Non-deterministic spurious ambiguity warning during make (in subarrays2.jl)

[carlobaldassi]

I was doing some work in Base and compiling a bunch of times. Once, non-deterministically, these ambiguity warning showed up during make:

subarray2.jl
Warning: New definition 
    getindex could not show value of type Tuple at subarray2.jl:84
is ambiguous with: 
    getindex could not show value of type Tuple at subarray2.jl:83.
To fix, define 
    getindex could not show value of type Tuple
before the new definition.

I wasn't doing anything related to subarrays or to getindex, and I checked that the warning disappeared when forcing make to run again without actually changing anything. Actually, I tried to reproduce it without success for at least 20 times. So I don't know what's the value of reporting, perhaps it's just the result of corruption in some totally unrelated area, but maybe someone can use this as a hint and find the bug (I don't know if this is related to other ambiguity issues, but I haven't found anything about non-deterministic ones).

For completeness: the warning refers to the following two lines in subarray.jl:

getindex{T,N,P,IV}(V::SubArray{T,N,P,IV}, I::AbstractArray{Bool,N}) = copy(sub(V, find(I)))
getindex{T,N,P,IV}(V::SubArray{T,N,P,IV}, I::Union(Real, AbstractVector)...) = getindex(V, to_index(I)...)

It seems the warning is spurious, since these two definitions do not lead to any ambiguity, although it's not trivial to see that, and in fact I haven't managed to reproduce the warning at the REPL. Plus, the other getindex definitions in subarray.jl (particularly this one) would also help disambiguate anyway.

I saw this while I was based on commit 3cc1752.

[JeffBezanson]

I only plan to work on this via #8974, which already lists 11 ambiguity-related issues. But you might be right that this indicates corruption elsewhere while loading that file, so that should maybe be investigated.

[timholy]

I've seen this quite a number of times, but you're right that it's rare. I've had the same experience with non-deterministic warnings on loading the Images.jl package (JuliaImages/Images.jl#59).

[timholy]

Given that subarray2 doesn't exist anymore, and the changes in #16125 and #16220, let's close this.