This is proving impossible to work around.
Firstly, I can't seem to produce a minimal example of the behaviour I just
mentioned. If the promote_rule_exists function is placed in the Nemo module
itself it returns a different value to if I define the function at the
REPL. But if I place it in a module created in the REPL, it behaves as
expected.
So instead I tried the following:
julia> using Nemo
julia> promote_type(Nemo.Poly{Nemo.Residue{Nemo.fmpq_poly}}, Nemo.fmpq_poly)
Nemo.PolyElem{T}
julia> @which promote_type(Nemo.Poly{Nemo.Residue{Nemo.fmpq_poly}},
Nemo.fmpq_poly)
promote_type{T,S}(::Type{T}, ::Type{S}) at promotion.jl:121
Why is Julia defining a completely generic promote_rule for types it knows
nothing about, and giving a completely invalid type as the result?
I thought of just defining:
function promote_rule_exists{T1, T2}(::Type{T1}, ::Type{T2})
return promote_type(T1, T2) == T1
end
Since all the promote_rules we create are of this form. But then it returns
true for cases where promote_type(T1, T2) is not T1!! This seems to happen
in cases where promote_type returns a parameterised type rather than a
concrete type (and thus presumably is coming from a promote_rule defined by
Julia.
But it doesn't consistently return true in such a case, making it
impossible to work around the problem.
Basically it looks to me like there is a large amount of breakage in
current master. Either that or promote_type, promote_rule, == and "in" are
now all so counterintuitive that I don't understand how they work. Or I
suppose I could just be being very stupid about something.
Bill.
