tab-complete can mutate objects!

Consider the class.

class X:
    T_cnt = []

    @property
    def T(self):
        self.T_cnt.append(len(self.T_cnt))

Thus X.T_cnt keeps track of the number of times that X().T was called.

The following demonstrates the bug:

>>> x = X()
>>> x.<tab>
>>> x.<tab>
>>> X.T_cnt
[0, 1]

Each time I use <tab> to bring up the list of attributes of x, x.T is called under the hood. This occurs in both the ipython console and Jupyter notebook.

I am using iPython 6.1.0

[takluyver]

It’s hard to do any useful introspection if we guarantee that tab completion has no side effects, because Python allows objects to override so much of their behaviour: almost anything we do might have side effects.

So we’ve generally tried to ensure that tab completion doesn’t do things which are likely to have side effects (like calling regular methods), but allowed it to do things which usually don’t (like attribute access).

I generally consider it bad design for property getters to have non-trivial side effects, because it’s surprising both for introspection software and developers when that happens on attribute access.

[Carreau]

I’ll have a look, but not soon. With PyBay comming and JupyterCon just after it is tough.

It’s on my todo list though.

On Mon, Jul 31, 2017 at 7:28 AM, Thomas Kluyver [e-mail] wrote:

Thanks 😃 It does sound like we (well, @Carreau https://github.com/carreau) might be able to improve on this. Or might even already have done so.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/ipython/ipython/issues/10720#issuecomment-319083685, or mute the thread https://github.com/notifications/unsubscribe-auth/AAUez9kQVTL3S0kt2-EbnJ1mEOsjKsB_ks5sTeSngaJpZM4Onaw5 .