Missing: binary operations with scalars about oldimmutablearrays.jl

[dronir]

The code to generate the types and operations is mostly past the horizon of my own metaprogramming skills, so I can't jibble with it much.

[twadleigh]

There will still probably need to be tweaking of exactly which binary operators are defined, and how, but that task should be a bit more accessible to all now.

[dronir]

This gives me a kilometer of ambiguity warnings when using the package, and division still doesn't work.

julia> using ImmutableArrays
Warning: New definition 
    +(Vector2{T},T) at /Users/owilkman/.julia/ImmutableArrays/src/ImmutableArrays.jl:75
is ambiguous with 
    +(T,Vector2{T}) at /Users/owilkman/.julia/ImmutableArrays/src/ImmutableArrays.jl:71.
Make sure 
    +(Vector2{T},Vector2{T})
is defined first.

[something like 50 of these warnings]

julia> a = Vector3{Float64}(0.1, 0.2, 0.3)
3-element Float64 Vector3:
 0.1
 0.2
 0.3

julia> a / 3
ERROR: no method ./(Vector3{Float64},Int64)
 in / at abstractarray.jl:260

julia> a / 3.0
ERROR: no method ./(Vector3{Float64},Float64)
 in / at abstractarray.jl:260

[twadleigh]

The warnings can be eliminated by commenting out one of the left or right applications of the scalar. This is also an issue, for instance, for the DataFrames package. There is currently no way to supress warnings.

[twadleigh]

Methods that apply automatic coersions for scalars of mixed type can be added but will require a further generalization of the method definition functions to allow for specifying a second scalar type.

[twadleigh]

I generalized the type of the scalar argument to Number. This had the benefit of both suppressing the warnings and causing the new test script to pass without complaint. Hopefully the JIT compilation will ensure that we don't loose any performance.

[dronir]

Last night I was very bored, but not in a very creative mood, so I took the trouble of writing an immutable array package up to size 4 completely explicitly: https://github.com/dronir/MyLittleArray

I probably won't develop it any further, and it may have bugs, but I'll use it to compare to the generated code we have here. I'm slightly surprised that with that latest change, ImmutableArrays.jl actually even has slightly better performance than MyLittleArray.

[twadleigh]

@dronir : It may be useful to have your hand-coded library (or some version of it) in the test suite to test for semantic or performance regressions. Thanks again for your contributions.

[dronir]

This project is dear to me, because most of what I do are things where these immutable arrays are usually very good.