various fixes for v0.7/1.0 (#340)

* various fixes for v0.7/1.0

* remove old depwarns that have hung around for more than a release cycle

* use ≈ instead of == for Dual auto-tests

Author: jrevels

REQUIRE

@@ -1,4 +1,4 @@
-julia 0.7-beta2
+julia 0.7-rc3
 StaticArrays 0.5.0
 DiffResults 0.0.1
 DiffRules 0.0.4

src/ForwardDiff.jl

@@ -1,5 +1,3 @@
-__precompile__()
-
 module ForwardDiff
 
 using DiffRules, DiffResults
@@ -20,7 +18,6 @@ include("derivative.jl")
 include("gradient.jl")
 include("jacobian.jl")
 include("hessian.jl")
-include("deprecated.jl")
 
 export DiffResults
 

src/deprecated.jl

@@ -254,10 +254,10 @@ end
 @inline Base.one(d::Dual) = one(typeof(d))
 @inline Base.one(::Type{Dual{T,V,N}}) where {T,V,N} = Dual{T}(one(V), zero(Partials{N,V}))
 
-@inline Base.rand(d::Dual) = rand(typeof(d))
-@inline Base.rand(::Type{Dual{T,V,N}}) where {T,V,N} = Dual{T}(rand(V), zero(Partials{N,V}))
-@inline Base.rand(rng::AbstractRNG, d::Dual) = rand(rng, typeof(d))
-@inline Base.rand(rng::AbstractRNG, ::Type{Dual{T,V,N}}) where {T,V,N} = Dual{T}(rand(rng, V), zero(Partials{N,V}))
+@inline Random.rand(d::Dual) = rand(typeof(d))
+@inline Random.rand(::Type{Dual{T,V,N}}) where {T,V,N} = Dual{T}(rand(V), zero(Partials{N,V}))
+@inline Random.rand(rng::AbstractRNG, d::Dual) = rand(rng, typeof(d))
+@inline Random.rand(rng::AbstractRNG, ::Type{Dual{T,V,N}}) where {T,V,N} = Dual{T}(rand(rng, V), zero(Partials{N,V}))
 
 # Predicates #
 #------------#

src/dual.jl

@@ -22,9 +22,8 @@ end
 
 @inline Base.@propagate_inbounds Base.getindex(partials::Partials, i::Int) = partials.values[i]
 
-Base.start(partials::Partials) = start(partials.values)
-Base.next(partials::Partials, i) = next(partials.values, i)
-Base.done(partials::Partials, i) = done(partials.values, i)
+Base.iterate(partials::Partials) = iterate(partials.values)
+Base.iterate(partials::Partials, i) = iterate(partials.values, i)
 
 Base.IndexStyle(::Type{<:Partials}) = IndexLinear()
 

src/partials.jl

@@ -9,7 +9,7 @@ using DiffTests
 
 include(joinpath(dirname(@__FILE__), "utils.jl"))
 
-srand(1)
+Random.seed!(1)
 
 ########################
 # test vs. Calculus.jl #

test/DeprecatedTest.jl

@@ -419,12 +419,12 @@ for N in (0,3), M in (0,4), V in (Int, Float32)
                     if isnan(actualdx)
                         @test isnan(partials(dx, 1))
                     else
-                        @test partials(dx, 1) == actualdx
+                        @test partials(dx, 1) ≈ actualdx
                     end
                     if isnan(actualdy)
                         @test isnan(partials(dy, 1))
                     else
-                        @test partials(dy, 1) == actualdy
+                        @test partials(dy, 1) ≈ actualdy
                     end
                 end
             end

test/DerivativeTest.jl

@@ -138,5 +138,4 @@ sresult3 = ForwardDiff.gradient!(sresult3, prod, sx, scfg)
 @test DiffResults.gradient(sresult2) == DiffResults.gradient(result)
 @test DiffResults.gradient(sresult3) == DiffResults.gradient(result)
 
-
 end # module

test/DualTest.jl

@@ -115,9 +115,11 @@ a = fill(1.0, N)
 jac0 = reshape(vcat([[fill(0.0, N*(i-1)); a; fill(0.0, N^2-N*i)] for i = 1:N]...), N^2, N)
 
 for op in (:.-, :.+, :./, :.*)
-    f = @eval x -> [$op(x[1], a); $op(x[2], a); $op(x[3], a); $op(x[4], a)]
-    jac = @eval ForwardDiff.jacobian(f, a)
-    @test isapprox(jac0, jac)
+    @eval begin
+        f = x -> [$op(x[1], a); $op(x[2], a); $op(x[3], a); $op(x[4], a)]
+        jac = ForwardDiff.jacobian(f, a)
+        @test isapprox(jac0, jac)
+    end
 end
 
 # NaNs #
@@ -133,8 +135,10 @@ h = ForwardDiff.hessian(y -> sum(hypot.(x, y)), y)
 @test h[1, 1] ≈ (x[1]^2) / (x[1]^2 + y[1]^2)^(3/2)
 @test h[2, 2] ≈ (x[2]^2) / (x[2]^2 + y[2]^2)^(3/2)
 @test h[3, 3] ≈ (x[3]^2) / (x[3]^2 + y[3]^2)^(3/2)
-for i in 1:3, j in 1:3
-    i != j && (@test h[i, j] ≈ 0.0)
+let i, j
+    for i in 1:3, j in 1:3
+        i != j && @test h[i, j] ≈ 0.0
+    end
 end
 
 ########
@@ -144,8 +148,8 @@ end
 # issue 267
 @noinline f267(z, x) = x[1]
 z267 = ([(1, (2), [(3, (4, 5, [1, 2, (3, (4, 5), [5])]), (5))])])
-let z = z267
-    g = x -> f267(z, x)
+let z = z267,
+    g = x -> f267(z, x),
     h = x -> g(x)
     @test ForwardDiff.hessian(h, [1.]) == fill(0.0, 1, 1)
 end

test/GradientTest.jl

@@ -36,10 +36,6 @@ for N in (0, 3), T in (Int, Float32, Float64)
         @test PARTIALS[i] == VALUES[i]
     end
 
-    @test start(PARTIALS) == start(VALUES)
-    @test N == 0 || (next(PARTIALS, start(PARTIALS)) == next(VALUES, start(VALUES)))
-    @test done(PARTIALS, start(PARTIALS)) == done(VALUES, start(VALUES))
-
     i = 1
     for p in PARTIALS
         @test p == VALUES[i]

test/MiscTest.jl

@@ -5,7 +5,7 @@ t = @elapsed include("PartialsTest.jl")
 println("done (took $t seconds).")
 
 println("Testing Dual...")
-# t = @elapsed include("DualTest.jl")
+t = @elapsed include("DualTest.jl")
 println("done (took $t seconds).")
 
 println("Testing derivative functionality...")
@@ -37,7 +37,3 @@ if Base.JLOptions().opt_level >= 3
     t = @elapsed include("SIMDTest.jl")
     println("done (took $t seconds).")
 end
-
-println("Testing deprecations...")
-t = @elapsed include("DeprecatedTest.jl")
-println("done (took $t seconds).")

test/PartialsTest.jl

@@ -5,7 +5,7 @@ using Random
 
 # seed RNG, thus making result inaccuracies deterministic
 # so we don't have to retune EPS for arbitrary inputs
-srand(1)
+Random.seed!(1)
 
 const XLEN = DEFAULT_CHUNK_THRESHOLD + 1
 const YLEN = div(DEFAULT_CHUNK_THRESHOLD, 2) + 1