support AbstractVector

[stevengj]

Rather than declaring arguments (and struct fields) as Vector{IntPoint}, it would be more generic to allow AbstractVector{IntPoint}. This can still be passed to C libraries as long as it supports a pointer conversion, and you could check that the stride(array,1) == 1 so that it is contiguous — these will throw a MethodError for array types that don't support these operations.

(It also might be convenient to use const IntPoint = SVector{2,Int64}.)

See also this discourse discussion

[plut]

I agree that passing vectors instead of structures is much better (I wrote some wrappers in my own code), since vectors natively support linear algebra for example.

However for point types there are two and a half possibilities that come to mind that would be even better than using a SVector{2,Int64}:

• use a labelled vector (https://github.com/SciML/LabelledArrays.jl), or
• even more Julian: use an abstract type (StaticArray{Tuple{2}} comes to mind — this would allow both SVector and labelled vectors).
• also, the coordinate type should also be left open (I use some fixed-point real numbers), as long as it is 64 bits long. This can even be enforced on the Julia side using type traits (dispatching via Val(sizeof(T))).