Extremely slow performance and high CPU usage for fastai library

I am using Neovim 0.5.0-dev+nightly with jedi-languange-server which works well and quickly for most packages like numpy for example. However when used with the fastai library the results take sometimes 10-30 minutes to show up when prompted.

To reproduce my issue you can install fastai via miniconda3: conda install -c fastai -c pytorch fastai and then install Jedi-language-server: conda install jedi-language-server

Then open a file in neovim with the conda-env enabled and jedi enabled, here is a short example file:

from fastai.vision.all import *

path = untar_data(URLs.PETS)/'images'

def is_cat(x): return x[0].isupper()
dls = ImageDataLoaders.from_name_func(
        path, get_image_files(path), valid_pct=0.2, seed=42,
        label_func=is_cat, item_tfms=Resize(224))

learn = cnn_learner(dls, resnet34, metrics = error_rate)

then try to bring up hover help on one of the fastai functions like ImageDataLoaders. For me this takes several minutes and makes one of my CPUs stay at 100% for that entire time.

I’m wondering if this has something to do with the way that fastai is designed to be imported with from fastai import * which is not typical practice for python libraries.

If there is anymore information I can give you please let me know. Thanks

[atticusmatticus]

Just in case someone else might find this useful, the lua code that got what @HansPinckaers was describing working for me with neovim 0.5 built-in LSP was this:

    require'lspconfig'.jedi_language_server.setup{
        init_options = {
            jediSettings = {
                autoImportModules = { "fastai", "fastcore" }
            }
        },
        on_attach = on_attach
    }

The on_attach settings are other general settings.

[pappasam]

Hmm, I can reproduce. This looks like the Jedi issue being tracked here: https://github.com/davidhalter/jedi/issues/1721