Anyone have ideas about how I might add functionality for the following call?

>>> import numpy as np
>>> np.random.randint(5, [10, None])

It makes sense to allow this call given that we allow high to be None currently. In my PR, it doesn't work because numpy will throw a type error when trying to convert the None element to long. However, I can't think of anything "fast" / segfault-safe that isn't iterating through both the low and high arrays.

np.random.randint(5, [10, None])

I wouldn't bother... just tell people to write randint([5, 0], [10, 5]) if that's what they want. (It'll be infinitely less confusing.)

Okay, that will make things a lot easier. Will add that to documentation.

Documentation added, and Travis is happy. I have no idea how this will gel with #6824 or #6869, but as it stands, unless there are some major style/functional issues I'm missing, I think it should be good to go.

In light of the email discussion, would it be safe to close this PR along with the related issue (#6745)?

☔ The latest upstream changes (presumably #6910) made this pull request unmergeable. Please resolve the merge conflicts.

@gfyoung I'll close this as it would need a complete rewrite at this point. I'll leave the issue open just in case someone wants to pursue it. Thanks for your work here.