Ilhan Polat

What I also noticed is that if I transpose and use the `uplo='U'` it doesn't have a problem.

Ah yes, sorry for the terse comment; what I meant is if you transpose the original matrix so that the interested part is on the upper triangle and then use...

A similar to the original problem namely dsyevr failure is reported again to SciPy ```python a = np.array([[0.9349689745783974, 0.0225235320981566, -0.05953165551175491, 0.014890259948252545, -0.12836839955836318, -0.0480224626614684, -0.06886406875564752, 0.04046562569985317, 0.03714156517414783, -0.018402704526518737, 0.07197525245559412], [0.02252353209815662, 0.9230010962223161,...

I'll cook up an example like the original post as soon as I switch to my other computer.

Here is an example that leads to info=2 in DSYEVR call on my machine ```python cimport cython cimport numpy as cnp import numpy as np from scipy.linalg.cython_lapack cimport dsyevr cdef...

This is the same machine as the original post (with OpenBLAS 3.11) the users are using MKL and OpenBLAS too. However @oamarques could also reproduce this with his analysis above...

They just forgot to add 3.9.0 to `lapack-release` probably. Seems like an SVN habit to have this as trunk and the other as the release copy. But hardly ever, people...

Numerically returning an identity matrix is a waste and it will take time to create that identity matrix for no reason if the matrix is empty. This would make many...

Ah indeed you are right, my bad. I've overlooked the JOB types issue. It still feels awkward but I see the point for cases where singular vectors are requested. However...

Can you also add which query leads to LWORK=0?