Hard-coded Float64 values

[glwagner]

I suspect there will be people cropping up (👀 @ali-ramadhan) want to use Float32. This will thwart them:

OceanBioME.jl/src/Models/AdvectedPopulations/NPZD.jl

Line 320 in 3c287d6

@inline redfield(::Union{Val{:P}, Val{:Z}, Val{:D}}, bgc::NPZD) = 6.56

All that's needed is

@inline redfield(::Union{Val{:P}, Val{:Z}, Val{:D}}, bgc::NPZD{FT}) where FT = convert(FT, 6.56)

The key is to never write decimals in source code without convert.

[johnryantaylor]

Thanks for the tip @glwagner!

[ali-ramadhan]

Not sure if nitpicking but would

@inline redfield(::Union{Val{:P}, Val{:Z}, Val{:D}}, bgc::NPZD{FT}) where FT = FT(6.56)

be better than

@inline redfield(::Union{Val{:P}, Val{:Z}, Val{:D}}, bgc::NPZD{FT}) where FT = convert(FT, 6.56)

to avoid extra layer(s) of method inference during compilation?

[glwagner]

Well, I thought they were the same!

In Oceananigans source code we have been trying to migrate to convert(FT, x) simply because we felt it was more semantically obvious than merely FT(x) which is easier to forget exactly what it does first thing in the morning or at the end of a long day (well, it converts x to FT of course).

[ali-ramadhan]

That makes sense! For floats it's a super simple method inference. Maybe for more complex types there's more extra inference. But agree it's good to be more semantically obvious!