add minimalvec (#22)

* add minimalvec

* fix test

* run invalidations on lts

* Remove non-minimalvec tests

Co-authored-by: Lukas Devos <[email protected]>

---------

Co-authored-by: Lukas Devos <[email protected]>

.github/workflows/Invalidations.yml

@@ -18,7 +18,7 @@ jobs:
     steps:
     - uses: julia-actions/setup-julia@v2
       with:
-        version: '1'
+        version: 'lts'
     - uses: actions/checkout@v4
     - uses: julia-actions/julia-buildpkg@v1
     - uses: julia-actions/julia-invalidations@v1

Project.toml

@@ -1,7 +1,7 @@
 name = "VectorInterface"
 uuid = "409d34a3-91d5-4945-b6ec-7529ddf182d8"
 authors = ["Jutho Haegeman <[email protected]> and contributors"]
-version = "0.4.9"
+version = "0.5"
 
 [deps]
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
@@ -10,7 +10,7 @@ LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 Aqua = "0.6, 0.7, 0.8"
 LinearAlgebra = "1"
 Test = "1"
-TestExtras = "0.2"
+TestExtras = "0.2,0.3"
 julia = "1"
 
 [extras]

src/VectorInterface.jl

@@ -43,4 +43,7 @@ include("namedtuple.jl")
 # General fallback implementation: comes with warning and some overhead
 include("fallbacks.jl")
 
+# Minimal vector type for testing
+include("minimalvec.jl")
+
 end

src/fallbacks.jl

@@ -24,7 +24,6 @@ function scalartype(T::Type)
     elT === T && throw(ArgumentError(_error_message(scalartype, T)))
     return scalartype(elT)
 end
-# should this try to use `eltype` instead? e.g. scalartype(T) = scalartype(eltype(T))
 
 # zerovector & zerovector!!
 #---------------------------
@@ -102,8 +101,8 @@ end
 function scale!(y, x, α::Number)
     T = Tuple{typeof(y),typeof(x),typeof(α)}
     @warn _warn_message(scale!, T) maxlog = 1
-    if applicable(LinearAlgebra.mul!, y, x, α)
-        return LinearAlgebra.mul!(y, x, α)
+    if applicable(LinearAlgebra.mul!, y, x, α, true, false)
+        return LinearAlgebra.mul!(y, x, α, true, false)
     else
         throw(ArgumentError(_error_message(scale!, T)))
     end
@@ -112,8 +111,9 @@ end
 function scale!!(y, x, α::Number)
     T = Tuple{typeof(y),typeof(x),typeof(α)}
     @warn _warn_message(scale!!, T) maxlog = 1
-    if applicable(LinearAlgebra.mul!, y, x, α) && promote_scale(y, x, α) <: scalartype(y)
-        return LinearAlgebra.mul!(y, x, α)
+    if applicable(LinearAlgebra.mul!, y, x, α, true, false) &&
+       promote_scale(y, x, α) <: scalartype(y)
+        return LinearAlgebra.mul!(y, x, α, true, false)
     else
         α_Ty = α * one(scalartype(y))
         if applicable(*, x, α_Ty)

