I was reading this Github page, and came across a post from StefanKarpinski, and he says:
…but you really mostly shouldn’t be doing much computation with tuples in the first place,…
I was wondering if someone could elaborate on why? Is it because tuples are immutable and you may need to construct a new tuple every time a value needs to be updated?
I think because they are not meant for it. If you want to do computations you should use StaticArrays.jl, which wraps tuples.
Notice that that comment is almost eight years old. At the time tuples had very different performance characteristics, predating the classic PR Tuples made me sad (so I fixed them) by Keno · Pull Request #4042 · JuliaLang/julia · GitHub, and dot broadcasting had not been invented.
You may be interested in https://github.com/JuliaArrays/StaticArrays.jl . SArray provides an array-like interface around data stored in a tuple.
https://github.com/JuliaArrays/StaticArrays.jl/blob/master/src/SArray.jl#L18-L30
The is mainly practical for small arrays.
One would have to implement all mathematically sensible computations for tuples. For example, this is not implemented:
julia> x = (1,2)
(1, 2)
julia> 2*x
ERROR: MethodError: no method matching *(::Int64, ::Tuple{Int64,Int64})
The solution to such implementation exists and is the StaticArrays package that everyone is mentioning. StaticArrays are the tuples for which the computations can be performed:
julia> using StaticArrays
julia> x = SVector{2,Float64}(1,2)
2-element SArray{Tuple{2},Float64,1,2} with indices SOneTo(2):
1.0
2.0
julia> 2*x
2-element SArray{Tuple{2},Float64,1,2} with indices SOneTo(2):
2.0
4.0
Dot broadcasting works just fine with plain tuples.
julia> 2 .* (1, 2)
(2, 4)
It’s extremely convenient for e.g. size computations.
Sure, but multiplication by a scalar does not, summing tuples does not, multiplication by a matrix does not etc. Probably if it was chosen to put tuples as center type for computations those should have been implemented as well.