GeezerVison

Note, that similar code could be shared also with test_long.py and test_float.py. That refactoring of the scope here, but maybe in this pr you could put these classes somewhere in the support module? // see #110956

I proposed this in #84310. But for some reasons this idea was not liked, so now we need to duplicate the code in many test files. cc @rhettinger

This pattern is quite common. Maybe there should be helper method like:

    def assertComplexesAreIdentical(self, x, y):
        self.assertFloatsAreIdentical(x.real, y.real)
        self.assertFloatsAreIdentical(x.imag, y.imag)

Yes, I seen this in your PR, and this looks helpful. There are also similar methods (with different names) in other test files. Perhaps we could move the best implementation in a common file and use it everywhere.

Naming nitpick: the Mock prefix on these classes doesn't feel appropriate: nothing's being mocked here. I'd probably go with something like ImplementsIndex or HasIndex.

I once added common FakePath class which implements the __path__ method, and it is now used in many tests for the path protocol. I proposed to add similar FakeIndex, FakeFloat, etc, but Raymond rejected this idea. Here I use the Mock prefix which looks more neutral and test specific than Fake. Different tests currently use similar classes with different names. The most common prefix is perhaps My. Prefix Supports cannot be used because it can conflict with typing.SupportIndex, etc. Suffix Like can conflict with os.PathLike.

I see that several tests use the With prefix: WithStr, WithRepr, WithIterAnext, etc. What do you think about WithIndex?

WithIndex works. I'm also happy for this to be left as-is - it's not a strong objection, just a niggling feeling that the name isn't quite right.

We could drop the parenthetical here, given that we're now requiring IEEE 754 for CPython.

Interesting - looks like these tests were already assuming IEEE 754 before.

@serhiy-storchaka, maybe it's time to drop this? Ditto for float's.

Returning non-complex(float) values was deprecated in v3.7.

This is a different issue.