Some background context before I comment on the patch: Printing warnings to a console is a flawed idea. It does not work on Windows when Idle is run with pythonw and there is no console to write to. This is the normal case on Windows and, I suspect, the majority of all Idle uses across all systems. The attempt to write (to sys.__stderr__ == None) raises OSError, which shuts Idle down unless caught. (Thomas experiences exactly this when Python attempted to write the traceback.)  Either way, there is no message to be seen. Even when writing to console is successful, the message will often be hidden by Idle windows. Maybe it will be seen later, but  possibly not. 

Idle is a gui program and should not assume a text console. We should use tk messages boxes instead. Or perhaps put warnings in an Output Window (either instead, or in addition). Or put an error.log in .idlerc/ (this would be 'in addition'. Or also write to console when there is one, but not as the only message delivery system.

For problems in config files, I think we should, if possible, fix, rename, or delete the file. In any case, make it so the same warning will not appear next time Idle it is started. If we do anything, the warning message should say what was done.

There are 6 places in configHandler with code like the following:
            if  <config problem>:
                # do default, print warning
                warning=...
                try:
                    sys.stderr.write(warning)
                except OSError:
                    pass

(In one of the 6, Idle exits instead of catching OSError; there is another issue to fix this.)  These blocks mix together situation-specific message creation with a common and flawed delivery system. The latter should be factored out and fixed in a separate function at the top of the file (which I can write a first version of).

def config_warn(message): ... 

The problem with the patch is that it does not follow existing configHandler code. First, it stores errors and shuffles them off to be dealt with in PyShell. They should instead be immediately dealt with where detected. Second, it uses idle_showwarning to format and emit the warning.  Formatting with idle_formatwarning is ok, but idle_showwarning is designed specifically for the shell, ending warnings with the shell prompt '>>> '. It also has the flaw of the code above.  By default, it sends messages to an often non_existent sys.__stderr__. The patch should use the new config_warn function.

I will look at the new tests later.

When GSOC is done, I want to look as Tal's extension config dialog. That should reduce errors from users hand-editing config-extensions. Saimadhav has also done some work on key validation that needs to be applied to configHandler and configDialog. That should reduce key errors a bit.