Sending a code object back to the parent process and having it deal with
the situation sounds very unpleasant indeed! I think a completely
different type of solution may be possible.

In general, I can't think of any reason for IDLE to pickle "user
objects" from the subprocess and send them to the parent process; it
should merely send back the output (as strings), with special cases for
exceptions and such. By "user objects" I mean objects "inside the
interpreter", as opposed to those used by IDLE itself.

I'll have to unwind the spaghetti in rpc.py, run.py and PyShell.py a bit
more to propose a specific set of changes; I hope to get to that tomorrow.