Int(::Dual) and Integer(::Dual)

[dlfivefifty]

This adds convert routines for integers. I need this to fix the following bug:

julia> length(UnitRange(Dual(1.0), Dual(2.0)))
ERROR: MethodError: no method matching Integer(::Dual{Nothing, Float64, 0})
Closest candidates are:
  (::Type{T})(::T) where T<:Number at boot.jl:760
  (::Type{T})(::AbstractChar) where T<:Union{AbstractChar, Number} at char.jl:50
  (::Type{T})(::Base.TwicePrecision) where T<:Number at twiceprecision.jl:243
  ...
Stacktrace:
 [1] unsafe_length(r::UnitRange{Dual{Nothing, Float64, 0}})
   @ Base ./range.jl:559
 [2] length(r::UnitRange{Dual{Nothing, Float64, 0}})
   @ Base ./range.jl:561
 [3] top-level scope
   @ REPL[4]:1

[codecov-io]

Codecov Report

Merging #508 (f89181b) into master (62512aa) will increase coverage by 0.02%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #508      +/-   ##
==========================================
+ Coverage   88.94%   88.96%   +0.02%     
==========================================
  Files          10       10              
  Lines         841      843       +2     
==========================================
+ Hits          748      750       +2     
  Misses         93       93              

Impacted Files	Coverage Δ
src/dual.jl	79.32% <100.00%> (+0.14%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 62512aa...f89181b. Read the comment docs.

[KristofferC]

I think this should only work if the dual part is zero (cf the behavior of complex numbers).

[dlfivefifty]

I think I agree:

julia> Integer(nextfloat(1.0))
ERROR: InexactError: Int64(1.0000000000000002)
Stacktrace:
 [1] Int64
   @ ./float.jl:723 [inlined]
 [2] Integer(x::Float64)
   @ Core ./boot.jl:786
 [3] top-level scope
   @ REPL[1]:1

But in that case I should also overload unsafe_length(::UnitRange{<:Dual}) so that

julia> UnitRange(Dual(1,1), Dual(2,0)) |> collect
2-element Vector{Dual{Nothing, Int64, 1}}:
 Dual{Nothing}(1,1)
 Dual{Nothing}(2,1)

works.

[dlfivefifty]

Ha! We're in luck, because partials(step(r)) == partials(first(r)) the default code for unsafe_length(::UnitRange) just works as is..

[ranocha]

Bump. This looks like a useful addition to me 👍

[dlfivefifty]

Can this be merged? Just ran into it, again

[dlfivefifty]

The test failure looks unrelated...

[KristofferC]

I agree that the test failures don't look directly related to this but when I look at the CI run for earlier commits on master (https://github.com/JuliaDiff/ForwardDiff.jl/commits/master) they all seem green so it is a bit surprising why this would suddenly start happening. And it happens for all Julia versions as well.

[KristofferC]

CI error also happens in #527. It seems to be something with the StaticArray tests so maybe it is a new version of StaticArrays that caused it?

The last PR on master ran with StaticArrays v1.1.0 and this one runs with StaticArrays v1.2.4,

[KristofferC]

I bisected and JuliaArrays/StaticArrays.jl#907 is when the ForwardDiff test failure was introduced.

[dlfivefifty]

Note the values depend on the sample in x = rand(3,3). If I do Random.seed!(0) I get the slightly different result:

julia> @test DiffResults.value(result1) == DiffResults.value(result)
Test Failed at REPL[39]:1
  Expression: DiffResults.value(result1) == DiffResults.value(result)
   Evaluated: 0.0 == 1.297445353195765e-6

[dlfivefifty]

I'm debugging this now. FYI I believe vector_mode_dual_eval should be renamed vector_mode_dual_eval!

ForwardDiff.jl/src/apiutils.jl

Line 40 in 909976d

function vector_mode_dual_eval(f!::F, y, x, cfg::JacobianConfig) where {F}

EDIT: The StaticArrays.jl#907 Issue already has a minimum working example.

[codecov-commenter]

Codecov Report

Merging #508 (74e59ac) into master (909976d) will increase coverage by 0.21%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #508      +/-   ##
==========================================
+ Coverage   84.62%   84.83%   +0.21%     
==========================================
  Files           9        9              
  Lines         826      831       +5     
==========================================
+ Hits          699      705       +6     
+ Misses        127      126       -1     

Impacted Files	Coverage Δ
src/dual.jl	72.72% <100.00%> (+0.84%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 909976d...74e59ac. Read the comment docs.