Type parameters on statics

[brendanzab]

Would this be possible?

pub enum Reply<V, StopReason> {
    Reply(V),
    Stop(StopReason),
}

pub static Continue<StopReason>: Reply<(), StopReason> = Reply(());

I guess it would be less useful than the C++14 proposal seeing as our constexprs are currently so limited.

[nikomatsakis]

This is...basically identical to writing

fn Continue<S>() -> Reply<(), S> { Reply(()) }

except that you don't need the ().

In general I'd say it's not "possible" in the sense of static things representing a value that is pre-allocated and lives in RO memory.

[nikomatsakis]

Closing as impossible, re-open if you disagree.

[huonw]

Out of curiousity, could this be implemented by monomorphising statics like we monomorphise functions? i.e. Continue doesn't appear in the executable, but things like Continue<u32> and Continue<Foo> do, if it's used in places where the type parameter is given those types.

[glaebhoerl]

@huonw that's exactly what the C++ proposal does (see link in OP).

I don't know whether there's compelling use cases as of yet -- it took C++ until now to get around to wanting them, and their templates and constexprs are much more sophisticated than Rust's -- but I think it's a reasonable idea.

[huonw]

One instance where it would be useful would be making a static initialiser for AtomicPtr... not sure if this is worth the trouble of a whole new language feature.

[brendanzab]

@nikomatsakis One issue is you can't pattern match on function calls. :(

[nikomatsakis]

Yes, I suppose we could monomorphize on statics, I'll reopen. I was lacking in vision.

[glaebhoerl]

A real use case, not quite the same but probably requiring similar things, would be static var: T in generic functions.

[alexcrichton]

Issues of this kind now belong in the RFC repo, so closing.

[mbrubeck]

Similar to rust-lang/rfcs#2130