Further improve coverage

Author: tansongchen

src/codegen.jl

@@ -2,17 +2,21 @@ using ChainRulesCore
 using SymbolicUtils, SymbolicUtils.Code
 using SymbolicUtils: Pow
 
+@scalar_rule +(x::Any) true
+@scalar_rule -(x::Any) -1
+@scalar_rule deg2rad(x::Any) deg2rad(one(x))
+@scalar_rule rad2deg(x::Any) rad2deg(one(x))
 @scalar_rule asin(x::Any) inv(sqrt(1 - x^2))
 @scalar_rule acos(x::Any) inv(-sqrt(1 - x^2))
 @scalar_rule atan(x::Any) inv(1 + x^2)
 @scalar_rule log(x::Any) inv(x)
-@scalar_rule log10(x::Any) inv(log(10.) * x)
+@scalar_rule log10(x::Any) inv(log(10.0) * x)
 @scalar_rule log1p(x::Any) inv(x + 1)
-@scalar_rule log2(x::Any) inv(log(2.) * x)
+@scalar_rule log2(x::Any) inv(log(2.0) * x)
 
 dummy = (NoTangent(), 1)
 @syms t₁
-for func in (asin, acos, atan, log, log10, log1p, log2)
+for func in (+, -, deg2rad, rad2deg, asin, acos, atan, log, log10, log1p, log2)
     F = typeof(func)
     # base case
     @eval function (op::$F)(t::TaylorScalar{T, 2}) where {T}

src/primitive.jl

@@ -11,8 +11,6 @@ import Base: hypot, max, min
 
 ## Delegated
 
-@inline +(t::TaylorScalar) = t
-@inline -(t::TaylorScalar) = -1 * t
 @inline sqrt(t::TaylorScalar) = t^0.5
 @inline cbrt(t::TaylorScalar) = ^(t, 1 / 3)
 @inline inv(t::TaylorScalar) = 1 / t
@@ -160,6 +158,8 @@ end
     end
 end
 
+raise(::T, df::S, t::TaylorScalar{T}) where {S <: Number, T <: Number} = df * t
+
 @generated function raiseinv(f::T, df::TaylorScalar{T, M},
                              t::TaylorScalar{T, N}) where {T, M, N} # M + 1 == N
     ex = quote

src/scalar.jl

@@ -55,19 +55,8 @@ end
 @inline value(t::TaylorScalar) = t.value
 @inline extract_derivative(t::TaylorScalar, i::Integer) = t.value[i]
 
-@generated function zero(::Type{TaylorScalar{T, N}}) where {T, N}
-    quote
-        $(Expr(:meta, :inline))
-        TaylorScalar($(zeros(T, N)...))
-    end
-end
-@generated function one(::Type{TaylorScalar{T, N}}) where {T, N}
-    quote
-        $(Expr(:meta, :inline))
-        TaylorScalar(one(T), $(zeros(T, N - 1)...))
-    end
-end
-
+@inline zero(::Type{TaylorScalar{T, N}}) where {T, N} = TaylorScalar{T, N}(zero(T))
+@inline one(::Type{TaylorScalar{T, N}}) where {T, N} = TaylorScalar{T, N}(one(T))
 @inline zero(::TaylorScalar{T, N}) where {T, N} = zero(TaylorScalar{T, N})
 @inline one(::TaylorScalar{T, N}) where {T, N} = one(TaylorScalar{T, N})
 

test/primitive.jl

@@ -1,10 +1,25 @@
 using FiniteDifferences
 
+@testset "No derivative or linear" begin
+    some_number = 3.7
+    for f in (+, -, zero, one, adjoint, conj, deg2rad, rad2deg), order in (2, 4)
+        @test derivative(f, some_number, order) ≈ 0.0
+    end
+end
+
 @testset "Unary functions" begin
     some_number = 3.7
     for f in (exp, expm1, exp2, exp10, sin, cos, sqrt, cbrt, inv), order in (2, 4)
         fdm = central_fdm(12, order)
-        @test derivative(f, some_number, order) ≈ fdm(f, some_number) rtol=1e-6
+        @test derivative(f, some_number, order)≈fdm(f, some_number) rtol=1e-6
+    end
+end
+
+@testset "Codegen" begin
+    some_number = 0.6
+    for f in (log, ), order in (2, 4)
+        fdm = central_fdm(12, order, max_range=0.5)
+        @test derivative(f, some_number, order)≈fdm(f, some_number) rtol=1e-6
     end
 end
 
@@ -13,10 +28,10 @@ end
     for f in (*, /), order in (2, 4)
         fdm = central_fdm(12, order)
         closure = x -> exp(f(x, another_number))
-        @test derivative(closure, some_number, order) ≈ fdm(closure, some_number) rtol=1e-6
+        @test derivative(closure, some_number, order)≈fdm(closure, some_number) rtol=1e-6
     end
     for f in (x -> x^7, x -> x^another_number), order in (2, 4)
         fdm = central_fdm(12, order)
-        @test derivative(f, some_number, order) ≈ fdm(f, some_number) rtol=1e-6
+        @test derivative(f, some_number, order)≈fdm(f, some_number) rtol=1e-6
     end
 end