Add a `TypeID` intrinsic

[Kimundi]

It's needed for things like an Any type, or safe dynamic loading.

It basically needs to be an generic intrinsic that returns some kind of data structure that uniquely identifies the generic type it has been instantiated with. Ideally it also includes information about the crate itself and it's meta data, so that two different named and/or versioned crates with the same type result in different TypeIDs.

Currently, the static pointers to type descriptors can be kinda used for this, but there is no guarantee that a type descriptor can not be duplicated, so using them for this purpose is incorrect.

Possible usage could look like this:

fn same_types<T, U>() -> bool {
    let type_id_t = intrinsics::type_id::<T>();
    let type_id_u = intrinsics::type_id::<U>();

    type_id_t == type_id_u
}

fn main() {
    assert!(same_types::<uint, uint>());

    let type_id = intrinsics::type_id::<std::util::Void>();
    assert!(type_id.path().to_str() == "<6df45e453>::std::util::Void");
    assert!(type_id.crate().name() == "std");
    assert!(type_id.crate().version() == "1.0");
    assert!(type_id.crate().hash() == "6df45e453");

}

See also https://gist.github.com/Kimundi/6802198 for some prototype implementation of a TypeID using type descriptors.

[brson]

Nominating.

[orenbenkiki]

Comparing TypeId for equality is great, but it would also be very nice if it were possible to access the names of the types, especially for error messages. Would it be possible for TypeId to offer a name() -> &'static str (or maybe ~str) method?

I know unique names of types across different crates are tricky, but even if given only the unique name within the defining crate, this would be extremely useful for debugging dynamic type errors.

[alexcrichton]

I don't think that the type id is the right location for the name, but there is a name field on the TyDesc, and that can be called for arbitrary types. Additionally, you may also be able to use the TyVisitor to build the name of a type.

[orenbenkiki]

Fair point... that said, it seems silly we have TypeId and TyDesc and no relationship between them. Doesn't it make sense to unify them both (e.g., add Eq to TyDesc)? Or, barring that, at least provide an easy way to get one from the other?

[alexcrichton]

That's not a bad idea unifying them. I think right now the blocker for that is that type_id is only defined for T: 'static where get_tydesc is defined for all T. If we can lift that restriction, I think it would be a bit easier to unify the two of them, but for the time being it may be difficult to do so.