Closed gully closed 2 years ago
Another issue is that some tutorials rely on gollum, and gollum
as currently implemented relies on local precomputed synthetic spectral models. So the muler tutorials that use gollum
fail.
So we may want to return to auto
rather than always
option and re-upload some notebooks. Hmm... I wonder how other projects deal with these friction points.
I've learned how to turn off rendering notebooks on an individual basis. So I am inclined to leave on the notebook rendering (or maybe turn it to auto
).
We have the issue that some of the notebooks I turned off have their output stripped with nbstripout
, but that's solvable.
I'll change the title to reflect the new bottleneck.
Ok, after lots of back-and-forths, I got readthedocs (RTD) to render the tutorials! Yay! 🥳
The one hiccup is that RTD does not play nicely with
importlib_resources
, I think it has something to do with how RTD makes paths to the source code it installs, relative vs. absolute paths, and other finicky things.My workaround is to simply (and heretically)
pip install
muler to render the documentation. That's heretical because it means any code change will not get instantaneously reflected into the live documentation, which is a big loss. Big enough that we may want to reconsider this choice. Feedback is welcomed.On the other hand it means that we have a renewed incentive to make regular pip releases in order to keep the docs up-to-date, which will make
pip
a more up-to-date option for those who don't want to do the full developer version.Originally posted by @gully in https://github.com/OttoStruve/muler/issues/93#issuecomment-1110230316