Resampler: IPython 5.0 triggers wrong FutureWarning

[jorisvandenbossche]

With the latest IPython:

In [1]: rng = pd.date_range('20130101', periods=10, freq='T')
   ...: ts = pd.Series(np.arange(len(rng)), index=rng)

In [3]: r = ts.resample('2T')

In [4]: r
Out[4]: c:\users\vdbosscj\scipy\ipython-joris\IPython\utils\dir2.py:65: FutureWa
rning: .resample() is now a deferred operation
use .resample(...).mean() instead of .resample(...)
  canary = getattr(obj, '_ipython_canary_method_should_not_exist_', None)
DatetimeIndexResampler [freq=<2 * Minutes>, axis=0, closed=left, label=left, con
vention=start, base=0]

In [5]: import IPython

In [6]: IPython.__version__
Out[6]: '5.0.0'

In [7]: pd.__version__
Out[7]: '0.18.1+184.g15c2a3a'

while using IPython 4.2:

In [1]: rng = pd.date_range('20130101', periods=10, freq='T')

In [2]: ts = pd.Series(np.arange(len(rng)), index=rng)

In [3]: r = ts.resample('2T')

In [4]: r
Out[4]: DatetimeIndexResampler [freq=<2 * Minutes>, axis=0, closed=left, label=l
eft, convention=start, base=0]

In [5]: import IPython

In [6]: IPython.__version__
Out[6]: '4.2.0'

In [7]: pd.__version__
Out[7]: '0.18.1+184.g15c2a3a'

[jorisvandenbossche]

@takluyver This seems to be related to ipython/ipython#9289 (previously, it was first checked if method is a callable, and only then it was checked if it could retrieve a non-existing attribute, while now the callable check is moved to the end of get_real_method).
The problem is that the deprecation warning is only triggered once you "do" something with it, and now it is triggered by displaying its repr

[takluyver]

That's unfortunate. But if our check for the canary attribute triggers this, won't our checks for _repr_pretty_, _repr_json_, etc. also trigger it?

[jorisvandenbossche]

Apparently not. Where are those checks in the IPython code?

It does trigger the warning if I manually do getattr(r, '_repr_json_'), but this does check does not then not seem to occur when displaying an object.

[takluyver]

IPython.core.formatters - they're somewhat hidden in infrastructure, but if you search for print_method in that module, it should be possible to work out what's going on.

[jorisvandenbossche]

Ah, I was in a terminal, so I suppose there only the plain text displayer will be used?
If I try it in a notebook, but using IPython 4.2, the same issue already shows up

[takluyver]

I thought the same machinery would check for _repr_pretty_ even in the terminal, but maybe it's done a bit differently.

[jorisvandenbossche]

@takluyver We solved it at our side! (at least, @chris-b1 did :-) )