Redesign of build process

[fredrikaverpil]

I'm considering changing the build process on all platforms:

• Use pre-built docker images on Linux, stored on dockerhub to speed up the pre-build phase
• Utilize JOM for parallel PySide2 build commands, speeding up the build process
• Conform to using/requiring conda on all platforms for building and then also installing/using PySide2
• Possibly publish the wheels on conda-forge allowing for very simple quickstarts, given the user already is using conda.

What would you guys say about using conda instead of the current setups?

[leycec]

Everything sounds fabulous, except that whole "Conform to using/requiring conda on all platforms for building and then also installing/using PySide2" part.

The [Ana|Mini]conda distribution is fairly heavyweight and specialized for use in scientific computing. While leveraging conda-forge to publish [Ana|Mini]conda packages would dramatically simplify PySide2 installation for [Ana|Mini]conda users (and hence would be a very good thing), doing so should ideally be done in addition to rather than instead of publishing PyPI-distributed wheels.

The two communities serve different needs and are largely orthogonal. While our downstream multiphysics simulator personally recommends use of [Ana|Mini]conda to simplify installation, we have a hardcore contingent of CLI-oriented users who prefer a manual installation leveraging some combination of pip3 and their system-wide package manager (e.g., apt). They are painfully loud and refuse to use [Ana|Mini]conda. In short, they are people like me.

It'd be great if we could continue to service both communities.

[fredrikaverpil]

Yes, I agree. I think adding conda-built wheels (so to speak), could be a nice addition rather than a replacement.

[fredrikaverpil]

@leycec PySide2 now available on Anaconda Cloud: https://github.com/conda-forge/pyside2-feedstock ❤️