Update options.jl to use @for_petsc macro

[jkozdon]

Couple changes that happen with this PR

• For PETSc arguments that have no input, nothing not true should be used, e.g., ksp_view = nothing.
• Add parse_options for parsing command line arguments

[jkozdon]

@boriskaus and @nicoberlie: All the other PRs shouldn't have changed functionality of to code much. This one might impact you more, since it changes the behavior of the options.

In general, I am 100% willing to help fix the drama I created with my slew of PRs today. If you push out the update doc strings, I can try to fix any other drama I have created for you with #119

[boriskaus]

if you have a large number of options, it is often useful to be able to temporarily deactivate one of them, say snes_monitor. Is that still possible now that we should use snes_monitor=nothing rather than snes_monitor=true/false?

[jkozdon]

if you have a large number of options, it is often useful to be able to temporarily deactivate one of them, say snes_monitor. Is that still possible now that we should use snes_monitor=nothing rather than snes_monitor=true/false?

My thought is that options like snes_monitor have no values, so setting nothing in julia makes more sense to me (since nothing represents no value).

So we use this

S = PETSc.SNES{Float64}(MPI.COMM_SELF; 
        snes_rtol=1e-12, 
        snes_monitor=nothing,
        snes_converged_reason=nothing);

instead of

S = PETSc.SNES{Float64}(MPI.COMM_SELF; 
        snes_rtol=1e-12, 
        snes_monitor=true, 
        snes_converged_reason=true);

If you want to disable it, you just wouldn't include it.

If that's not going to work with your workflow I am fine undoing this.

[boriskaus]

yes I see.
The way I would normally do this in PETSc is to slightly change the name (e.g., call it -snes_monitor1 instead of -snes_monitor) to temporarily disable that. We can ofcourse do that here as well or, as you say, not include the option to start with. In a complex setup, however, you will have many options and this last case would require you to delete lines. That is why I believe setting it to false to disable is a handy feature to have. Is it possible to check if an option is nothing or false, and in case of false not include it in the options DB?

[jkozdon]

Yes, we can certainly do that. Will update the PR with this behavior.