RTD failing yet again

[dcherian]

memory consumption errors as usual.

@keewis I remember you had an idea for using pip instead of conda?

[keewis]

we could try moving everything except cartopy to the pip section in the environment file (iris is scitools-iris on PyPI). That seems to reduce the maximum memory consumption quite a bit (the conda-forge channel has grown almost too much).

[mathause]

This did indeed work for my project (regionmask), https://github.com/mathause/regionmask/blob/master/docs/environment_rtd.yml

[keewis]

I tried it, but it still fails: https://readthedocs.org/projects/xray/builds/10502148/ We can't use pip directly since cartopy does not upload wheels and I'd expect the source install to fail.

If we look at the issue tracker of RTD, we can see a lot of issues regarding memory errors, so maybe this is a general issue with conda (when using conda-forge) and RTD.

Edit: seems I have to conda-install cf-units since otherwise it would need to be built on install.

[DocOtak]

The docs seem to build ok on Azure piplines, is it possible to get the built docs from that and publish somewhere? I do know this is possible with Travis, but haven't actually done it myself since my docs don't (yet?) have a memory problem.

[keewis]

yeah, that would probably be the way forward. See #3697 for more discussion.

[dcherian]

No luck with pinned versions and the stable build image: https://readthedocs.org/projects/xray/builds/10547287/

[keewis]

due to the work of the RTD maintainers, our builds pass again. However, trying to build stable still fails, but instead of a memory error we get a sphinx failure: https://readthedocs.org/projects/xray/builds/10609593/

I think we already fixed that in master?

Also, we might want to switch back to the latest image in readthedocs.yml?