Allowing the user to pass input using several formats usually is handled within julia by defining specialized methods for each function and argument type(s). This can prove to be inefficient when the amount possible combinations of these increases. AbstractCollection is a Dict meant to enable the user to profit from a fixed interpretation describing different collections of mathematical objects, while also simplifying the life of the developer, also resulting in less code duplication.


Commonly the same kind of information, e.g. an amount of PointVectors, is accepted as argument for many different functions. The user can chose from different types (coming with an interpretation of their content) to use when calling one of these functions to describe the data. This data is then converted to a type and format Polymake.jl (and thus indirectly the polymake kernel) supports.


Usually in polymake, a collection of points is displayed as a matrix of row-vectors. Such a matrix is always created from the input information. When writing a new function accepting an object x of type AbstractCollection[PointVector] (note that, with AbstractCollection being a Dict, its entries are accessed using square brackets; the keys are the Oscar types of the elements of the collection), the necessary conversion can (and should) be called at the beginning. These conversion functions already exist and support all of the types stated in Type compatibility. In this case the function is homogenized_matrix(x, 1).

RayVectors and their collections work about the same; the main difference for the programmer is that homogenized_matrix(x, 0) is called.

When looking at the beginning of the convex_hull method, the corresponding conversions of the three arguments V, R and L can be seen:

function convex_hull(::Type{T}, V::AbstractCollection[PointVector], R::Union{AbstractCollection[RayVector], Nothing} = nothing, L::Union{AbstractCollection[RayVector], Nothing} = nothing; non_redundant::Bool = false) where T<:scalar_types
    # Rays and Points are homogenized and combined and
    # Lineality is homogenized
    points = stack(homogenized_matrix(V, 1), homogenized_matrix(R, 0))
    lineality = isnothing(L) || isempty(L) ? zero_matrix(QQ, 0, size(points,2)) : homogenized_matrix(L, 0)


Conversion functions

So effectively supporting AbstractCollections only requires to know when to apply which conversion function. The following table explains this for AbstractCollection[T]:

TTarget formatConversion function
PointVectormatrix of row-vectorshomogenized_matrix(*, 1)
RayVectormatrix of row-vectors (linear setting)unhomogenized_matrix(*)
RayVectormatrix of row-vectors (affine setting)homogenized_matrix(*, 0)
LinearHalfspace/LinearHyperplaneinequality/equation matrix (linear setting)linear_matrix_for_polymake(*)
AffineHalfspace/AffineHyperplaneinequality/equation matrix (affine setting)affine_matrix_for_polymake(*)