Write Docs in Jupyter Notebooks with Sphinx
MaxwellBloch now has docs, including a usage guide and a set of examples. These are mostly written in Jupyter notebooks for two reasons. First, notebooks encourage a ‘show don’t tell’ approach. Second, if something in the library interface changes and I forget to update the docs, the build will fail and the changes won’t be merged before it’s fixed.
extensions = [ 'nbsphinx', 'sphinx.ext.mathjax' ]
The MathJax item renders LaTeX blocks. I’ve set some configuration options to ensure that long-running cells complete, that any cell errors cause the docs build to fail, and that notebooks are always executed even if they already have partial or full output1.
nbsphinx_timeout = 3600 nbsphinx_allow_errors = False nbsphinx_execute = 'always'
The docs are hosted at Read the Docs. I tried setting this up last year but found Read the Docs could not at that time support the high memory requirement in setting up the Conda environment. They’ve since looked specifically at support for scientific project documentation and increased the memory available. So it works now, a great service. The time constraint on the build means I still have to cache the output of longer-running example problems rather than have the solver run at build time.
Write Docs Early
In developing MaxwellBloch I prioritised getting the features I wanted built before looking to write usage docs and just focused on having some examples available. From the support emails I get I realise now this was a bad idea. Many emailers who asked how something works would have been helped just by having a clear guide, and there are no doubt more people who have just given up and not used the package even if it might have been of value to them. In future projects I’ll write usage docs as soon as there’s something there to use.
Though I have
gitclear any output before staging