The CI error looks unrelated.

No output has been received in the last 10 minutes, this potentially indicates a stalled build or something wrong with the build itself.

Looks mostly good to me. The leggauss function calls legcompanion whose docstring recommends using eigvalsh to find the eigenvalues for the basis polynomials, and the leggauss comment says that it uses the fact that the companion matrix is symmetric to get better eigenvalues... so something must have been lost somewhere.

Also, the sort on the next line can be deleted because the values returned by eigvalsh are already sorted (unlike the values returned by eigvals).

Oops ;) Might be worth checking the other polynomial types to see if the error is generic.

Looks like Laguerre suffers from the same problem.

And they all have the extra sort.

EDIT: but the eigvalsh documentation says "not necessarily ordered", so either the documentation is wrong, or the sort is needed.

The np.eigvalsh are not necessarily in order, while scipy.linalg.eigvalsh` claims to return them in increasing order. Might be worth a check to see if the numpy documentation is correct.

oh you're right, I must have been looking at the scipy eigvalsh docs.

Might be worth a check to see if the numpy documentation is correct.

Well the docs say that _ssyevd and _heevd are used
https://github.com/numpy/numpy/blob/maintenance/1.9.x/numpy/linalg/linalg.py#L943

and those docs both say that the returned eigenvalues are sorted in ascending order
http://www.netlib.org/lapack/explore-html/da/da8/ssyevd_8f.html
http://www.netlib.org/lapack/explore-html/d6/dee/zheev_8f.html

Thanks @ajdawson!