mybinder.org sagemath does not work any more

[dimpase]

https://mybinder.org/v2/gh/sagemath/more-sagemath-tutorials/master?filepath=_build/html/tutorial-start-here.ipynb

errors out.

[mmosmond]

It seems like this is still/again an issue? My SageMath notebooks via mybinder are also failing over at https://github.com/mmosmond/eeb430. Any ideas what is going on? Everything was working fine in early September but stopped working later in the month. Some updates have broken things? See also calculuslab/Calculus_Lab#1

[dimpase]

IMHO it's not us, it's mybinder that's guilty here. Maybe @betatim knows...

[betatim]

Have a look at jupyterhub/binderhub#1409 (comment) and the forum threads. It seems likely that this is because of the recent changes to use jupyterlab by default on mybinder.org.

[behackl]

While there is a conda-based approach for SageMath + mybinder by now, I still wonder what exactly it is that causes mybinder to fail starting the environment when using a simple Dockerfile like

FROM sagemath/sagemath:9.7
COPY --chown=sage:sage . .

I played around with the setup a bit and found something somewhat odd: if we don't use the jupyter / notebook installed to SageMath's venv, but instead install Python 3 and notebook separately, it is possible to use the Docker image: with the Dockerfile provided at https://gist.github.com/behackl/a5d25c901ec557565f99331b130f09a2 (apologies, it's not particularly pretty) the mybinder environment does start up normally and the kernel works as expected (as far as I could tell):

Why? Is there some notebook setting that is overwritten or changed in our venv that clashes with how mybinder creates the environment? I'll also share this with sage-devel, perhaps someone has an idea about this. (Edit: done.)

[dimpase]

IMHO Sage should not vendor Jupyter in the 1st place.

Note that even not vendoring gcc is still open, cf https://trac.sagemath.org/ticket/32532 - where I propose the remove the capacity to build gcc, retaining the capacity to build gfortran.

[behackl]

IMHO Sage should not vendor Jupyter in the 1st place.

It is nice that sage -n jupyter just works, but I am not sure whether it is necessary to actually vendor notebook for that. I suppose it might be possible for sage to do something intelligent regarding looking for Jupyter installed on the system side and linking its kernel to it, but if it isn't it could still be installed in Sage's venv.

My best guess at this point is that either Sage ships with some specific setting that makes the notebook server unresponsive when started by mybinder (none of the settings feel like they could be responsible though; but that's something that could be tested) -- or mybinder is not happy with its bash commands being executed via sage -sh -c.

Note that even not vendoring gcc is still open, cf https://trac.sagemath.org/ticket/32532 - where I propose the remove the capacity to build gcc, retaining the capacity to build gfortran.

I'm not familiar with the build toolchain well enough to make a meaningful comment about it, but I don't find it unlikely that this discussion might be easier with packages that have a little less, well, gravitas.

[dimpase]

fixed by #16