I'm slowly working on (2), there's a few mathtext-related PRs.
(1) is known sphinx-doc/sphinx#5460 and milestoned for sphinx 1.8.1, I agree with blacklisting 1.8.0.

This means I need to re-generate and re-publish the docs tonight?

@tacaswell No pressure. As soon as you have time. I think the search function proves useful especially when a new version is out, so waiting for 2 more weeks might be a bit too long.

Btw. is there any data available on how often which site is used? I mean, maybe there are only 9 people other then me actually using the search? 😆

In the long run, how realistic is it to generate the docs automiatically from the respective vx.y.z-doc branch every couple of days, similar to the devdocs?

Yes, we have google analytics on the site, looks like ~50k hits to search a month...

manually backported to v3.0.0-doc as ea13253

I just pushed the updated docs, should be live in the next 10-15 minutes.

@tacaswell It doesn't look like search is working in the updated docs at matplotlib.org even though it says to have been built with sphinx 1.7.9.

• Search at matplotlib.org (not working)
  

• Search of this PR's circle-ci build (working)
  

I noticed that there are also other differences between the docs on the website and the ones created via circle-ci, e.g. the version number seems to be different, but also the "note" box in the examples that was removed in #11428 is still there on matplotlib.org (cf. ci-build vs. web). When I build the docs locally they are in line with the ci-builds. Any idea on a possible difference?

Now, sphinx released version 1.8.1 yesterday without fixing the search. So effectively the sphinx!=1.8.0 introduced here is already useless. New ci-builds will use sphinx 1.8.1, again introducing the non-working search and red symbols for mathtext.

Concerning search, the fix is not done in sphinx itself as it stands, but one has to modify layout.html to include some additional javascript. I will try to make a PR with that shortly.

Have you tried refeshing your browser? I had the problem that even with 1.7.9 it used a cached erroneous javascript or something.

😞 on sphinx 1.8.1. So we should sphinx<1.8.0 for the time being.

@timhoffm Try now? I just cleared the cloudflare cache as well.

@tacaswell Thanks. It works now, though I'm not sure if I had the problem on this machine.

Wow... so there are remains from the past in my browser, and they are hidden deep.
It's rather disturbing to find out that my browser executes some javascript from a site I visited yesterday...
Completely clearing the browser cache did the trick though.

Here is the fix for the search: #12216
I don't know how far the activities on fixing the red mathtext symbols have moved and whether it would hence make sense to pin sphinx to <=1.7.9 or not.

I think I know how to fix the red mathtext symbols (you "just" need to patch the suitable docutils dicts and/or send a patch to docutils) but it's not going to be a priority for me, pinning to 1.7.9 in the meantime is fine.

Math issue is now tracked in #12532.