There are still sites in the CPython code where all errors of failed attribute lookup are silenced or overridden by other exception. This can hide such exceptions like MemoryError, RecursionError or KeyboardInterrupt and lead to incorrect result (as the attribute was just absent instead of looking it up was interrapted by side causes). Only AttributeError is expected to signal an absence of an attribute, and only it can be silenced.

The proposed PR fixes most of such cases.

There are still few sites where *all* errors are ignored (for example when report an error to the stderr or like), they should be considered separately.