exit( and quit( in Qt console produces phantom signature/docstring popup, even though quit() or exit() raises NameError

[dwf]

A very minor oddity I noticed in master while running the (very impressive) Qt console. Title pretty much says it all. Being unfamiliar I typed "quit(" (which is how I usually quit on the ancient IPython 0.8.3 installed at the lab) and was given an incorrect information about its call signature (but told it had no docstring). Attempting to call either raises a NameError.

[takluyver]

Two issues there. At a guess, the tooltip is popping up because the information request is being handled without the modification to the builtins that we make for execution (see IPython/core/builtin_trap.py).

As for calling either: we override exit and quit with magic commands, so you can just type exit. It would be good to do something more helpful if you try to call a magic command (exit()) though.

[dwf]

Yep, I figured that last part out. I guess in general it would be good to detect collisions between builtins and magics that shadow them if you try to call them, but giving the user false information via incorrect call tips is probably of more immediate concern. :)

I'm not sure of the right place to it though. I think I've got a pretty good idea of what's going on (client passes object_info message and replies with the result of object_inspect from InteractiveShell), but there doesn't seem to be a "right" place to deal with this magic corner case.

It seems that what needs to happen is either that first message has to carry back some information with it (specifically whether the shell instance is in automagic mode and has a magic attribute that shadows a builtin with that name, etc.), or there needs to be a more messages exchanged. I am sufficiently ignorant of the design philosophy you guys have taken to know exactly what the right approach is, though.

[takluyver]

Automagics are, I believe, on or off at the kernel level, so we don't need any more info for that. I think we just need ensure that object_info requests are handled inside the builtin_trap context manager; that way it won't find the builtin exit or quit functions. But the kernel-client model isn't something I've worked on a lot.

[dwf]

You're right. I'm just sufficiently unfamiliar with context managers that I didn't see that as an obvious solution. :) I've submitted a pull request.

[minrk]

This was closed by PR #339