Broadcast max is not supported

[weymouth]

This was a bad MWE! Nevermind.

Perhaps I'm doing something silly, but it looks like broadcasted max is not supported as a differentiable function?

> using ForwardDiff
> f(x::Vector) = max.(x,0.);
> g = x-> ForwardDiff.gradient(f,x);
> x = rand(3)
0.326712
0.362656
0.832167
> f(x)
0.326712
0.362656
0.832167
> g(x)
MethodError: no method matching extract_gradient!(::Type{ForwardDiff.Tag{typeof(Main.workspace397.f),Float64}}, ::Array{Array{ForwardDiff.Dual{ForwardDiff.Tag{typeof(Main.workspace397.f),Float64},Float64,3},1},1}, ::Array{ForwardDiff.Dual{ForwardDiff.Tag{typeof(Main.workspace397.f),Float64},Float64,3},1})

Closest candidates are:

extract_gradient!(::Type{T}, ::AbstractArray, !Matched::ForwardDiff.Dual) where T at /Users/weymouth/.julia/packages/ForwardDiff/qTmqf/src/gradient.jl:79

extract_gradient!(::Type{T}, ::AbstractArray, !Matched::Real) where T at /Users/weymouth/.julia/packages/ForwardDiff/qTmqf/src/gradient.jl:78

extract_gradient!(::Type{T}, !Matched::DiffResults.DiffResult, !Matched::ForwardDiff.Dual) where T at /Users/weymouth/.julia/packages/ForwardDiff/qTmqf/src/gradient.jl:72

...

vector_mode_gradient(::typeof(Main.workspace397.f), ::Array{Float64,1}, ::ForwardDiff.GradientConfig{ForwardDiff.Tag{typeof(Main.workspace397.f),Float64},Float64,3,Array{ForwardDiff.Dual{ForwardDiff.Tag{typeof(Main.workspace397.f),Float64},Float64,3},1}})@gradient.jl:101
gradient(::Function, ::Array{Float64,1}, ::ForwardDiff.GradientConfig{ForwardDiff.Tag{typeof(Main.workspace397.f),Float64},Float64,3,Array{ForwardDiff.Dual{ForwardDiff.Tag{typeof(Main.workspace397.f),Float64},Float64,3},1}}, ::Val{true})@gradient.jl:19
gradient(::Function, ::Array{Float64,1}, ::ForwardDiff.GradientConfig{ForwardDiff.Tag{typeof(Main.workspace397.f),Float64},Float64,3,Array{ForwardDiff.Dual{ForwardDiff.Tag{typeof(Main.workspace397.f),Float64},Float64,3},1}})@gradient.jl:17
(::Main.workspace397.var"#1#2")(::Array{Float64,1})@Other: 1
top-level scope@Local: 1