Re: [Quantlib-dev] Curve interpolation

[Luigi B. <lui...@gm...>]

Hi Simon,

On Fri, 2008-02-15 at 09:57 +0000, Simon Ibbotson wrote:
> I'm a little confused about the QuantLib curve classes ForwardCurve,
> DiscountCurve and ZeroCurve. 
>  
> In the three class definitions we have the member variable
> mutable Interpolation interpolation_; 
> 
> [...] often the object allocated is an object of a derived class e.g.
> CubicSpline.
> 
> Now, I know pointers and references can be polymorphic. But in this
> case a derived class is being allocated to a base class instance... 
> 
> I know most information required for interpolation is contained within
> the Interpolation::impl_ object but I'm wondering whether:
> 
> a) my C++ knowledge is lacking and the base class instance (e.g.
> ForwardCurve::interpolation_) can be polymorphic somehow. Or...

No, you're right---it's not polymorphic. As you surmised, the
polymorphic behavior is delegated to the inner Interpolation::impl_
object, which is copied by pointer and continues to have the correct
type even when the containing Interpolation object is sliced.


> b) why a pointer to the derived Interpolation object isn't returned by
> the factory class (e.g. Cubic::interpolate) - to obviate the need for
> a polymorphic Interpolation::impl_ member variable?

I confess that it's a while since we coded it, so I might be wrong. But
I think it's the same reason why we coded, say, the Calendar and
DayCounter classes in the same way---to avoid writing the shared_ptr
part. See <http://quantlib.org/quep/quep001.html> for the full rationale
(the code samples are outdated, but the reasoning still holds.)

Luigi


-- 

Olmstead's Law: 
After all is said and done, a hell of a lot more is said 
than done.