Installing Additional Packages at runtime

[psychemedia]

Is there any way for a user to installing additional packages from pip into a running xeus-python kernel?

[martinRenou]

It's not available (yet?). I guess it could be done for pure Python packages, but for now you'll need to rely on conda-forge and emscripten-forge packages.

[DerThorsten]

It's not yet working but we working on integrating micropip and a very slim mamba (picomamba) into xeus-python.

[martinRenou]

I guess the original question was concerning pre-installed packages?

[DerThorsten]

Is there any way for a user to installing additional packages from pip into a running xeus-python kernel?

but his question is about installing in a "running" kernel, ie not pre-install?

[psychemedia]

Original comment was more along the lines of: I am in a notebook running a xeus-python kernel; how do I do an equivalent of %pip install newPackage.

[psychemedia]

Elsewhere, I learned that the default pyodide kernel now supports %pip magic.

[jtpio]

Elsewhere, I learned that the default pyodide kernel now supports %pip magic.

Yes this was added in 0.1.0b14: https://jupyterlite.readthedocs.io/en/latest/changelog.html#b14

[jtpio]

For reference micropip should now be available on conda-forge: https://github.com/conda-forge/micropip-feedstock

It would be interesting to get it to install packages on the fly in the xeus python kernel.

[jtpio]

It would be interesting to get it to install packages on the fly in the xeus python kernel.

Gave this a try with the following environment.yml:

name: test
channels:
- https://repo.mamba.pm/emscripten-forge
- https://repo.mamba.pm/conda-forge
dependencies:
- micropip

Importing micropip works:

import micropip
micropip.__version__
# '0.3.0'

Installing a package with:

micropip.install('emoji')

Then importing it with:

import emoji

Gives the following:

[psychemedia]

Any updates on the ability to install packages at runtime?

(To a certain extent, NOT supporting this might be useful in an educational context, if you want students to be limited to using just and only what is provided...)

Having micropip available might be handy when working out what's required when putting toigether a new environment, though...

[jtpio]

(To a certain extent, NOT supporting this might be useful in an educational context, if you want students to be limited to using just and only what is provided...)

Probably this would be the case depending on whether the micropip (or equivalent) package is available in the kernel environment, which would be controlled by the site deployer.

[martinRenou]

The coming update to the new emscripten-wasm32 platform may allow using micropip? @DerThorsten

[psychemedia]

Thinking about the environment.yaml config file, the "normal" config file spec allows you to install pip as a dependency and then provide a sublist of pip packages to install from PyPI. Is there any sense in which you can installpip or micropip packages as part of the xeus-python-kernel's environment.yaml file?

[martinRenou]

As of recently xeus-python allows partial support for installing pip dependencies: https://xeus-python-kernel.readthedocs.io/en/latest/configuration.html#pip-packages

Though you should not specify pip in the dependency list. We should probably take care of this case though. I'll open an issue for that.

[pierre-haessig]

Hello, I'm wondering if the support for installing pure Python package with pip (https://xeus-python-kernel.readthedocs.io/en/latest/configuration.html#pip-packages) is still considered experimental.

In the meantime, I see that jupyterlite-xeus-python is being deprecated over jupyterlite-xeus (https://xeus-python-kernel.readthedocs.io/en/latest/migration.html#migration), but I see the corresponding doc at https://jupyterlite-xeus.readthedocs.io/en/latest/environment.html#pip-packages with about the same content.

[martinRenou]

It is indeed still experimental.

The biggest downside of it is that it installs your package with the --no-deps option passed to pip so you'll have to install the dependencies yourself, either from Pypi or conda-forge or emscripten-forge.

Maybe one day we'll have a better approach for this.