Test failure on 1.0 appears to be caused by this list not being empty:

julia> using Test, EllipsisNotation; detect_ambiguities(EllipsisNotation, Base, Core)
2-element Array{Tuple{Method,Method},1}:
 (getindex(A::AbstractArray, ::EllipsisNotation.Ellipsis) in EllipsisNotation at /Users/me/.julia/dev/EllipsisNotation/src/EllipsisNotation.jl:25, getindex(v::JSON.Parser.PushVector, i) in JSON.Parser at /Users/me/.julia/packages/JSON/d89fA/src/pushvector.jl:15)                        
 (getindex(A::AbstractArray, ::EllipsisNotation.Ellipsis) in EllipsisNotation at /Users/me/.julia/dev/EllipsisNotation/src/EllipsisNotation.jl:25, getindex(a::GenericArray, i...) in Test at /Users/osx/buildbot/slave/package_osx64/build/usr/share/julia/stdlib/v1.0/Test/src/Test.jl:1553)

I think the offending line is this:

Base.getindex(A::AbstractArray, ::Ellipsis) = A

This list is not empty on 1.3 either, but that didn’t cause an error. I’m not entirely sure why this sees the package JSON at all.

JSON was being loaded by BenchmarkTools in my statup. On 1.0 the remaining ambiguity warning is from getindex(a::GenericArray, i...) defined by Test, which was must have been removed from later versions. I’ve simply disabled that test on 1.0 now.

This looks fine for me, just bump the version number.

Great! Is a minor increment enough, or should it go to 0.4?

Definitely 0.4

The downside of this change is that EllipsisNotation.jl now depends on ArrayInterface.jl:

https://github.com/ChrisRackauckas/EllipsisNotation.jl/blob/master/Project.toml

So this adds around 5000 lines of code, not 20. If this package wants to stay lightweight, then we should consider reverting this.

Yeah, that's not great. Let's deal with the issue over in SciML/EllipsisNotation.jl#25

This PR is reverted in #83, btw.