missing measurements

[Benoît Pasquier]

I'm not sure it makes sense, but wouldn't it be nice to be able to deal with missing measurements?

E.g., have

x = 1.0 ± 2.0
y = 1.0 ± missing
z = missing ± missing

not throw errors and just propagate the missings for things like x+y or x+z?

[Mosè Giordano]

I've been thinking about best do this, but I think it's a bit tricky. My initial idea was to allow Measurement type to have parameter of type Union{AbstractFloat,Missing}, instead of just AbstractFloat. Note that this is the type of both the measurement value and its uncertainty, so

y = 1.0 ± missing

wouldn't be allowed, or at least it'll be automatically converted to

y = missing ± missing

I don't see great value in separating the types of the value and the uncertainties, this helps us keeping type-stability in many situations.

The idea proposed above is doable, it just requires some work to be done -- help is always much appreciate! What worries me most about my suggestion is that you can have a missing measurement Measurement{Missing}, but this will be subtype of AbstractFloat instead of Missing: we can certainly define

Base.ismissing(::Measurement{Missing}) = true

but any code relying on a missing value being subtype of Missing would fail to detect it.

[Mosè Giordano]

Well, it turned out that this can be done in a much simpler way, see #62. Thanks to @simeonschaub!