Are there more pages that need the directive?

LGTM.

For the curious, this PR does not change the number of warnings in our pedantic mode doc build, both before and after there are 183 warnings about "reference not found"

Just a note, since I was credited with this: I'm not certain this is the correct fix.

What I am certain of is that certain ufuncs in the NumPy documentation currently do not correctly generate cross-reference hyperlinks. For example, if cross-links were generated correctly, I believe every entry in the third column of this table in NumPy's docs would have a hyperlink, but for some reason only divmod does.

We see the same thing in JAX when we use intersphinx to attempt to generate cross-links to these function, but note that this is an issue in NumPy even without any reference to the JAX project.

I haven't checked whether the suggested fix here fixes the issue, but I came up with it as a possible culprit after a few minutes of browsing the doc source and trying to figure out what may have been different between divmod and the other functions in that table.

@rossbar if you think this is probably right, lets give it a shot?

According to the objects.inv, the ufunc link targets are listed under the :py:data: role, not :func:. I went ahead and tried switching the roles from :func: to :py:data: in NEP 13 (see @jakevdp 's comment above) and sure enough it fixes the links (note that the divmod link that "works" actually points to the Python builtin divmod, not numpy's).

I do think it would be an improvement if the generated ufunc stubs were recognized by the :func: role, but I'm not sure exactly why/how this works without taking a deeper look. I'm not sure that this change will fix the problem ID'ed in jax-ml/jax#11578. FWIW I tried the following procedure to see if this would fix the problem in NEP 13:

1. Added a row to the table in NEP 13 for the cos function
2. Update the intersphinx mapping in the NEPS conf.py to point to the objects.inv from the circleCI artifacts in this PR
3. Built the NEPS

The link to cos was still not working despite the updated objects.inv. @ilemhadri can you confirm that this patch fixes the problem for you? If so I'd be fine with adding it (even though I don't fully understand how it fixes the link issue).

Alternatively, if it's easy for you to update the role which is used to link the numpy ufuncs in the jax docs to :py:data:, that might be another solution.

@ilemhadri can you confirm that this patch fixes the problem for you?

I don't have a way of testing this locally, but @jakevdp might be able to check!

@rossbar is there a point in doing this whether it helps or not? In that case, maybe we should just merge, otherwise maybe simply close (can always reopen/open a new PR)?

@seberg I would vote in favor of just merging!

It seems the fun trick to make it py:function role would be do the same as in gh-23039. I wonder how much doc linking that would break, to "fix" it to that role, though?

Going to close the PR, nobody seems sure that this makes any difference. The role of ufuncs is currently data, which could be changed (see gh-23039). So if JAX uses :func: it can't work to begin with.

I am happy to change the role :data: as said above, @rossbar what do you think?