Fix v[i] = -0.0 and support reinterpret (#296)

* assign whenever v !== zero(eltype(sparsecollection))

* test -0.0

* allow reinterpret

* test reinterpret

Author: LilithHafner

src/abstractsparse.jl

@@ -75,13 +75,6 @@ issparse(S::AbstractSparseArray) = true
 
 indtype(S::AbstractSparseArray{<:Any,Ti}) where {Ti} = Ti
 
-function Base.reinterpret(::Type, A::AbstractSparseArray)
-    error("""
-          `reinterpret` on sparse arrays is discontinued.
-          Try reinterpreting the value itself instead.
-          """)
-end
-
 # The following two methods should be overloaded by concrete types to avoid
 # allocating the I = findall(...)
 _sparse_findnextnz(v::AbstractSparseArray, i) = (I = findall(_isnotzero, v); n = searchsortedfirst(I, i); n<=length(I) ? I[n] : nothing)

src/sparsematrix.jl

@@ -3048,7 +3048,7 @@ function _setindex_scalar!(A::AbstractSparseMatrixCSC{Tv,Ti}, _v, _i::Integer, _
     end
     # Column j does not contain entry A[i,j]. If v is nonzero, insert entry A[i,j] = v
     # and return. If to the contrary v is zero, then simply return.
-    if _isnotzero(v)
+    if v isa AbstractArray || v !== zero(eltype(A)) # stricter than iszero to support A[i] = -0.0
         nz = getcolptr(A)[size(A, 2)+1]
         # throw exception before state is partially modified
         !isbitstype(Ti) || nz < typemax(Ti) ||

src/sparsevector.jl

@@ -390,7 +390,7 @@ end
     if 1 <= k <= m && nzind[k] == i  # i found
         nzval[k] = v
     else  # i not found
-        if _isnotzero(v)
+        if v isa AbstractArray || v !== zero(eltype(x)) # stricter than iszero to support v[i] = -0.0
             insert!(nzind, k, i)
             insert!(nzval, k, v)
         end

test/issues.jl

@@ -628,6 +628,36 @@ end
     @test A[1,1] == 2
 end
 
+@testset "-0.0 (issue #294, pr #296)" begin
+    v = spzeros(1)
+    v[1] = -0.0
+    @test v[1] === -0.0
+
+    m = spzeros(1, 1)
+    m[1, 1] = -0.0
+    @test m[1, 1] === -0.0
+end
+
+@testset "reinterpret (issue #289, pr #296)" begin
+    s = spzeros(3)
+    r = reinterpret(Int64, s)
+    @test r == s
+
+    r[1] = Int64(12)
+    @test r[1] === Int64(12)
+    @test s[1] === reinterpret(Float64, Int64(12))
+    @test r != s
+
+    r[2] = Int64(0)
+    @test r[2] === Int64(0)
+    @test s[2] === 0.0
+
+    z = reinterpret(Int64, -0.0)
+    r[3] = z
+    @test r[3] === z
+    @test s[3] === -0.0
+end
+
 # As part of the migration of SparseArrays.jl into its own repo,
 # these tests have been moved from other files in julia tests to
 # the SparseArrays.jl repo
@@ -753,6 +783,6 @@ g12063() = f12063(0, 0, 0, 0, 0, 0, 0.0, spzeros(0,0), Int[])
     @test String(take!(io)) == "transpose(sparse([1, 2, 1, 2], [1, 1, 2, 2], [1, 3, 2, 4], 2, 2))"
 end
 
-end
+end # SparseTestsBase
 
 end # module

test/sparsevector.jl

@@ -463,7 +463,7 @@ end
         @test Vector(x1) == collect(x)
     end
 end
-@testset "vec/reinterpret/float/complex" begin
+@testset "vec/float/complex" begin
     a = SparseVector(8, [2, 5, 6], Int32[12, 35, 72])
     # vec
     @test vec(a) == a