what to do about one and zero ?

Currently, we have this:

julia> one(Diagonal(fill(1.0, 2)))    # 1
2×2 Array{Float64,2}:
 1.0  0.0
 0.0  1.0

julia> zero(Diagonal(fill(1.0, 2)))  # 2
2×2 Diagonal{Float64,Array{Float64,1}}:
 0.0   ⋅ 
  ⋅   0.0

julia> one(Diagonal(Fill(1.0, 2)))  # 3
2×2 Array{Float64,2}:
 1.0  0.0
 0.0  1.0

julia> zero(Diagonal(Fill(1.0, 2)))  # 4
2×2 Diagonal{Float64,Array{Float64,1}}:
 0.0   ⋅ 
  ⋅   0.0

one and zero behave differently, but they probably should behave analogously.
The documentation for one says

If possible, one(x) returns a value of the same type as x,

And for zero

additive identity element for the type of x

This says to me that, of the above examples, only number 2 is correct.

No source file, test, or entry point is named. Start with the four Julia REPL examples and the documented contracts for one and zero; done requires an agreed behavior and corresponding implementation and tests, but the issue does not specify which outcome to choose.