src/minimalvec.jl

@@ -0,0 +1,110 @@
+"""
+    MinimalVec{M,V<:AbstractVector}
+    MinimalVec{M}(vec::V) where {M,V<:AbstractVector}
+
+Wraps a vector of type `V<:AbstractVector` in such a way that the wrapper only supports the
+minimal interface put forward by VectorInterface.jl. The type parameter `M` can take the
+values `true` or `false` and determines whether the vector behaves as a mutable vector that
+supports in-place operations (`M == true`) or whether it behaves as an immutable or
+static vector (`M == false`).
+
+This wrapper can be used to test whether an algorithm is implemented using only the minimal
+interface of VectorInterface.jl, without relying on other methods that would for example
+be available for `AbstractVector` or `AbstractArray`.
+
+To unwrap the contents of a `v::MinimalVec` instance, the field access `v.vec` can be used.
+
+See also [`MinimalMVec`](@ref) and [`MinimalSVec`](@ref) for convenience constructors.
+"""
+struct MinimalVec{M,V<:AbstractVector}
+    vec::V
+    function MinimalVec{M,V}(vec::V) where {M,V}
+        M isa Bool || throw(ArgumentError("first type parameter must be `true` or `false`"))
+        return new{M,V}(vec)
+    end
+    MinimalVec{M}(vec::V) where {M,V} = MinimalVec{M,V}(vec)
+end
+"""
+    const MinimalMVec = MinimalVec{true}
+    MinimalMVec(v::AbstractVector)
+
+Type alias for `MinimalVec{true}`, representing a vector wrapper that implements the minimal
+interface of VectorInterface.jl, including in-place operations (!-methods).
+
+See also [`MinimalVec`](@ref) and [`MinimalSVec`](@ref).
+"""
+const MinimalMVec{V} = MinimalVec{true,V}
+"""
+    const MinimalSVec = MinimalVec{false}
+    MinimalSVec(v::AbstractVector)
+
+Type alias for `MinimalVec{false}`, representing a vector wrapper that implements the
+minimal interface of VectorInterface.jl, excluding in-place operations (!-methods).
+
+See also [`MinimalVec`](@ref) and [`MinimalMVec`](@ref).
+"""
+const MinimalSVec{V} = MinimalVec{false,V}
+
+MinimalMVec(v::AbstractVector) = MinimalVec{true}(v)
+MinimalSVec(v::AbstractVector) = MinimalVec{false}(v)
+
+_ismutable(::Type{MinimalVec{M,V}}) where {V,M} = M
+_ismutable(v::MinimalVec) = _ismutable(typeof(v))
+
+scalartype(::Type{<:MinimalVec{M,V}}) where {M,V} = scalartype(V)
+
+function zerovector(v::MinimalVec, S::Type{<:Number})
+    return MinimalVec{_ismutable(v)}(zerovector(v.vec, S))
+end
+function zerovector!(v::MinimalMVec{V}) where {V}
+    zerovector!(v.vec)
+    return v
+end
+zerovector!!(v::MinimalVec) = _ismutable(v) ? zerovector!(v) : zerovector(v)
+
+function scale(v::MinimalVec, α::Number)
+    return MinimalVec{_ismutable(v)}(scale(v.vec, α))
+end
+function scale!(v::MinimalMVec{V}, α::Number) where {V}
+    scale!(v.vec, α)
+    return v
+end
+function scale!!(v::MinimalVec, α::Number)
+    if _ismutable(v)
+        w = scale!!(v.vec, α)
+        return w === v.vec ? v : MinimalMVec(w)
+    else
+        return scale(v, α)
+    end
+end
+function scale!(w::MinimalMVec{V}, v::MinimalMVec{W}, α::Number) where {V,W}
+    scale!(w.vec, v.vec, α)
+    return w
+end
+function scale!!(w::MinimalVec, v::MinimalVec, α::Number)
+    if _ismutable(w)
+        wvec = scale!!(w.vec, v.vec, α)
+        return wvec === w.vec ? w : MinimalMVec(wvec)
+    else
+        return scale(v, α * one(scalartype(w)))
+    end
+end
+
+function add(y::MinimalVec, x::MinimalVec, α::Number, β::Number)
+    return MinimalVec{_ismutable(y)}(add(y.vec, x.vec, α, β))
+end
+function add!(y::MinimalMVec{W}, x::MinimalMVec{V}, α::Number, β::Number) where {W,V}
+    add!(y.vec, x.vec, α, β)
+    return y
+end
+function add!!(y::MinimalVec, x::MinimalVec, α::Number, β::Number)
+    if _ismutable(y)
+        yvec = add!!(y.vec, x.vec, α, β)
+        return yvec === y.vec ? y : MinimalMVec(yvec)
+    else
+        return add(y, x, α, β)
+    end
+end
+
+inner(x::MinimalVec, y::MinimalVec) = inner(x.vec, y.vec)
+LinearAlgebra.norm(x::MinimalVec) = LinearAlgebra.norm(x.vec)

