Support for Missings

[cstjean]

We're transitioning to missing values in our code base, instead of NaN, and find that there are some... missing... methods, like uconvert(::Unitful.FreeUnits, ::Missing). I assume that it just hasn't been implemented yet? If so, I'll make a PR.

[ajkeller34]

Missing support sounds like a good idea but I haven't been following along with the latest thinking about it. What would you propose this method should do? Just return missing?

[cstjean]

Yeah. We already have that 1u"m" * missing === missing, because Quantity <: Number.

[ajkeller34]

Sounds good, thanks!

[nicoleepp]

Additionally methods like:

*(y::Missing, x::T) where {T<:Units} = missing
ustrip(x::Missing) = missing

would be good to have

[rafaqz]

Any idea of the set of methods needed for a complete Unitful missings PR?

[cstjean]

I've been using

Unitful.uconvert(::Any, ::Missing) = missing
Unitful.ustrip(::Missing) = missing
Unitful.unit(::Type{Union{Missing, T}}) where T = unit(T)
Base.:*(::Missing, ::Unitful.Units) = missing
Base.:/(::Missing, ::Unitful.Units) = missing

and it works fine. It has been heavily tested. If anyone wants to PR that code, please do.

[rafaqz]

Also

Base.:^(::Unitful.Units, ::Missing) = missing

?

[rafaqz]

Should unit(x::Missing) return missing or NoUnit?

The same question also applies to absoluteunit, anddimension(::Missing) returning missing or NoDims

[cstjean]

missing. Otherwise unit.([1.0m, missing, 3.3m]) would be strange.

[ajkeller34]

Closed by #208