[slozier]

_imp.acquire_lock was only counting and not blocking, so replace it with a real lock (test_lock).

importlib._bootstrap was modified to resolve issues:

• _SpecMethods.load on 3.4 could raise an exception on a None spec, leaving the lock acquired.
• _get_module_lock set a callback on the weakref, which runs on the finalizer thread: on 3.4 it could delete a newer lock, and on 3.6 it took the lock, which could deadlock with gc.collect() (test_lock_gc_collect).

[BCSharp]

The code uses a private lock object. Maybe rather than using System.Object unconditionally, use System.Threading.Lock on .NET 9.0 and greater? See src/dlr/src/core/GlobalUsings.cs. I'd like to introduce this mechanism in all suitable places in IronPython as well, but we have to start somewhere.

[slozier]

Maybe rather than using System.Object unconditionally, use System.Threading.Lock on .NET 9.0 and greater

Sure, I can do that

[slozier]

@BCSharp I didn't go with the global usings approach but used a polyfill instead. Might be worth doing the same in the DLR now that we're using polyfills there...

[BCSharp]

Might be worth doing the same in the DLR now that we're using polyfills there...

Sure, I can look into it. If I remember correctly, the polyfill at the time had some inefficiencies so I didn't take it.