zero() and one() for incomplete and abstract types

[ivarne]

The current state of zero and one seems unexpected for incomplete and abstract types. I think all these cases should result in an error.

julia> zero(Real)
0
julia> typeof(ans)
Int64

julia> one(Real)
1
julia> typeof(ans)
Int64

julia> zero(Complex)
0 + 0im
julia> typeof(ans)
Complex{Int64} (constructor with 1 method)

julia> one(Complex)
1 + 0im
julia> typeof(ans)
Complex{Int64} (constructor with 1 method)

Some relevant discussion in #3728, but I'm opening a new one because it is marked as fixed.

Edited to add:
I also want errors for arrays of incomplete and abstract types, but i have not figured out the relationship between zero and zeros and one and onses. That might be a separate issue.

[johnmyleswhite]

+1 for making all of these errors.

[StefanKarpinski]

I agree that these should be errors for abstract types.

[ivarne]

@StefanKarpinski What about incomplete types like Complex{T<:Real}?

[StefanKarpinski]

It seems like anything that's not a complete concrete type is problematic since, while you may be able to pick some zero value that's of that type, it's always somewhat arbitrary since there will be other ones that could be picked.

[Keno]

would zero(::Any)=false make sense?

[StefanKarpinski]

Currently that only really makes sense for Number, not Any. I'm skeptical about one and zero for Any.

[jiahao]

zero(::Any)=false isn't sensible for arrays.

[goretkin]

If this issue hasn't been discussed elsewhere, could it get a breaking label and perhaps a 2.0 milestone?

This came up in this thread recently: https://discourse.julialang.org/t/why-isa-x-1-y-1-array-tuple-stuff-number-1-false/55777