Getting started

Installation

To install directly from the repository before a possible Julia General registry release:

using Pkg
Pkg.add(url = "https://github.com/yolhan83/GooseArrays.jl")

Dynamic and static arrays

using GooseArrays, LinearAlgebra

result = with_datastack(1 << 20) do s
    x = GooseVector{Float64}(s, undef, 4)
    fill!(x, 2.0)
    y = StaticGooseVector{Float64,4}(s, undef)
    y .= 2 .* x .+ 1
    (sum(y), Array(y))  # independent copy outside the arena
end
result

(20.0, [5.0, 5.0, 5.0, 5.0])

GooseArray{T,N} stores dimensions outside the type, making it suitable for lengths chosen at runtime. StaticGooseArray{T,N,S} encodes the shape in its type, trading additional specialization for shape-aware code.

Scopes and invalidation

with_datastack(4096) do s
    outer = GooseArray(s, [1, 2, 3])
    escaped = with_scope(s) do scope
        a = GooseArray(scope, [4, 5, 6])
        copy(a)  # copy within the same arena; invalid after scope exit
    end
    (sum(outer), try
        sum(escaped)
    catch err
        typeof(err)
    end)
end

(6, InvalidatedArrayError)

To retain data beyond with_scope or with_datastack, call Array(a) before the scope ends. Never retain or reuse raw pointers into an arena after it has been closed or reset.

Reductions and linear algebra

map, map!, broadcasting, and standard reductions work with Goose arrays. Out-of-place map creates its result in the first Goose operand's arena; numeric mapreduce can avoid materializing an intermediate array. LinearAlgebra also provides mul!, dot, axpy!, axpby!, and compatible static-shape products.

Runnable examples are available in examples/.