Deprecate @ syntax, as with ~ syntax.

[alexcrichton]

We currently still allow @ syntax without allowing ~ syntax. We should apply the same lang-item approach for ~ to @ and start using Gc<T> everywhere.

We would also be able to remove the feature gate because the Gc structure would be marked #[experimental]

Nominating.

[huonw]

Would it be unreasonable to just remove @ entirely?

(Temporarily implementing Gc<T> as a wrapper around Rc<T>? Or even just removing it too, it's doing nothing useful in its current form: mainly just pretending that Rust actually has a GC.)

[alexcrichton]

I would think that because Gc<T> collects cycles when the task ends that we'd need to keep the infrastructure for now.

I believe that there are good portions of the compiler that are leaking their memory currently because Rc<T> will leak on cycles, and it would be a shame to impose something like this on servo as well. This is mostly just a guess, though.

[huonw]

Hasn't servo moved to using the spidermonkey GC?

[pnkfelix]

sounds fine to just shift the existing infrastructure over to Gc<T>.

Assigning P-low; not a 1.0 blocker so not on the 1.0 milestone.

[huonw]

For reference, grep doesn't reveal any uses of @ (as a GC'd pointer) in Servo.