test/minimalmvec.jl

@@ -0,0 +1,160 @@
+module Minimal
+using VectorInterface
+using VectorInterface: MinimalVec, MinimalMVec
+using Test
+using TestExtras
+
+deepcollect(x::MinimalVec) = x.vec
+deepcollect(x::Number) = x
+
+x = MinimalMVec(randn(3))
+y = MinimalMVec(randn(3))
+
+@testset "scalartype" begin
+    s = @constinferred scalartype(x)
+    @test s == Float64
+end
+
+@testset "zerovector" begin
+    z = @constinferred zerovector(x)
+    @test all(iszero, deepcollect(z))
+    @test all(deepcollect(z) .=== zero(scalartype(x)))
+    z1 = @constinferred zerovector!!(deepcopy(x))
+    @test all(deepcollect(z1) .=== zero(scalartype(x)))
+    z2 = @constinferred zerovector!(deepcopy(x))
+    @test all(deepcollect(z2) .=== zero(scalartype(x)))
+
+    z3 = @constinferred zerovector(x, ComplexF64)
+    @test all(deepcollect(z3) .=== zero(ComplexF64))
+    z4 = @constinferred zerovector!!(deepcopy(x), ComplexF64)
+    @test all(deepcollect(z4) .=== zero(ComplexF64))
+    @test_throws MethodError zerovector!(deepcopy(x), ComplexF64)
+end
+
+@testset "scale" begin
+    α = randn()
+    z = @constinferred scale(x, α)
+    @test all(deepcollect(z) .== α .* deepcollect(x))
+
+    z2 = @constinferred scale!!(deepcopy(x), α)
+    @test deepcollect(z2) ≈ (α .* deepcollect(x))
+    xcopy = deepcopy(x)
+    z2 = @constinferred scale!!(deepcopy(y), xcopy, α)
+    @test deepcollect(z2) ≈ (α .* deepcollect(x))
+    @test all(deepcollect(xcopy) .== deepcollect(x))
+
+    z3 = @constinferred scale!(deepcopy(x), α)
+    @test deepcollect(z3) ≈ (α .* deepcollect(x))
+    xcopy = deepcopy(x)
+    z3 = @constinferred scale!(zerovector(x), xcopy, α)
+    @test deepcollect(z3) ≈ (α .* deepcollect(x))
+    @test all(deepcollect(xcopy) .== deepcollect(x))
+
+    α = randn(ComplexF64)
+    z4 = @constinferred scale(x, α)
+    @test deepcollect(z4) ≈ (α .* deepcollect(x))
+    xcopy = deepcopy(x)
+    z5 = @constinferred scale!!(xcopy, α)
+    @test deepcollect(z5) ≈ (α .* deepcollect(x))
+    @test all(deepcollect(xcopy) .== deepcollect(x))
+    @test_throws InexactError scale!(xcopy, α)
+
+    α = randn(ComplexF64)
+    xcopy = deepcopy(x)
+    z6 = @constinferred scale!!(zerovector(x), xcopy, α)
+    @test deepcollect(z6) ≈ (α .* deepcollect(x))
+    @test all(deepcollect(xcopy) .== deepcollect(x))
+    @test_throws InexactError scale!(zerovector(x), xcopy, α)
+
+    xz = @constinferred zerovector(x, ComplexF64)
+    z6 = @constinferred scale!!(xz, xcopy, α)
+    @test deepcollect(z6) ≈ (α .* deepcollect(x))
+    @test all(deepcollect(xcopy) .== deepcollect(x))
+
+    z7 = @constinferred scale!(xz, xcopy, α)
+    @test deepcollect(z7) ≈ (α .* deepcollect(x))
+    @test all(deepcollect(xcopy) .== deepcollect(x))
+
+    ycomplex = zerovector(y, ComplexF64)
+    α = randn(Float64)
+    xcopy = deepcopy(x)
+    z8 = @constinferred scale!!(ycomplex, xcopy, α)
+    @test z8 === ycomplex
+    @test all(deepcollect(z8) .== α .* deepcollect(xcopy))
+end
+
+@testset "add" begin
+    α, β = randn(2)
+    z = @constinferred add(y, x)
+    @test all(deepcollect(z) .== deepcollect(x) .+ deepcollect(y))
+    z = @constinferred add(y, x, α)
+    # for some reason, on some Julia versions on some platforms, but only in test mode
