Vscode-jupyter: "Run Cell" does not pickup on jupyter installed via virtualenv but conda works

[sotte]

Environment data

• VS Code version: 1.29.0 5f24c93878bd4bc645a4a17c620e2487b11005f9 x64

• Extension version (available under the Extensions sidebar): 2018.10.1

• OS and version: Ubuntu 18.04
• Python version (& distribution if applicable, e.g. Anaconda): pipenv with python 3.7
• Type of virtual environment used (N/A | venv | virtualenv | conda | ...): virtualenv
• Relevant/affected Python packages and their versions: -

Actual behavior

When using a virtualenv/pipenv environment clicking on "Run Cell" yields the error msg:
Running cells requires Jupyter notebooks to be installed.

(But it works with conda.)

Expected behavior

Cell is executed.

Steps to reproduce:

mkdir hello
cd hello
pipenv --python 3.7
pipenv shell
pipenv install jupyter
which jupyter
# /home/somebody/.local/share/virtualenvs/hello-UXM3mEkX/bin/jupyter
# jupyter notebook also works on its own

Open vscode, select the interpreter for the "hello" pipenv environment, create a file demo.py with the following content:

#%%
print("hello")

Clicking on "Run Cell" yields the error msg:
Running cells requires Jupyter notebooks to be installed.

Relevant excerpt from Pipfile.lock

        "jupyter-client": {
            "version": "==5.2.3"
        },
        "jupyter-console": {
            "version": "==6.0.0"
        },
        "jupyter-core": {
            "version": "==4.4.0"
        },

Logs

Output for Python in the Output panel (View→Output, change the drop-down the upper-right of the Output panel to Python)

No output.

Output from Console under the Developer Tools panel (toggle Developer Tools on under Help)

/usr/share/code/resources/app/out/vs/workbench/workbench.main.js:294 [Extension Host] Python Extension: Error: Error executing Jupyter command 'nbconvert': [Errno 2] No such file or directory

t.log @ /usr/share/code/resources/app/out/vs/workbench/workbench.main.js:294
/usr/share/code/resources/app/out/vs/workbench/workbench.main.js:294 [Extension Host] Python Extension: Error: Running cells requires Jupyter notebooks to be installed.
t.log @ /usr/share/code/resources/app/out/vs/workbench/workbench.main.js:294
/usr/share/code/resources/app/out/vs/workbench/workbench.main.js:2429 Running cells requires Jupyter notebooks to be installed.

nbconvert is not available via the venv, but jupyter-nbconvert is.

Maybe Related

See https://github.com/Microsoft/vscode-python/issues/3341 and https://github.com/Microsoft/vscode-python/issues/3332

[sotte]

Update: It seem to be related to how you start vscode.

• When vscode is started from the activated virtualenv code . then "run cell" works.
• When vscode is started without any environment (and jupyter is not available in the current PATH)
  
  
  
  • "run cell" works with a conda interpreter
  
  
  • "run cell" does not work with the interpreter from the virtualenv
  
  

conda and virtualenv seem to be handled differently. Is this a bug or accepted behavior?

[mrvkino]

I have the same setup as @sotte except I am on MacOS Mojave 10.14.1. But, event if I start vscode within the environment started by Pipenv, vscode is not able to find Jupyter.

The Jupyter Python Package from @DonJayamanne is able to find the Jupyter executable.

[rchiodo]

I can't repro @sotte's bug on Linux, but I can on Windows. Hopefully the fix I have out will address the problem.

[sotte]

@rchiodo thanks for the quick reply!

I just checked 1.30.0-insider 24b0f75b1e33b88f30f2ed9d5b17ca60062c270b x64 and I have the same experience. It was a fresh insider install with no extensions other than the python one.

[rchiodo]

@sotte Ah, your venv doesn't support nbconvert so you can't do an import. There's a bug in the code that it checks for nbconvert instead of notebook being available.

This line here:
https://github.com/Microsoft/vscode-python/blob/40554c8effdba308744ff68de24e5fd83ee84f12/src/client/datascience/history.ts#L492

That line should say:
if (!(await this.jupyterExecution.isNotebookSupported())) {

Sorry took me so long to notice you didn't have nbconvert installed in your venv.

I'll submit a fix for this right away.

[rchiodo]

Should be fixed now in https://github.com/Microsoft/vscode-python/commit/d2a5e21ecd8e8ad5ecaf27f864f54f27b6a604af

[sotte]

@rchiodo thanks, you're great!

[leonjonathan]

@sotte Were you able to take advantage of the fix? I have very similar issue (macOS 10.14.1) and I've been navigating the related issues that've been merged, but can't make out if the fix is available

[sotte]

Yes, it worked for me. The fix is only available through the nightly/insider build though @leonjonathan

[leonjonathan]

Thanks @sotte, no luck for me though. Looks like the commit with the fix has already been pushed to Extension Version 2018.11.0, which is what I am using. After looking at microsoft/vscode-jupyter#3774 it's worth mentioning that I'm on python 3.6.7 (pyenv install) and my virtual environment is created with venv (via pyenv-virtualenv), so maybe worth opening my own issue? @rchiodo

[rchiodo]

@leonjonathan if you're using a pip venv then microsoft/vscode-jupyter#3774 is likely the culprit. If you start code from the terminal that has the environment activated, does it work then?

[leonjonathan]

@rchiodo Yes, works as expected when starting code with environment activated in terminal. Not using pipenv though (as in microsoft/vscode-jupyter#3774). If I understand correctly, pipenv uses VirtualEnv, while pyenv-virtualenv uses native python3 -m venv despite its name