change the definition of fastcholesky for numbers and uniform scaling

src/FastCholesky.jl

@@ -32,13 +32,21 @@ function fastcholesky(input::AbstractMatrix)
     # The `PositiveFactorizations.default_δ` should small enough in majority of the cases
     return cholesky(PositiveFactorizations.Positive, Hermitian(input), tol = PositiveFactorizations.default_δ(input))
 end
+
+fastcholesky(input::Number) = cholesky(input)
 fastcholesky(input::Diagonal) = cholesky(input)
 fastcholesky(input::Hermitian) = cholesky(PositiveFactorizations.Positive, input)
-fastcholesky(input::Number) = input
-fastcholesky!(input::Number) = input
 
-fastcholesky(x::UniformScaling) = sqrt(x.λ) * I
-fastcholesky!(x::UniformScaling) = sqrt(x.λ) * I
+function fastcholesky(x::UniformScaling)
+    return error(
+        "`fastcholesky` is not defined for `UniformScaling`. The shape is not determined."
+    )
+end
+function fastcholesky!(x::UniformScaling)
+    return error(
+        "`fastcholesky!` is not defined for `UniformScaling`. The shape is not determined."
+    )
+end
 
 function fastcholesky(input::Matrix{<:BlasFloat})
     C = fastcholesky!(copy(input))

test/fastcholesky_tests.jl

@@ -63,8 +63,8 @@ end
         @test chollogdet(I) ≈ zero
         @test all(cholinv_logdet(one * I) .≈ (one * I, zero))
         @test_throws ArgumentError chollogdet(two * I)
-        @test fastcholesky(I) ≈ I
-        @test fastcholesky!(I) ≈ I
+        @test_throws ErrorException fastcholesky(I)
+        @test_throws ErrorException fastcholesky!(I)
     end
 end
 
@@ -73,7 +73,8 @@ end
 
     for Type in SupportedTypes
         let number = rand(Type)
-            @test fastcholesky(number) === number
+            @test size(fastcholesky(number).L) == (1, 1)
+            @test all(fastcholesky(number).L .≈ sqrt(number))
             @test cholinv(number) ≈ inv(number)
             @test cholsqrt(number) ≈ sqrt(number)
             @test chollogdet(number) ≈ logdet(number)