Avoid panics when exiting thread

[ghost]

If we try to call pin() when exiting the thread, it is possible that we'll access the thread-local HANDLE after it has already been destroyed.

This is sometimes a problem when using crossbeam-epoch while the thread is unwinding from a panic. In that case, double panics cause aborts.

To fix the problem, we register a temporary Handle in case HANDLE has already been destroyed.

[ghost]

Note: servo/servo#21325 is waiting on this PR.

[Thomasdezeeuw]

Maybe FnOnce would be better, since it's only called once anyway? This might require a slight change in the code however.

[ghost]

Unfortunately, that doesn't compile and I don't know how to do better.

error[E0382]: capture of moved value: `f`
  --> src/default.rs:49:50
   |
49 |     HANDLE.try_with(|h| f(h)).unwrap_or_else(|_| f(&COLLECTOR.register()))
   |                     ---                          ^ value captured here after move
   |                     |
   |                     value moved (into closure) here
   |
   = note: move occurs because `f` has type `F`, which does not implement the `Copy` trait

[Thomasdezeeuw]

Sorry I thought that it might work with a match statement, however after trying it I see that it doesn't work.

[Thomasdezeeuw]

But I think you can pass f directly to try_with right? It doesn't need an extra closure.

[ghost]

We need an extra closure. :) Otherwise we get this:

error[E0382]: capture of moved value:
  --> src/default.rs:49:39
   |
49 |     HANDLE.try_with(f).unwrap_or_else(|_| f(&COLLECTOR.register()))
   |                     -                 ^^^ value captured here after move
   |                     |
   |                     value moved here
   |
   = note: move occurs because `f` has type `F`, which does not implement the `Copy` trait

[Thomasdezeeuw]

Is this only meant to not abort? If so a comment explaining that would be useful, now the test doesn't really make much sense.

[Vtec234]

This is removed in #84.