Should logging.Handler.lock allow None?

[gsnedders]

Since python/cpython@74723e1 (in CPython 3.13 and later), a variety of calls of Handler.acquire() and Handler.release() got converted into with Handler.lock.

As a result, unless you're overriding Handler.handle, Handler.lock cannot be None, which it could be previously.

Note that logging.NullHandler does actually set lock to None in the stdlib, but it does override Handler.handle.

I’m unsure about the policy for typeshed is about instance variables whose valid values depend on whether a particular method is overridden.

[JelleZijlstra]

Sounds to me like the attribute can indeed be None, and our stubs should reflect that.