Add a marker trait for non GC types

[ZainlessBrombie]

It would be nice to have the following:

1. An unsafe trait called "GcDeadEnd" (or something), marking something as having no Gc dependencies.
2. impl GcDeadEnd for e.g. String, primitives, Rc, RefCell etc.
   3) impl Trace for GcDeadEnd

Cheers! :)

[Manishearth]

Currently we just achieve this by a blank Trace impl. A problem is that coherence kinda rears its head if you add a Trace blanket impl based on GcDeadEnd.

[ZainlessBrombie]

Sorry, could you elaborate what you mean by "coherence kinda rears its head"?

[ZainlessBrombie]

Additional note: There would be impl Trace for Rc where T: GcDeadEnd
(Things like Gc)

[Manishearth]

Sorry, could you elaborate what you mean by "coherence kinda rears its head"?

You'd need a blanket impl Trace for GcDeadEnd impl, and this locks out certain Trace impls from existing due to coherence.

Additional note: There would be impl Trace for Rc where T: GcDeadEnd

You won't be able to have both this and an impl Trace for non-deadend contents.

[ZainlessBrombie]

Right thanks. I realized:
A) impl Trace for GcDeadEnd is not needed. We only need:
-> impl GcDeadEnd for e.g. String
-> seperate Trace for String
-> impl GcDeadEnd for e.g. Rc<GcDeadEnd>
-> impl Trace for Rc<GcDeadEnd>
B) This still means that we need specialization. (The soundness of which has not been entirely figured out unfortunately)

But with specialization it works in my fork :)
Edit: Just in case you are wondering: Lazy me just included #![feature(specialization)] unconditionally in my fork because I didn't put a #[cfg] on the default-impls ^^

[ZainlessBrombie]

Nevermind, implemented my own Gc using safe non-experimental Rust only over at https://github.com/ZainlessBrombie/rs-safe-gc
Still missing the derive macro, but it's 3 am ^^
It does not encounter the problem mentioned here and allows moving out of Gc'ed values, a major painpoint for me.

🎉 c:
Either way, hats of to you and all of the Rust core team, I'm already addicted haha ;)

[ZainlessBrombie]

Ah right I forgot: I'm running an MIT license and used some small snippets of your code (macros, thanks!), is that compatible with your license? I think not... off the top of your head, do you know if that's much of a problem here?

[Manishearth]

is that compatible with your license? I think not... off the top of your head, do you know if that's much of a problem here?

yeah that's fine

[andersk]

Even without specialization, a GcDeadEnd trait could still be used for a checked version of manual unsafe_ignore_trace annotations, right? (I recognize unsafe_ignore_trace is not literally unsafe as per #52, but there’s still value in checking that these annotations don’t cause leaks.)

[Manishearth]

Even without specialization, a GcDeadEnd trait could still be used for a checked version of manual unsafe_ignore_trace annotations, right?

Proc macros don't have access to trait information, though I guess we could generate assertions if people are still annotating them.