Explicit default boundary conditions

[ali-ramadhan]

This PR gets rid of default and non-default boundary conditions. Default boundary conditions are now explicitly Periodic in x,y and FreeSlip in z.

This will make implementing multiple wall-bounded dimensions easier/cleaner.

[codecov]

Codecov Report

Merging #288 into master will decrease coverage by 0.07%.
The diff coverage is 42.85%.

@@            Coverage Diff             @@
##           master     #288      +/-   ##
==========================================
- Coverage   68.36%   68.29%   -0.08%     
==========================================
  Files          22       22              
  Lines         882      880       -2     
==========================================
- Hits          603      601       -2     
  Misses        279      279

Impacted Files	Coverage Δ
src/boundary_conditions.jl	80.64% <100%> (-1.18%)	⬇️
src/time_steppers.jl	84.09% <33.33%> (ø)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 40dbd96...51bc9e0. Read the comment docs.

[codecov]

Codecov Report

Merging #288 into master will decrease coverage by 0.29%.
The diff coverage is 14.28%.

@@            Coverage Diff            @@
##           master     #288     +/-   ##
=========================================
- Coverage   68.36%   68.06%   -0.3%     
=========================================
  Files          22       22             
  Lines         882      880      -2     
=========================================
- Hits          603      599      -4     
- Misses        279      281      +2

Impacted Files	Coverage Δ
src/Oceananigans.jl	100% <ø> (ø)	⬆️
src/time_steppers.jl	82.95% <0%> (-1.14%)	⬇️
src/boundary_conditions.jl	80.64% <100%> (-1.18%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 40dbd96...4db4b31. Read the comment docs.

[glwagner]

Looks good --- though, anticipating a replacement of the current boundary condition implementation to one using halo regions, I think we should not introduce a FreeSlip condition, but instead use a Flux condition with condition=0, because this is simpler, reduces code length, and should not produce a performance detriment.

[ali-ramadhan]

Thanks for taking a look. Good point. FreeSlip is technically not correct either as it only applies to the velocity field. I've removed it as a boundary condition and instead specify a no-flux boundary condition.

So now only doubly Periodic boundary conditions default to nothing as they're enforced via halo filling elsewhere in the code for now. The no-flux BCs will just add zeros but the performance hit should be negligible. I guess we only apply BCs in the z-dimension for now, so would be good to benchmark when x and y BCs are applied.

[glwagner]

I think its correct that Periodic boundary conditions have condition=nothing.

Shouldn't we fill halos in all 3 directions? That way everything is neat and simple and there's no special directions. It will also allow us to easily implement Periodic in the z direction at some point in the future, which may be useful for both science and validating the model against established results.

[ali-ramadhan]

Shouldn't we fill halos in all 3 directions?

That would be nice and I have the halo filling figured out for the z-direction already, but it's not within the scope of this PR.

[glwagner]

I wasn't suggesting that z-halos becomes part of the PR --- I'm referring to this part of your comment:

I guess we only apply BCs in the z-dimension for now, so would be good to benchmark when x and y BCs are applied.

I think this comment could be incorrect, because our future strategy will involve filling halos to satisfy the boundary conditions, rather than "applying" them, as is currently done in the z-direction.

[ali-ramadhan]

I think this comment could be incorrect, because our future strategy will involve filling halos to satisfy the boundary conditions, rather than "applying" them, as is currently done in the z-direction.

Ah that's true, yes, if we fully switch to halos first.