Conversions to and from RationalFunction broken

[jishnub]

Julia nightly demands that convert(T, x) produce a result of type T. This is currently not the case with RationalPolynomials, where promote_rule at various places hard-codes Polynomial instead of using a proper promotion. As a consequence, the following is broken (and tests fail on Julia nightly in general):

julia> r = SparsePolynomial([1,2], :x)
(SparsePolynomial(1 + 2*x)

julia> rr = r // (r-1)
(1 + 2*x) // (2*x)

julia> oftype(rr, r)
ERROR: TypeError: in typeassert, expected RationalFunction{Int64, :x, SparsePolynomial{Int64, :x}}, got a value of type RationalFunction{Int64, :x, Polynomial{Int64, :x}}
Stacktrace:
 [1] oftype(x::RationalFunction{Int64, :x, SparsePolynomial{Int64, :x}}, y::SparsePolynomial{Int64, :x})
   @ Base ./essentials.jl:641
 [2] top-level scope
   @ REPL[12]:1

[jverzani]

Thanks for letting me know!

[jishnub]

Note that there are other instances like

Polynomials.jl/src/rational-functions/common.jl

Lines 52 to 64 in 263b965

	function Base.convert(::Type{PQ}, q::LaurentPolynomial) where {PQ <: AbstractRationalFunction}
	m = firstindex(q)
	if m >= 0
	p = convert(Polynomial, q)
	return convert(PQ, p)
	else
	z = variable(q)
	zᵐ = z^(-m)
	p = convert(Polynomial, zᵐ * q)
	return rational_function(PQ, p, convert(Polynomial, zᵐ))
	end
	end

where the convert explicitly uses Polynomial, which may not be the correct type

[jverzani]

Yes, that one was deliberate. Perhaps it should just be deprecated and the user should convert to a polynomial type without negative powers before using.

Thanks again! I really appreciate your time and effort to point these out.