DEP: replace interp2d by stub

[h-vetinari]

Towards #18583, details & references in that issue.

[h-vetinari]

The last commit here can be omitted if we merge #19904 first.

[h-vetinari]

So now we have the issue that the transition guide wants to execute interp2d, which is obviously not possible anymore. Should we simply delete the guide and in the error/docstring refer to the 1.13 transition guide?

[ev-br]

This is an interesting fallout :-). Basically, this page should be static. And well discoverable, esp after interp2d starts raising hard errors.
Whether it's back to ReST or link to 1.13 docs --- whichever works.

[h-vetinari]

Is there an easy way to translate the current website in main to a static RST/MD file, where we don't re-execute the removed import? CC @tupui

[j-bowhay]

Could we move the offending code cells to be markdown cells with the content surrounded with backticks and then just put the images in?

[tupui]

Is there an easy way to translate the current website in main to a static RST/MD file, where we don't re-execute the removed import? CC @tupui

I am not sure what you mean. If it's about the API doc itself, I am not sure you can generate it and save it for later builds (besides the normal caching from Sphinx, but I never managed to get this caching working in our CI, I had 2-3 PRs trying various things and gave up. Yes this is 💯 ridiculous that we have to do 20 mins runs for the doc on all PRs 😅😭).

[h-vetinari]

Could we move the offending code cells to be markdown cells with the content surrounded with backticks and then just put the images in?

Wouldn't we have to turn it from a notebook into a regular markdown page? I guess we could save the images from a local run with 1.13 (or main).

I am not sure what you mean.

It might go through many layers, but the end result is a static HTML page. If we can keep showing that webpage, the goal would be achieved; preferably without stripping off too many layers of abstraction.

However I'm tending towards just referring to the 1.13 docs. Those will continue to be a notebook that works fine, and it avoids a large effort for (IMO) questionable gain.