should x // y such that iszero(x) && iszero(y) throw an ArgumentError or a DivideError?

[nsajko]

The fact that dividing zero by zero throws DivideError in some cases, but ArgumentError in other cases seems inconsistent:

julia> 0 // 0
ERROR: ArgumentError: invalid rational: zero(Int64)//zero(Int64)
Stacktrace:
 [1] __throw_rational_argerror_zero(T::Type)
   @ Base ./rational.jl:30
 [2] Rational
   @ ./rational.jl:32 [inlined]
 [3] Rational
   @ ./rational.jl:48 [inlined]
 [4] //(n::Int64, d::Int64)
   @ Base ./rational.jl:84
 [5] top-level scope
   @ REPL[15]:1

julia> (0 // 1) // 0
ERROR: DivideError: integer division error
Stacktrace:
 [1] div
   @ ./int.jl:297 [inlined]
 [2] div
   @ ./int.jl:229 [inlined]
 [3] divgcd(x::Int64, y::Int64)
   @ Base ./rational.jl:54
 [4] //(x::Rational{Int64}, y::Int64)
   @ Base ./rational.jl:87
 [5] top-level scope
   @ REPL[16]:1

FWIW:

julia> Base.infer_exception_type(//, Tuple{Int, Int})
Union{DivideError, ArgumentError, OverflowError}

julia> Base.infer_exception_type(//, Tuple{Int, Rational{Int}})
Union{DivideError, OverflowError}

julia> Base.infer_exception_type(//, Tuple{Rational{Int}, Int})
Union{DivideError, OverflowError}

julia> Base.infer_exception_type(//, Tuple{Rational{Int}, Rational{Int}})
Union{DivideError, OverflowError}

Should 0 // 0 be changed to throw DivideError? Should (0 // 1) // 0 be changed to throw ArgumentError?

The specific reason I'm interested in this inconsistency is that I'm implementing a type very similar to Rational in a package. So I need to decide what to throw there.

[nhz2]

Linking a related discussion in: #25702