Problems with matrices and vectors

[jonasrauber]

I have multiple problems when using SIUnits with matrices and vectors.

A minimal working example of the underlying problem is shown in the following code. I would expect the second statement returning the same type as the first, but it doesn't.

julia> typeof(collect(1Hz:5Hz))
Array{SIQuantity{Int64,0,0,-1,0,0,0,0},1}

julia> typeof(collect(1:5)Hz)
Array{SIQuantity{Int64,m,kg,s,A,K,mol,cd},1}

[timholy]

This is actually a julia problem, not a SIUnits problem. The underlying issue is that promote_rule simply isn't flexible enough for SIUnits.

See JuliaLang/julia#8027 and JuliaLang/julia#8974.

[tomasaschan]

I was just about to file the same issue; I've hit some problems because of it over in JuliaMath/Interpolations.jl#19.

It becomes even more apparent that something is broken here considering the following example:

julia> A = rand(10) * Meter
10-element Array{SIQuantity{Float64,m,kg,s,A,K,mol,cd},1}:
   ...

julia> B = zeros(A)
10-element Array{SIQuantity{Float64,m,kg,s,A,K,mol,cd},1}:
   ...

julia> A + B
ERROR: Unit mismatch. Got (m ) + ()
 in error at error.jl:21
 in + at /home/tlycken/.julia/v0.3/SIUnits/src/SIUnits.jl:156
 in + at array.jl:721

@timholy, you're faster than me! :P

Could SIUnits work around this by manually tightening array types when possible (i.e. when all elements in an array result has the same unit), or would that lead to too great a performance problem?

[timholy]

I suppose SIUnits could redefine all the .*, .+, .-, and ./ methods in array.jl. That seems to be a potentially-viable solution. It's a bandaid, but not a crazy one since it's unclear when the underlying problem will be fixed (it's kinda waiting on JuliaLang/julia#8974).

[tomasaschan]

I made an attempt at defining .+ and .-, but I'm getting a bunch of ambiguity warnings about collisions between e.g. .+(x::SIQuantity{T,m,kg,s,A,K,mol,cd}, Y::StridedArray{S<:Number}) and .+(x::Number, Y::BitArray{N<:Number}), asking me to define .+(x::SIQuantity{T,m,kg,s,A,K,mol,cd}, BitArray{N<:Number}). Does that method even make sense to define?

[timholy]

You can always create error methods for such functionsmethods. Someday, julia will create those automatically, but for now you have to do it manually.

[jonasrauber]

#25 describes the same problem