Jedi doesn't show available methods for instances

[tarjeiba]

(This is a repost from the emacs-jedi issues, Issue #212.)

My configuration of jedi won't show available methods for an instance of a class, only for the class itself. The following will give no completion at the point

import matplotlib.pyplot as plt

fig = plt.figure()
fig.

while a corresponding plt.Figure instead of fig. will show all available methods. Does anyone have an idea of what may be wrong?

[davidhalter]

Hmm, what's your Jedi version? Have you tried the latest master branch?

[tarjeiba]

Yes, I have tried using the latest master branch in a fresh python virtual environment with only matplotlib installed.

[tarjeiba]

Could this be related to Issue 372? I get autocomplete for a. in the snippet below, but not for b.

import numpy as np
a = [1,2]
a.
b = np.array([1,2])
b.

[davidhalter]

Can you give me the completion output of just plt.figure?

[tarjeiba]

>>> import jedi
>>> source1 = '''
... import matplotlib.pyplot as plt
... fig = plt.figure()
... fig.'''
>>> source2 = '''
... import matplotlib.pyplot as plt
... plt.figure.'''
>>> script1 = jedi.Script(source1, 3, len('fig.'), 'example1.py')
>>> script2 = jedi.Script(source2, 3, len('plt.figure.'), 'example2.py')
>>> completions1 = script1.completions()
>>> print completions1[0].complete
abs
>>> completions2 = script2.completions()
>>> print completions2[0].complete
func_closure

...so everything seems to be in order. Still, no completion shows when used in Emacs. I find it strange that it finds completions for other, "simpler", types, though. As the list example above. I'm not sure whether this is an emacs or emacs-jedi issue.

[davidhalter]

What you missed is that func_closure is the completion of a function object, not an actual class. People tend to think that numpy/matplotlib uses classes, that you instantiate. This is not true. Most of the time you call a function that returns you an object of something, but you don't call the object itself (because they need to handle the content of the object first).

So what's happening here is simple, plt.figure is a function and therefore gets executed. The execution is too complicated for Jedi (i.e. matplotlib uses a weird caching mechanism that Jedi doesn't understand and therefore cannot evaluate the return value).

This is not a bug - it's simply one of the things Jedi - at the moment - cannot understand, so I'm closing it.

[tarjeiba]

That explains it. Thanks for the thorough follow-up.

And this is the same for numpy arrays?

[davidhalter]

If they are not working - yes! (I know that it's a function, so...)