RalphAS

The cause of the failure reported by DSTEMR for certain matrices of large norm is an inconsistency in scaling used in a test in DLARRF: https://github.com/Reference-LAPACK/lapack/blob/54261473ab03f5c943317934611680b6dfddfd31/SRC/dlarrf.f#L467-L477 `SMLGROWTH` scales with the...

Try more iterations. It currently fails about 1/500 times with spurious eigenvalues and terrible residuals.

Increasing the iteration count in the script above to a few thousand reliably gets a few failures on my system with the current master branch. Note that ArnoldiMethod normally starts...

IIUC this arises because the new Pkg system doesn't handle OS-conditional dependencies yet. Similar annoyances arise for Linux users. I'd like to ask @staticfloat if it is reasonable to wrap...

There's a Krylov-subspace approximant option using Arpack; I've only done minimal testing of matrix-free methods (see linmap.jl in the test folder). I'll plan to make that usable without all the...

Thanks! When I wrote most of this (several years ago) I found working with Recipes very frustrating and didn't pursue them, but if you're able to get them to handle...

I agree in principle, but so far I have been unable to write recipes that plot line art on top of contour plots, whereas that is easy with top-level Plots...

Thanks, I'll put that in the next patch.

It has been handy to have a separate SVD for compatibility with GenericSchur. I have recently tried to make the latter compatible with GenericLinearAlgebra; if that effort is successful then...

I agree with @andreasnoack position above: I don't think the pirate packages should be dependencies when they are just "friends", and end users should be directed to add them by...