Tomás Chor

I did a few tests with some criteria for timestep-skipping with a couple of my own simulations in addition to the MWE included here. In summary: 1. Criterion `sim.Δt /...

@glwagner are you okay if I just add `min_Δt` as a property of `NonhydrostaticModel` and maintain the strategy of skipping the timestep is Δt is smaller than that? I think...

> This isn't a problem of the nonhydrostatic model specifically. Why would we add it as a property there? This PR is supposed to fix the output of the pressure,...

> I think we should fix the problem once. Otherwise we'll end up with unnecessary code somewhere that has to be deleted. @glwagner Can you please be clearer? Does that...

> Looks good, maybe bump the minor version Done! I wanna test this code in my actual simulation before sending it off for reviews though. Also, I'm thinking of using...

I tested this branch on my simulation and things work fine. However, when trying to include a test I came across some weird behavior. Namely the snippet below fails: ```julia...

> > So there seems to be some attribute of `model` (presumably `model.clock`?) that's leading to different `Simulation` behavior. Is this expected? I couldn't quite figure out what was happening....

@wenegrat @zhazorken This should fix the pressure issues when writing to disk. All you need to do is add the keyword argument `minimum_relative_step=1e-10` (or some other very small positive number)...

> [@tomchor](https://github.com/tomchor) The issue is the preconditioner used in `ConjugateGradientPoissonSolver`. I changed them a little bit to be SPD (mandatory for Krylov methods) in [#4041](https://github.com/CliMA/Oceananigans.jl/pull/4041). > > Maybe the modifications...

> We added a new solver based on Krylov.jl but didn't changed the `ConjugateGradientPoissonSolver`. The only thing that I modified and could impact this solver is the preconditioner. Thanks for...