Tab completion in IPython 6 on Categorical shows warning

[wesm]

With pandas 0.20.1, IPython 6.0.0 (stack from conda-forge) I have:

In [31]: obj = pd.Categorical(['foo', 'bar', 'baz'])

In [32]: obj.<TAB>
/home/wesm/miniconda/envs/book-env/lib/python3.6/site-packages/jedi/evaluate/compiled/__init__d
  getattr(obj, name)

I'm not sure the origin or meaning in the output, but it's a bad smell that may confuse users

[TomAugspurger]

IPython 6 switched over to jedi for completion, which must execute at least some code? Categorical.labels is a property. It looks like we can get around this by specifying __dir__.
I think I've seen it in at least one other place.

@Carreau does this sound like an IPython bug? For a simpler example:

In [32]: import warnings

In [33]: class Foo:
    ...:     @property
    ...:     def bad(self):
    ...:         warnings.warn("bad")
    ...:     @property
    ...:     def good(self):
    ...:         return "good"

In [34]:

In [34]: f = Foo()

/Users/taugspurger/Envs/pandas-dev/bin/ipython:4: UserWarning: bad
  import re
In [35]: f.<TAB>

If so I'll file one on IPython. Either way I'll put up a PR to silence these using __dir__.

[Carreau]

Yes it's a Jedi bug. There are 2 other issues open in IPython and spark. Tab completion triggers jobs on spark. I'll try to have a look this week, but we'll need a release of Jedi. You can deactivate Jedi completion though, i'm on my phone so can't tell you the option right now.

…

On May 21, 2017 14:28, "Tom Augspurger" ***@***.***> wrote: IPython 6 switched over to jedi for completion, which must execute at least some code? Categorical.labels is a property. It looks like we can get around this by specifying __dir__. I think I've seen it in at least one other place. @Carreau <https://github.com/carreau> does this sound like an IPython bug? For a simpler example: In [32]: import warnings In [33]: class Foo: ...: @Property ...: def bad(self): ...: warnings.warn("bad") ...: @Property ...: def good(self): ...: return "good" In [34]: In [34]: f = Foo() /Users/taugspurger/Envs/pandas-dev/bin/ipython:4: UserWarning: bad import re In [35]: f.<TAB> If so I'll file one on IPython. Either way I'll put up a PR to silence these using __dir__. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#16409 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAUez31uWSfD5tdFQMNDWW6Q2RTj1E47ks5r8KxogaJpZM4NhuR9> .

[wesm]

FWIW, the Jedi completion in IPython 6 feels like a gigantic step backwards in UX in the IPython shell. I'm seeing 500ms or more tab completion lag on objects when using pandas; having used IPython since version 0.8.4 and never had this problem, a bit of a bummer =( I suspect pandas is a case of a "big" API that stretches some of the Jedi worst case scenarios. Is there a way to revert to the old completer?

[Carreau]

c.Completer.jedi_compute_type_timeout=<Int>
    Experimental: restrict time (in milliseconds) during which Jedi can compute
    types. Set to 0 to stop computing types. Non-zero value lower than 100ms may
    hurt performance by preventing jedi to build its cache.
    Default: 400

and

c.Completer.use_jedi=<Bool>
    Experimental: Use Jedi to generate autocompletions. Default to True if jedi
    is installed
    Default: True

These two options can be of help, the old completer has not been deactivated and completions are merged with jedi ones.
Actually increasing the jedi timeout may make things faster as if you give enough time to jedi it will save a cache of the type-inference tree on disk and re-use that. Which should make most things ~quasi instantaneous (at least for me on matplotlib).

Dev branch of jedi is also way faster (but still have some bug).

[TomAugspurger]

Thanks @Carreau. ipython/ipython#9094 is the IPython issue for properties executing code.

[jorisvandenbossche]

For me it happens on tab completing modules with deprecations as well. Eg pd.lib.<TAB> displays a huge list of warnings (but that might also be related to the way we implemented the shim, but in any case was not visible in IPython < 6.0)

[DGrady]

FWIW I see this (only the first time?) when I try to tab complete on a data frame:

Python 3.6.1 |Anaconda custom (x86_64)| (default, May 11 2017, 13:04:09)
Type 'copyright', 'credits' or 'license' for more information
IPython 6.0.0 -- An enhanced Interactive Python. Type '?' for help.

In [1]: import pandas as pd

In [2]: df = pd.DataFrame()

In [3]: df.gro2017-05-23 10:27:43.062 python[28545:3846149] !!! BUG: The current event queue and the main event queue are not the same. Events will not be handled correctly. This is probably because _TSGetMainT.
upby

I pressed TAB after typing df.gro.

[TomAugspurger]

@DGrady that seems unrelated. Is it just on DataFrames that happens, or any python object? If you could narrow it down a bit, I'm sure the IPython devs would appreciate a bug report.