> And plotting implicitly starts from a fixed number of
> samples, so we could potentially miss poles if the pole is
> narrow enough and the sampling coarse enough. Don't think we
> could ever be smart enough to find a pole like that.
The right way to think of plotting is *not* evaluation at points, but
evaluation on intervals. With intervals, narrow poles and sin(1/x) and
all that "just work". But of course we need to implement intervals
first....
-s