BoundsError on edge case for empty input

[felipenoris]

I'm upgrading a few projects from Julia v0.6 to Juila 1.1 and I'm getting this interesting edge case error.

This used to work on Julia v0.6, where the gradient takes an empty input.
I think Julia v0.6 version is stuck on ForwardDiff v0.7.5.

julia> f(x) = 1.0
f (generic function with 1 method)

julia> g = x -> ForwardDiff.gradient(f, x)
(::#1) (generic function with 1 method)

julia> g([1.0])
1-element Array{Float64,1}:
 0.0

julia> g(zeros(Float64, 0))
0-element Array{Float64,1}

On Julia v1.1, I'm using ForwardDiff v0.10.3. And the same example triggers a BoundsError.

julia> f(x) = 1.0
f (generic function with 1 method)

julia> g = x -> ForwardDiff.gradient(f, x)
#3 (generic function with 1 method)

julia> g([1.0])
1-element Array{Float64,1}:
 0.0

julia> g(zeros(Float64, 0))
ERROR: BoundsError: attempt to access 12-element Array{ForwardDiff.Chunk,1} at index [0]
Stacktrace:
 [1] getindex at ./array.jl:729 [inlined]
 [2] ForwardDiff.Chunk(::Int64, ::Int64) at /home/felipenoris/.julia/packages/ForwardDiff/N0wMF/src/prelude.jl:17
 [3] Type at /home/felipenoris/.julia/packages/ForwardDiff/N0wMF/src/prelude.jl:22 [inlined] (repeats 2 times)
 [4] ForwardDiff.GradientConfig(::typeof(f), ::Array{Float64,1}) at /home/felipenoris/.julia/packages/ForwardDiff/N0wMF/src/config.jl:121
 [5] gradient(::Function, ::Array{Float64,1}) at /home/felipenoris/.julia/packages/ForwardDiff/N0wMF/src/gradient.jl:15
 [6] (::getfield(Main, Symbol("##3#4")))(::Array{Float64,1}) at ./REPL[6]:1
 [7] top-level scope at none:0

[KristofferC]

The bug is pretty obvious:

ForwardDiff.jl/src/prelude.jl

Line 17 in 958711d

N <= DEFAULT_CHUNK_THRESHOLD && return CHUNKS[N]

Needs to handle the case where N == 0.

[KristofferC]

Fixed