derivative of ParametricSpline is type-unstable and allocating

[stevengj]

The return value of derivative is inferred as Any (i.e., it is type-unstable). This seems like a bug?

Furthermore, it allocates a Vector{Float64} for its return value. Since the dimensionality is known from the ParametricSpline type, couldn't we use SVector{2,Float64} instead? StaticArrays.jl is pretty widely used these days, so it is not crazy to add a dependency on it. (Alternatively, you could add an API that returns a tuple.)

julia> using Dierckx, Test

julia> spl = ParametricSpline(1:10, rand(2, 10))
ParametricSpline(knots=[1.0,3.0 … 8.0,10.0] (8 elements), k=3, extrapolation="nearest", residual=0.0)

julia> derivative(spl, 5.1) # allocates
2-element Vector{Float64}:
 -0.6756394220200724
  0.1203327362940756

julia> @inferred derivative(spl, 5.1) # type-unstable
ERROR: return type Vector{Float64} does not match inferred return type Any

[stevengj]

Looks like the type-instability comes from the dims=[2] pattern that is used in the source, which is type-unstable because the length of the dims array is not known until runtime.

The fix should be simple: use dims=2 instead of dims=[2]. (Or if there are multiple dims, use a tuple.)