Eric Wieser

I think the only fix available to us is to always use `np.longlong` for default integer values, rather than `np.int_` which is sometimes 64-bit and sometimes 32-bit. We can't make...

> Understood. Then I will suggest that when the user passes an array of identical values of the same size as the f, numpy could provide a warning instead. Right...

bors r-

Also, can you update the PR description? It's no longer accurate

Thanks! bors d+ Feel free to merge once CI is happy

bors merge

Right now I think the problem is that `np.equal.types` contains only `XX->?` loops, and you want an `OO->O` loop. I think it would be pretty reasonable for numpy to add...

> Hup, it looks like there's a method numpy.set_numeric_ops (that has __doc__, but I can't find its online documentation), referenced in this post Note that just the other day [an...

Let me try and be a little more precise about the route I'm envisaging: 1. numpy should implement the `'OO->O'` loop for the comparison functions, but not expose it via...

> we'd just need to override __array_ufunc__ to use OO->O to get a Formula from expr1 == expr2, and for AutoDiffXd, we'd not need to override anything to get OO->B...