Barry, That accomplishes essentially the same thing as Lisandro's proposed SNESSetLinearSolve() but is more consistent with the existing framework of PETSc. In fact, it seems *inconsistent* right now that there is a way to use a shell PC but not a way to use a shell KSP. Since it is not always "natural" to think of every custom solver as a "preconditioner" (Lisandro provides an example), it seems like there ought to be a KSP shell option. But what might this mess up?
Just my 0.5 cents. --Richard Barry Smith wrote: > > You seem to be saying you want something much like a KSP shell?
