For the record, I don't think the weakref support is likely to be particularly useful either, and if we'd used __slots__ from the beginning, I expect we would have left it out.

It's specifically the idea of *taking weakref support away* when weakrefs currently work that seems dubious to me.

It may be worth asking the question on Python dev, as it's possible I'm being overly cautious, and it would be reasonable to drop the weakref support in 3.5.0, while mentioning in the 3.5 porting notes that we'd be open to the idea of adding weakref support back in a 3.5.x maintenance release if anyone reports its removal as a regression from 3.4.