Table of Contents

Currently I am writing all my Python code using Neovim and a couple of plugins to provide auto-completion, linting, etc.

For code auto-completion, I use deoplete as my completion engine and deoplete-jedi as the completion source for Python. They work great most of the time. However, occasionally I can not get any auto-completion for some object instances. For example, auto-completion for object instances returned by the Image.open() method in the package Pillow does not work. I decide to take a look.

At first, I think it is an issue with deoplete or deoplete-jedi. However, I find that auto-completion for Image instances returned by other methods, e.g., Image.new(), works as expected. If this is an issue with deoplete or deoplete-jedi, then auto-completion should not work anymore. Maybe it is due to the static type checker Jedi that deoplete-jedi relies on?

To find out the real reason, I use the following script to check if Jedi can provide completion for Image instances returned by Image.open() and Image.new() method¹:

import jedi

source1 = '''
from PIL import Image
im = Image.new('test.jpg', (128, 128))
im.
'''

script1 = jedi.Script(source1, 4, len('im.'), 'example1.py')
print(script1.completions())

source2 = '''
from PIL import Image
im = Image.open('test.jpg')
im.
'''
script2 = jedi.Script(source2, 4, len('im.'), 'example2.py')
print(script2.completions())

As expected, the first print() shows a list of completion results, while the second print() shows an empty list.

So this is actually an issue with Jedi. I filed an issue on the Jedi GitHub repo. After a while, I got a reply from the maintainer of this project. The root cause is that Jedi can not infer the type returned by Image.open() method since it is too dynamic to infer. The author suggested that I may use type hints to help Jedi infer the types of the return objects. I tried two ways based on the type hinting documentation of Jedi.

The first way is to use function annotations. I find the source file of Image.open() on my system and change the its definition from

def open(fp, mode="r") :

def open(fp, mode="r") -> Image:

Now the auto-completion works for instance returned by Image.open() method. While this method works, it is cumbersome to use since it requires changing the source files of various packages.

The second way is to use type hints in inline comments. But the auto-completion does not work anymore. My test script is:

import jedi

source = '''
from PIL import Image
im = Image.open('test.png')  # type: Image
im.
'''

script = jedi.Script(source, 4, len('im.'), 'example.py')
print(script.completions())

The print function shows an empty list.

So for now, you may stick to function annotations if you really want type hints to work for Jedi.