ENH: switch to the mystmd enginery for rendering

[bsipocz]

No description provided.

[bsipocz]

Well, apparently having the make file would make sense so we run the same in conda and in tox.

OTOH, do we really still need conda, or should nuke it and redo env management in pixi instead?

[rossbar]

OTOH, do we really still need conda, or should nuke it and redo env management in pixi instead?

👍 this is my thought too - especially since the environment resolution takes longer than the notebook execution 🙃 . There exists tooling for this exact problem now so let's transition to it.

I'd propose to remove the conda job in this PR then re-do/beef up CI after updating tooling.

[bsipocz]

We're still seeing issues with the preview, so this should not be merged until that is solved.

[stefanv]

We're still seeing issues with the preview, so this should not be merged until that is solved.

How do you pass the BASE_URL to tox? It may need something like passenv.

[bsipocz]

I'm not sure what we consider blocking for this PR, there are some issues I see. For anything that is a follow-up we may want to open an issue so we don't forget them.

• build time on CircleCI is significantly longer than it was with JB1. I could easily consider this as a follow-up though.
• cleaning up build errors and warnings -- currently we don't run with strict (and we didn't run with a similar manner as we were not failing JB1 builds with these broken links either)
• deployment not to a gh-pages branch -- this could probably be done as a follow-up, too and now just use the file we have. (I need to make my other repos consistent, too so can do this repo along with those).

[stefanv]

• build time on CircleCI is significantly longer than it was with JB1. I could easily consider this as a follow-up though.

Do you have a sense of why this is? Parallel execution of notebooks, setup time, etc.?

[stefanv]

Last build failed with timeout :/

[bsipocz]

build time on CircleCI is significantly longer than it was with JB1. I could easily consider this as a follow-up though.

Do you have a sense of why this is? Parallel execution of notebooks, setup time, etc.?

Unfortunately I don't have any good guesses. My experience with the IRSA notebooks were that the build is faster, but in reality those are the worst benchmarks ever as all of them are about querying data on remote servers and that has a genuine huge scatter in runtime (and gazillion timeouts that are due to the server side).

[bsipocz]

I think this is working now.

I would need decisions on the checkpoints listed in the comment above, do we want one big giant PR with everything or rolling it out with follow-ups. And either case, will need a review and approval (btw, how do you feel about requiring approvals in this repo -- I could argue both ways as I tend to do quick infrastructure fixes that needs to go asap, but even though could benefit from being forced to be looked at by someone else)

[bsipocz]

Now that the conda job has been removed this file is not being tested to work anywhere.

@rossbar - do you have preferences of how to progress?

[bsipocz]

I would nuke the mentions of it, but again I wasn't the conda user. I would think we should provide an env that works ™️ and passes the test, but this one didn't so we may want to jump ships and recommend pixi anyways.

OTOH, I would block the merging of this on the condition that the launch in binder or somewhere else should work, so provide at least one way to users to try the content without mandating them to fix their local installs.