Flips sign for gravity_unit_vector to match its description (attempt #2)

[tomchor]

This is the second attempt at flipping the sign of gravity_unit_vector. The first attempt (#2963) had an issue where the doctests were getting stuck with no apparent cause.

For this attempt, I proceeded more carefully, step by step to try and pinpoint what was happening. The good news is that the doctests are now passing locally. The bad news is that I still don't know why they were getting stuck in #2963.

Since tests pass locally for me (I tested everything except the examples) I'm hopefully that this time they'll also pass on buildkite.

@navidcy your modifications to #2963 are also included here!

[navidcy]

apparently something's wrong -- docs are running for 6hrs 8.5hrs and counting...... :(

[tomchor]

apparently something's wrong -- docs are running for 6hrs 8.5hrs and counting...... :(

Then I truly don't understand. The docs are building locally for me. The only thing I'm not testing locally are the examples. Could the culprit be there somehow?

@navidcy do you mind if I use buildkite for some experiments? e.g. trying to compile the docs without examples, etc?

[navidcy]

go for it

[navidcy]

It actually seems that docs are doing something... they are not stuck... but somehow much much much slower, which for me points to something that causes long compilations times...

[tomchor]

It actually seems that docs are doing something... they are not stuck... but somehow much much much slower, which for me points to something that causes long compilations times...

It's possible... but so far in this PR the docs are passing in a reasonable time locally for me so I have no way to investigate this locally.

[tomchor]

I just pushed the branch exactly the way I just tested them here (i.e. with a bunch of things in docs/make.jl commented out). The docs were successfully built for me in about 30 min, so it'd be a red flag if it takes much longer than that to build it on buildkite.

[tomchor]

Docs built successfully with f2e22b8! Nice. Next commit (8572c24) reintruduces the tilted bbl example and was also successfully tested locally.

[navidcy]

#2996 might be relevant..

[tomchor]

@navidcy can you wait until the docs are finished (or running for more than 8 hours) before pushing things to this PR? We can't see if docs pass in commit c24a45d because they never finished running. Since the docs normally are now apparently taking 7h, iterations on this are super slow.

[navidcy]

Sure. Sorry. :(

[tomchor]

Docs are building! 🎉

[navidcy]

I'll fix the unit tests

[glwagner]

boom

[tomchor]

What's weird is that even though the docs pass after I included things carefully one-by-one, I still don't know why they weren't passing before. As far as I can tell I did nothing different in the end. Weird...