native doesn't crash on this test case right? so maybe we shouldn't either? (well i guess we didn't crash either, but hard to imagine releasing things in this order can cause uaf)
Maybe not this test case, but according to my undersatnding, doing more or less the same thing—releasing the ddraw before the device—has crashed native before. Since the device doesn't hold a reference to the ddraw I'm not convinced it makes sense to do so here. I think we just fix the relevant test to release the ddraw last.