t: taylor(sin(x+y),[x,y],0,5)
=> y+x-....
ratcoeff(t,x,1) => 0+...
but
ratcoeff(ratdisrep(t),x,1) => (y^4-12*y^2+24)/24
The ratdisrep case should be taken as the definition. If
we don't want to do that, it should at least give an
error, saying that it doesn't handle multivariate series.
It would be nicer, though, if ratcoeff took multiple
arguments, e.g.
ratcoeff(ratcoeff(t,x,1),y,0)
=> x+y
ratcoeff(t,[x,y],1) => x+y
ratcoeff(t,[x,y],2) => 0
ratcoeff(t,[x,y],3) =>
-(y^3+3*x*y^2+3*x^2*y+x^3)/6
but that's another matter....
