Remove redundant GIL acquisition during PyObject release

[atifaziz]

This PR removes a redundant GIL acquisition during the release of a PyObject.

[tonybaloney]

The GIL is required and I want to avoid a possible race condition from the check above.

The scenario is a custom __del__ method on a heap allocated class which would be called when the object is disposed and the GIL assertion would cause a hard crash. It's possible we're just not testing that right now (or aren't testing the race)

[atifaziz]

The GIL is required and I want to avoid a possible race condition from the check above.

The scenario is a custom __del__ method on a heap allocated class which would be called when the object is disposed and the GIL assertion would cause a hard crash. It's possible we're just not testing that right now (or aren't testing the race)

If it's this tricky, then I think we should add a comment, but I am still having a tough time seeing the race condition. The GIL class is designed with TLS in mind for avoiding nested acquisitions for the same thread. If GIL.IsAcquired returned ture, it means that the calling thread already has the GIL. If there's a thread-switch between the GIL.IsAcquired check and GIL.Acquire then it wouldn't matter because other threads would see their own state of GIL acquisition. When the interrupted thread resumes, the result of its GIL.IsAcquired won't have changed. Am I missing something?