I had to write a couple of hacks for generating the notebooks, in particular for the first notebook cell and also to replace the "plot" commands by "iplot".

The different CIs are failing because of installation problems with the orca executable used by plotly to generate static images, I'll work on this issue later.

Seeing half a dozen commits already for tweaking CIs to get them to run here (something that plotly must have solved on their own internally, and already maintains, etc.) makes me think more strongly that #441 (comment), or some other variant of "keep code for other scrapers in the packages that need scraping", is a better approach. Before working too hard on the CIs here, can you see if you agree?

@larsoner now that I've started the PR I'd like to debug the CI :-). But of course I'm fine with the approach you recommend and which is described in more details in #441. It's just that the doc of sphinx-gallery says "If you've developed a custom scraper for Sphinx-gallery that would be useful to the broader community, we encourage you to contribute it to the list of natively-supported scrapers located in the scrapers module. We welcome PRs!". Since plotly is a quite popular plotting library and it's been discussed in a couple of issues, I thought it qualified as "useful for the broader community" ;-). But of course you'll need to limit the number of scrapers for maintainability. Maybe this bit in docs/advanced.rst should be updated?

If you've developed a custom scraper for Sphinx-gallery that would be useful to the broader community, we encourage you to contribute it to the list of natively-supported scrapers located in the scrapers module. We welcome PRs

Yeah I think we should update this. On further inspection I think it wasn't as good a plan as I/we originally thought.