Add MNELAB (requires Qt 6)

[hoechenberger]

As soon as we have Qt6 on conda-forge

To keep track: conda-forge/qt-main-feedstock#14

[hoechenberger]

An update on this one:

Updating Qt/PyQt to 5.15 was a huge deal of work, so on the Spyder team we won't be looking at Qt6 for at least a year, perhaps more. And I don't know who else would be willing to do this as a paid effort, sorry.

conda-forge/qt-main-feedstock#14 (comment)

cc @cbrnr

[cbrnr]

Yes, I saw that. 🤷‍♂️? It can be installed with pip, even in a conda env. I don't know why it is necessary to create conda packages when this is so much effort and when official binaries exist on PyPI.

[hoechenberger]

Because the installer ships much more than just Python packages

[cbrnr]

Yeah, it's a pity the installer can't include pip packages.

[hoechenberger]

We can ship custom files, so we could include packages and install them via a post-install script, but it would require some fiddling.

[cbrnr]

I think it would be worth a try. Maybe we can pip install in the post-install script?

[hoechenberger]

I think it would be worth a try. Maybe we can pip install in the post-install script?

Yes. It's trivial to just run pip install in the post-install script, or to include a desktop shortcut / .app that opens a shell and does that thing automatically.

The problem with this is that it would require network / internet access.

To avoid this, we'd need to bundle all PyPI dependencies, drop them into a separate folder, and run something like

pip install --index-url /path/to/pypi-packages ...

To ease things, we could simply bundle all "pip requirements" in a .zip file that we include in the installer.

I'm still reluctant to do so as I believe the last time I tried to get PyQt6 working with my MNE conda environment, things got all sorts of "twisted-up" and I ran into some issues. I don't remember details though.

In any case, I won't have time to look into this for at least a couple of weeks.

[hoechenberger]

Another approach:
Re-bundle the "pip requirements" as conda packages. We won't be able to publish those on conda-forge, but we could upload them to a custom mne channel on anaconda.org and list this package source in our construct.yaml. Then we won't have to fiddle with any post-install magic.

https://docs.conda.io/projects/conda-build/en/latest/user-guide/wheel-files.html#building-a-conda-package-from-a-wheel-file

[cbrnr]

I don't think that pip installing those packages into a conda env will mess up anything – these are packages that only exist on PyPI. Last time I tried it worked, but I didn't test with the full environment used by the installer. But we just have to install from the installer and then pip install mnelab to know if there are any conflicts. Do we have any tests that we can run?

Re internet access, we just have to bundle the source directory and then install via pip install <mnelab-source-directory>.

[hoechenberger]

I think converting the wheels to conda packages and then using these with the installer will save us a lot of headaches. We just need someone to volunteer and create an exhaustive list of all dependencies of MNELAB that we currently cannot ship through conda-forge. I can then create conda packages for all of them and publish them to a custom channel.

[cbrnr]

I can do that. The dependencies that are not available on conda-forge are PySide6, which itself requires PySide6-Addons, PySide6-Essentials, and shiboken6. That's it.

[hoechenberger]

Qt6 and PySide6 are now part of conda-forge (PySide6 is being built as we speak)

Once it's landed for all platforms, we can try to revive the MNELAB feedstock.

See https://anaconda.org/conda-forge/pyside6

[cbrnr]

Meanwhile, there is another problem that needs to be fixed first. Matplotlib doesn't work with the PySide6 backend, which MNELAB requires. They have already fixed it (matplotlib/matplotlib#24158), but we are waiting for a new release. Therefore, it's maybe best to wait for Matplotlib v3.6.2.

cbrnr/mnelab#362