I agree that testing GUI apps is problematic, especially if you are
working with an intermediate version which isn't being used heavily.
Besides code review, you essentially have to run at least a subset of an
user-conducted acceptance test which tries to explore the corner cases
and generally monkey test the app.  It seems the best way beyond that is
to use the new version of IDLE to develop code for awhile; that usually
exposes at least the superficial problems.  I don't have an issues with
correcting minor bugs, but I expect the corrective patch to be small.
We have one going on CallTips which has become problematic for several
reasons, and I'll comment on it at msg59009.

Unless a person wants to fork, he has to make a deliberate effort to
stay close to the main line of development.  That can mean forgoing
things that would be 'nice to have' or non-essential refactorings. It's
a problem with distributed VCS that divergence is encouraged.  I think
you see the problem!

Are the differences in your autocomplete code significant functional
differences, or can we bring our versions into alignment at some point?
Please take into account the comments I'm going to make at msg59009
regarding 2.6/3.0 compatibility.

I really do appreciate your interest in IDLE and the work you are doing
to make it better.  I think you understand my interest in keeping it
minimal and intuitive.  It doesn't have to be like all the rest and
continually accrete features which are of limited use to the target
audience, especially at the cost of reliability and performance.