Currently for every class that defines __getattr__ to retrun a callable, I have to define these to work around ipython: __repr__, _getAttributeNames, __methods__, trait_names.
First three could be worked around by disallowing underscore names in __getattr__, but there's hardly any generic logic that could exclude trait_names.
I think that if ipython thinks my object might be trait-able (whatever that is) it should first interrogate some underscore method or type or something else.
In any event ipython cannot arbitratily call an innocently named method like trait_names() because ipython doesn't know what side effects that might cause.
Example:
# usage:
# server = rpc_caller("http://....")
# rv = server.get_customer_info(arg1, arg2, ...)
class rpc_caller:
def __init__(self, base):
self.base = base
def __getattr__(self, name):
return lambda *args, **kwargs: do_rpc_call(self.base, name, args, kwargs)
# initialize some caller
rpc = rpc_caller("http://...")
after this, tab-expansion in ipython results in aweful side-effects, because ipython eventually calls rpc.trait_names() which the system is completely unprepared for.
also globals() may fail with exception
another case where this caused a serious bug for me was when __getattr__ returned a gtk widget
last checked against ipython 0.8.1
