Locking strategy is wrong

[ztane]

Now that the locking cached_property was adapted into Python standard library, I'd note here too that the threaded_cached_property locking strategy is wrong for many use cases, and I've written an issue in bugs.python.org.

The reason is that a descriptor is not instantiated for every object instance - they indeed exist only once for each class. Thus if there are multiple instances of the class, then all these would share the same one lock, even though the caches are independent. And if there is only one instance of a class then it is rather Pythonic to not write a class at all...