JupyterLab installed for Python 3.9 instead of 3.10 in Universal image

[dmgardiner25]

When trying to go through the codespaces-getting-started-ml sample repo in JupyterLab, we are failing to import some packages. This is due to us installing jupyterlab with python 3.9 instead of 3.10. The same does not happen in VS Code.

Repro:

1. Create a codespace using the repo mentioned above
2. Connect to JupyterLab using gh cs jupyter
3. Attempt to run the first cell and notice that it fails to import torch

Notes:

• You can verify that this is the issue by running pip list to see that jupyterlab is missing from the list of python 3.10 packages and /usr/local/python/3.9.14/bin/pip list to see that it's installed under python 3.9`

cc. JP Ungaretti (@jungaretti) Samruddhi Khandale (@samruddhikhandale)

[jungaretti]

pip, pip3, python, and python3 are all pointing to the correct installation of Python:

which pip && which pip3 && which python && which python3 && python --version

/home/codespace/.python/current/bin/pip
/home/codespace/.python/current/bin/pip3
/home/codespace/.python/current/bin/python
/home/codespace/.python/current/bin/python3
Python 3.10.7

Notice that the output of pip list does not include jupyterlab:

pip list | grep jupyter

(No output)

However, /usr/local/python/3.9.14/bin/pip list | grep jupyter does include jupyterlab:

/usr/local/python/3.9.14/bin/pip list | grep jupyter

jupyter_client       7.3.5
jupyter-core         4.11.1
jupyter-server       1.18.1
jupyterlab           3.4.7
jupyterlab-pygments  0.2.2
jupyterlab_server    2.15.1

I suspect this is a bug in the Python feature that installs JupyterLab to the "additional version" of Python that we're installing here:

images/src/universal/.devcontainer/devcontainer.json

Line 27 in 4690ca1

"additionalVersions": "3.9",

[jungaretti]

We use the Python feature's install_user_package helper to install JupyterLab: https://github.com/devcontainers/features/blob/ceb4544cc38d4882a0294c79f4ef8123435cb027/src/python/install.sh#L316-L319

install_user_package installs packages to whatever Python installation is specified by $INSTALL_PATH. Both install_from_source and install_using_oryx overwrite this variable. When we use install_python to install the extra version of Python (3.9), we overwrite $INSTALL_PATH to that installation.

When we install JupyterLab at the end of the script, it uses 3.9's $INSTALL_PATH instead of 3.10.

[jungaretti]

This is fixed in version 2.0.16 of the Universal image.