Mark `Id` as thread-safe about idcontain.rs

[cristicbz]

Thanks a lot for the thorough post! I think this makes sense; the only issue is that I do not want to introduce any unsafe (which this would do for the manual impls for Send and Sync).

We could use &'static T, but that would require the T: 'static bound, which isn't great. What about doing PhantomData<fn(T) -> T>: this makes Id<T>: Send + Sync and invariant in T which I think is a good idea.

[ElusiveMori]

Thanks for the swift reply!

I think PhantomData<fn(T) -> T> may look like a bit of a hack, but I definitely understand trying to avoid unsafe, and I personally can't think of any more alternatives. It would probably be worth leaving a comment about this in the code, too.

I will test this approach in my code (which requires Id<T> to be thread safe and without lifetime bounds), verify that it compiles, and report back.

[ElusiveMori]

Another variant, I think, could be PhantomData<Arc<Mutex<*const T>>>, but it is hardly less hacky than just PhantomData<fn(T) -> T>, although it could be argued that it conveys the intent better:

• Arc + Mutex provide Send + Sync traits
• *const T ignores lifetimes.

[cristicbz]

I don't think fn(T) -> T is that hacky; it's what https://doc.rust-lang.org/nomicon/phantom-data.html recommends for non-lifetime bound invariance.

You could argue fn() -> T is actually how an Id behaves: it's a function that given an IdSlab it returns a T.