care if something is a pointer or an actual value.

I would like to see the case for single constructor enums separated out and handled separately instead.