> What the problem tries to solve PR 3508?

The two test cases added demonstrate what was impossible to pickle before and is now.

> Swallowing all exceptions looks like an antipattern to me.

This is the only thing that we can do faced with custom `__getattr__()` implementations, especially when `copy.deepcopy()` creates new objects with `cls.__new__()`, something that most class implementers won't expect.

> Rather than failing and allowing the programmer to fix the picleability of its class, this can silently produce incorrect pickle data.

Can you give me an example where this would lead to incorrect pickle data?

> By swallowing MemoryError and RecursionError this changes the behavior of objects in an environment with limited resources: lack of memory or calling deep in the calling stack. This adds heisenbugs.

This is what `hasattr()` in Python 2 did.  This is why in Python 2 the `RecursionError` example I added to the tests was actually working just fine.