krylov space re-use [BifurcationKit jacobian-free methods optimization]

[rveltz]

Hi,

I am (kinda re-) opening an issue because I think it is different from the issues on memory allocations that are still open. In BifurcationKit.jl, we solve repeatedly linear systems and compute eigenvalues. I am now improving BifurcationKit for Jacobian-free methods by avoiding allocations when possible. I love KrylovKit because it is robust and have the most general matrix-free eigen solver.

Thus, it would be nice to allow

• for inplace linear operator
• re-use of Krylov subspace (memory) ; this could be increased if needed by linsolve or eigsolve. Is it something you have in mind?

[Jutho]

Yes I agree that this is becoming more pressing. I was kind of hoping that work on Julia's GC would make this unnecessary, but it seems like this is not going to happen any time soon.