Should be working now

@zasdfgbnm what's your opinion on name conflicts between input and output arguments? An alternate resolution is to say that, the output name always has to be distinct from input names (and then we have to pick a name like e and V, which I guess is worse than eigenvalues and eigenvectors for non-mathematicians, but is not totally unprecedented, since we have a bunch of other very shortly named identifiers.)

This PR looks functionally correct but I want to discuss the matter of _return some more

@ezyang I am OK with short names in general, like Q, R, but personally I don't like the name e and V for this one. Because the e and V are not as widely adopted as something like Q and R for QR decomposition. IMHO, it's good to use abbreviations only when there is a common way to abbreviate.

I don't thin there is a common way to abbreviate eigenvalues and eigenvectors, below are a list of abbreviations in different materials (these are all top ranking result in Google when searching eigen values):

• lambda and X: http://mathworld.wolfram.com/Eigenvalue.html
• lambda and v: https://en.wikipedia.org/wiki/Eigenvalues_and_eigenvectors
• lambda and eta: http://tutorial.math.lamar.edu/Classes/DE/LA_Eigen.aspx

SGTM. @soumith suggests we use pivot instead of piv though :)

@ezyang Just renamed piv -> pivot

Need to rebase after landing #16186, because we need to update the whitelist of the new unit test added in #16186.

landed

@ezyang Please review #17093 before #17195. As the more and more operators returning structseq, the test test_namedtuple_return becomes long. I will do a refactor on test_namedtuple_return in #17195 to reduce its size.