Message12418

The more I try to increase test coverage, the more dubious I get about the way we have added the Python mapping API to Java Map objects.

I've noticed that our modified version of test_dict, while it uses an instance of the type under test some of the time, obtained from self._make_dict(), where it refers to the type itself, still refers to dict. This happens particularly where it needs to define a sub-class of the type under test. Or rather used to. But correcting this leads to many more test failures.

Sub-classing works really badly: test_fromkeys uses subclassing extensively, but the overrides of __new__ and __setitem__ do not get called. This should probably not be a surprise, as there is no *Derived class to hook the special methods. At least, I don't think there is.