I haven't looked much in detail but this is a great idea to define the API with respect to Time missing values and scalars. You may want to put [skip ci] in each commit until tests are passing locally.

General discussion in #6509 - here just a note that if you want np.isna[nt] to work on Time, you'll need to define an __array_ufunc__ to tell how it should treat Time instances. If you'd also want things like np.nanmean to work, one would likely also need to define __array_function__ (which is planned a bit anyway, for np.linspace - see #8610.

All that said, it would be good if all these tests passed! (Except as @taldcroft noted, remove tests on private properties - those should remain implementation details - possibly including how to get isnan to work!)

@mhvk yes, thanks for the pointers.

This is what I started with, but I have not yet a working version

@mhvk I started working on this again. But defining array_ufunc is tricky. I defined it only for isnan and isnat but now stuff starts failing because what used to work, like:

np.equal(time, 0) now also calls ufunc.

Do I now have to implement every single numpy ufunc in Time.__array_ufunc. There has to be a better way.

@maxnoe - indeed, every ufunc has to be dealt with. There is a mixin to let all operators go via the same path, but I'm not sure that is too helpful in this case.

That said, I actually hadn't appreciated any ufuncs worked at all on Time! Are there any tests that are actually failing?

But on reflection/inspection, I see what happens: the ufunc creates an array of Time scalars and let those subtract:

In [9]: t = Time(np.arange(50000., 50003.), format='mjd')                                                                                                                

In [10]: np.subtract(t, t)                                                                                                                                   

Out[10]: 
array([<TimeDelta object: scale='tai' format='jd' value=0.0>,
       <TimeDelta object: scale='tai' format='jd' value=0.0>,
       <TimeDelta object: scale='tai' format='jd' value=0.0>],
      dtype=object)

So, dealing with those will make things better... (right now, for 10 elements t==t is 100 times faster than np.equal(t, t)).

In any case, my sense would be to first of all think of ufuncs where we're adding new functionality, and then worry about those that we've broken.

Hello. I converted your PR to a draft PR due to #10706 .

Hi humans 👋 - this pull request hasn't had any new commits for approximately 1 year, 14 days. I plan to close this in 30 days if the pull request doesn't have any new commits by then.

In lieu of a stalled pull request, please consider closing this and open an issue instead if a reminder is needed to revisit in the future. Maintainers may also choose to add keep-open label to keep this PR open but it is discouraged unless absolutely necessary.

If this PR still needs to be reviewed, as an author, you can rebase it to reset the clock.

If you believe I commented on this pull request incorrectly, please report this here.

Hi humans 👋 - this pull request hasn't had any new commits for approximately 1 year, 14 days. I plan to close this in 30 days if the pull request doesn't have any new commits by then.

In lieu of a stalled pull request, please consider closing this and open an issue instead if a reminder is needed to revisit in the future. Maintainers may also choose to add keep-open label to keep this PR open but it is discouraged unless absolutely necessary.

If this PR still needs to be reviewed, as an author, you can rebase it to reset the clock.

If you believe I commented on this pull request incorrectly, please report this here.

Sorry for the duplicate bot posts. I fixed the bug upstream, hopefully.

I'm going to close this pull request as per my previous message. If you think what is being added/fixed here is still important, please remember to open an issue to keep track of it. Thanks!

If this is the first time I am commenting on this issue, or if you believe I closed this issue incorrectly, please report this here.