On Wed, Nov 10, 2021 at 07:32:53PM +0100, Timo Röhling wrote: > Hi Andreas, > > * Andreas Tille <andr...@an3as.eu> [2021-11-10 18:14]: > > nbsphinx.NotebookError: TypeError in user_guide/style.ipynb: > > 'coroutine' object is not subscriptable > This seems to be an issue with nbconvert 5.6.1, which apparently is > not playing nice with updated Jupyter notebook packages [1]. >
The proximate cause was updating jupyter-client to 7, which changes some existing APIs to be asynchronous. This requires patches or new versions to some other packages (nbconvert included), and probably some other packages which import jupyter bits and pieces directly. I'm working on it. The jupyter ecosystem is unfortunately suffering from a bit of runaway complexity, which is making keeping up with all the bits (particularly the web-connected bits, like notebook and ipywidgets) difficult. > > I wonder what you think here whether it makes sense to stop > > providing the docs for this package. > Even though I usually look up documentation online, I'm a bit > hesitant to drop the documentation completely, because I do need to > work "in the field" without Internet connection from time to time. > > Maybe, as a compromise, we can cut out all the notebooks^H bells and > whistles and limit the offline documentation to the API reference > itself, which is arguably the most useful part? I don't know > how easy it is to trim down the documentation, though. > This I would consider. When working on python packages, my feeling is that the documentation is a frequent cause of problems (sphinx extensions, privacy breaches in templates, unpackaged themes, anything that is meant to embed code), and honestly I don't think compiling it to HTML adds very much value to users. If the documentation source is in something which can be directly human-consumed like markdown or restructuredtext, I'm inclined to think having that in the source package is sufficient for the majority of use-cases. Gordon