+    # there is a small floating point discrepancy, which makes the following test fail:
+    # @test all(deepcollect(z) .== muladd.(deepcollect(x), α, deepcollect(y)))
+    @test deepcollect(z) ≈ muladd.(deepcollect(x), α, deepcollect(y))
+    z = @constinferred add(y, x, α, β)
+    # for some reason, on some Julia versions on some platforms, but only in test mode
+    # there is a small floating point discrepancy, which makes the following test fail:
+    # @test all(deepcollect(z) .== muladd.(deepcollect(x), α, deepcollect(y) .* β))
+    @test deepcollect(z) ≈ muladd.(deepcollect(x), α, deepcollect(y) .* β)
+
+    α, β = randn(2)
+    xcopy = deepcopy(x)
+    z2 = @constinferred add!!(deepcopy(y), xcopy)
+    @test deepcollect(z2) ≈ (deepcollect(x) .+ deepcollect(y))
+    @test all(deepcollect(xcopy) .== deepcollect(x))
+    z2 = @constinferred add!!(deepcopy(y), xcopy, α)
+    @test deepcollect(z2) ≈ (muladd.(deepcollect(x), α, deepcollect(y)))
+    @test all(deepcollect(xcopy) .== deepcollect(x))
+    z2 = @constinferred add!!(deepcopy(y), xcopy, α, β)
+    @test deepcollect(z2) ≈ (muladd.(deepcollect(x), α, deepcollect(y) .* β))
+    @test all(deepcollect(xcopy) .== deepcollect(x))
+
+    α, β = randn(2)
+    z3 = @constinferred add!(deepcopy(y), xcopy)
+    @test deepcollect(z3) ≈ (deepcollect(y) .+ deepcollect(x))
+    @test all(deepcollect(xcopy) .== deepcollect(x))
+    z3 = @constinferred add!(deepcopy(y), xcopy, α)
+    @test all(deepcollect(xcopy) .== deepcollect(x))
+    @test deepcollect(z3) ≈ (muladd.(deepcollect(x), α, deepcollect(y)))
+    z3 = @constinferred add!(deepcopy(y), xcopy, α, β)
+    @test deepcollect(z3) ≈ (muladd.(deepcollect(x), α, deepcollect(y) .* β))
+    @test all(deepcollect(xcopy) .== deepcollect(x))
+
+    α, β = randn(ComplexF64, 2)
+    z4 = @constinferred add(y, x, α)
+    @test deepcollect(z4) ≈ (muladd.(deepcollect(x), α, deepcollect(y)))
+    z4 = @constinferred add(y, x, α, β)
+    @test deepcollect(z4) ≈ (muladd.(deepcollect(x), α, deepcollect(y) .* β))
+
+    α, β = randn(ComplexF64, 2)
+    z5 = @constinferred add!!(deepcopy(y), xcopy, α)
+    @test deepcollect(z5) ≈ (muladd.(deepcollect(x), α, deepcollect(y)))
+    @test all(deepcollect(xcopy) .== deepcollect(x))
+    z5 = @constinferred add!!(deepcopy(y), xcopy, α, β)
+    @test deepcollect(z5) ≈ (muladd.(deepcollect(x), α, deepcollect(y) .* β))
+    @test all(deepcollect(xcopy) .== deepcollect(x))
+
+    α, β = randn(ComplexF64, 2)
+    z5 = @constinferred add!!(deepcopy(y), xcopy, α)
+    @test deepcollect(z5) ≈ (muladd.(deepcollect(x), α, deepcollect(y)))
+    @test all(deepcollect(xcopy) .== deepcollect(x))
+    z5 = @constinferred add!!(deepcopy(y), xcopy, α, β)
+    @test deepcollect(z5) ≈ (muladd.(deepcollect(x), α, deepcollect(y) .* β))
+    @test all(deepcollect(xcopy) .== deepcollect(x))
+
+    α, β = randn(ComplexF64, 2)
+    @test_throws InexactError add!(deepcopy(y), xcopy, α)
+    @test_throws InexactError add!(deepcopy(y), xcopy, α, β)
+end
+
+@testset "inner" begin
+    s = @constinferred inner(x, y)
+    @test s ≈ inner(deepcollect(x), deepcollect(y))
+
+    α, β = randn(ComplexF64, 2)
+    s2 = @constinferred inner(scale(x, α), scale(y, β))
+    @test s2 ≈ inner(α * deepcollect(x), β * deepcollect(y))
+end
+
+end