Albert Steppi

> This is one example PR showing how to adjust the CI timeout limit: #8564 Great! Thanks @leofang!

> @steppi some of the CI failures are due to timeout, but at least one has scipy.special related failures: https://ci.preferred.jp/cupy.linux.cuda118/193866/ Could you go through the CI logs? Sure. I'll investigate.

It turns out the problem was missing execution space annotations, which are added in `xsf` with the macro `XSF_HOST_DEVICE`. There were a few functions that were missing them. When using...

> Looks like the fix works, we just run into timeouts. Should I bump the timeout times a further?

> I think it’s a very good initiative! Not only does it gives proper credit, it also helps the (curious) user to locate the source of the implementation easily without...

~Looking into it a little. I think this is actually a Sphinx bug which has been fixed at least in main. It would be weird if Sphinx wouldn't accept it's...

Thanks @agriyakhetarpal. I’m off this week and have limited availability to look at this. I think it’s just a deprecation warning that gets turned to an error when building the...

@agriyakhetarpal, that looks like the exact same deprecation warning. I think you did reproduce it. Have you confirmed the warning goes away with the fix you made in jupyterlite-sphinx?

Walking back my quote from above. In a way, the docs already function as both visual test and integration test. The reason 0.17.0 was shipped broken though was because we...

> @steppi This issue proposes to define the PMF of the negative binomial distribtion with n = 0 successes as a point mass with probability 1 at k = 0...