Promote problems

[diegozea]

I have a type which only accept Float64 values defined as:

@auto_hash_equals immutable Coordinates <: FixedVectorNoTuple{3, Float64}
    x::Float64
    y::Float64
    z::Float64
    function Coordinates(a::NTuple{3, Float64})
        new(a[1], a[2], a[3])
    end
end

I getting this errors trying to divide it with an Int:

I have solved it in the meanwhile using Float64(length(...))

[c42f]

Yup, this is one of many promotion problems which arise when working with mixed types. I've started work on a rewrite of the map functionality which will hopefully address these type woes. See #105.

[c42f]

Ah darn! #105 was going to fix this, but in the process of fixing a whole bunch of other things, your specific type of FSA subtype no longer works, since similar is used in a rather central way in the new implementation, and the default similar doesn't work well with your Coordinates.

We'll need to add something like Coordinates to the test cases, and come up with a coherent story around the role and API of similar for FSAs. In the meantime, if you add the following (caveat, the sz parameter may be a bad API choice for FSAs and may need to change):

Base.similar(::Type{Coordinates}, ::Type{Float64}, sz::Tuple) = sz[1] == 3 ? Coordinates : Vec{sz[1],Float64}
Base.similar{T}(::Type{Coordinates}, ::Type{T}, sz::Tuple) = Vec{sz[1],T}

Then you can do things like:

julia> points = [rand(Coordinates) for _ =1:10]
10-element Array{Coordinates,1}:
 Coordinates(0.7776985119194051,0.29719853453510536,0.10251763435148575)
 Coordinates(0.7379174083324038,0.3296379139698149,0.36391270520448904)
 Coordinates(0.4283095362624538,0.7821447202577865,0.14133737599972518)
 Coordinates(0.7555352303842677,0.520333541723871,0.9614920004560281)
 Coordinates(0.7255490300531384,0.8876056197246123,0.27401014626516274)
 Coordinates(0.0388837350288056,0.9984448841471119,0.6013870295464301)
 Coordinates(0.3917672241105896,0.5867953116084399,0.8013279650490599)
 Coordinates(0.3705099593706276,0.06483885178084137,0.8097820812141754)
 Coordinates(0.9703794802422856,0.934004150420159,0.11295307373487251)
 Coordinates(0.2978305066721063,0.9149579629444147,0.9108417622883043)

julia> mean(points)
Coordinates(0.5494380622376084,0.6315961491112156,0.5079561774109733)

julia> points[1] .< points[2]
Vec(false,true,true)

[diegozea]

the default similar doesn't work well with your Coordinates.

Is this problem caused by the use of Float64 instead of using a T parameter?

[c42f]

Is this problem caused by the use of Float64 instead of using a T parameter

Yes, this is why our tests didn't pick it up (the tests use an RGB{T} type as the prototypical FixedVectorNoTuple), but I don't think you're actually doing anything wrong, I think FixedSizeArrays needs some fixes to make your use case work nicely.

[c42f]

We're trying to figure out what to do about this to make your use case smooth - see #116 for API discussion.

[c42f]

This should now all be sorted out in #118, and you won't even need to overload similar :-)

You get the following:

julia> Coordinates(1,2,3) .< Coordinates(1,2,4)
Vec(false,false,true)

julia> mean([Coordinates(1,1,1) for i=1:10])
Coordinates(1.0,1.0,1.0)

julia> sum([Coordinates(1,1,1) for i=1:10])
Coordinates(10.0,10.0,10.0)

[diegozea]

Awesome! Thanks!! :D