Preconditioners do not work correctly

[termi-official]

Describe the example

Taking the example from the docs here and putting NaN into the preconditioner does not change the solver behavior at all.

Minimal Reproducible Example 👇

import LinearSolve as LS
import LinearAlgebra as LA

n = 25
A = rand(n, n)
b = rand(n)

precs = Returns((LS.InvPreconditioner(NaN*ones(n)), LA.Diagonal(NaN*ones(n))))

prob = LS.LinearProblem(A, b)
sol1 = LS.solve(prob, LS.KrylovJL_GMRES(verbose=1,precs));
sol2 = LS.solve(prob, LS.KrylovJL_GMRES(verbose=1));

Error & Stacktrace ⚠️

None. Which is the problem.

Not Working Environment (please complete the following information):

[7ed4a6bd] LinearSolve v3.48.0

[termi-official]

Seems like we either regressed or the update to the new interface was only partial in first place, as the following errors at least.

sol3 = LS.solve(prob, LS.KrylovJL_GMRES(verbose=1); Pl=Returns(LS.InvPreconditioner(NaN*ones(n))));

[oscardssmith]

Shouldn't this be sol1 = LS.solve(prob, LS.KrylovJL_GMRES(verbose=1; precs)); (i.e. precs is a keyword arg?)

[oscardssmith]

did that fix it?

[termi-official]

I am still tracking it down. But this part seems to be fixed for now.