numpy, eigen: Consider syncing up with NumPy typing?

[EricCousineau-TRI]

NumPy currently has some different sub-issues regarding this campaign:
numpy/numpy#7370 (comment)

Depending on that resolution, is it worthwhile to converge on using those type annotations (if/when they accommodate multiple dimensions, being writeable, storage order, etc.)?

Thoughts?
@rwgk @wjakob @bstaletic @YannickJadoul

FWIW I like these annotations the most so far:
ramonhagenaars/nptyping#27

[YannickJadoul]

Haven't read the whole discussion, but yes, converging on a(n) (unofficial) standard that makes sense for pybind11 sounds good to me!
I can't rid myself of the feeling that the whole numpy header is a full-blown hornets' nest, so the question is whether kicking it is a smart idea (especially right now, at a point where we're still trying to clean up, as far as I know) ?

[EricCousineau-TRI]

Oh yeah, we should leave it where it lay for now.
AFAICT, the NumPy stuff probably won't get resolved for a couple of years, at least to the extent that it'd be able to express what pybind11 needs.

Perhaps nptyping will reach that point of maturity, but yeah, still, we can definitely hold off for a while.

[YannickJadoul]

Right, OK. Still good to have somewhere on the TODO list ;-)

[rwgk]

Oh yeah, we should leave it where it lay for now.
AFAICT, the NumPy stuff probably won't get resolved for a couple of years, at least to the extent that it'd be able to express what pybind11 needs.

Perhaps nptyping will reach that point of maturity, but yeah, still, we can definitely hold off for a while.

Thanks for pointing this out Eric. Interesting ... but I'll have to watch from the sidelines until it's part of NumPy proper, or my pytype team mates start talking about it, whatever comes first :-)

[bstaletic]

Depending on that resolution, is it worthwhile to converge on using those type annotations (if/when they accommodate multiple dimensions, being writeable, storage order, etc.)?

Eventually, sure. Why not.