Re: [boost] [mp_math] a multi precision signed integer library

From: Maarten Kronenburg (M.Kronenburg_at_[hidden])
Date: 2008-04-01 09:36:43

"Steven Watanabe"  wrote in message
> AMDG
>
> Maarten Kronenburg wrote:
> > In my opinion an unsigned integer is a, works like a and can be
substituted
> > as an integer, and so is a modular integer, and as far as I'm concerned
also
> > an allocated integer. This is why I choose for derivation, and then with
the
> > base pointer to derived object (as quoted above) to runtime
polymorphism.
> >
>
> I strongly disagree.
>
> 1)  An unsigned integer only works like an integer for non-mutating
> operations.  If clients take parameters of const integer&, sure,
> unsigned_integer
> will work fine in this context.  Treating non-const integers
> polymorphically is a very bad
> idea, though.  Code that does so has to be very carefully designed to
> make sure that
> it can handle all the combinations.  If you have to use inheritance,
> make both integer and
> unsigned_integer inherit from a common base.
>

In  my design the integer class is the value class, which can also be used
as data member of other (possibly template) classes, and the
unsigned_integer class is polymorphically derived so that a base pointer to
a derived object still works. So in my opinion this design has the best of
both worlds.
But there are of course many other ways to do it with other pros and cons.
But in my opinion these days templates are used in designs by default, and
that's not a good thing.
Users cannot derived from the STL string because it has no virtual
destructor, but for that class that scenario was not so important. For the
class integer I think we cannot afford to overlook this. Don't look at me if
somewhere in the future this will bubble up.

> 2) I would expect an integer class to be a value object. virtual
> functions don't play
> well with value objects, because to prevent slicing, everything has to
> be held by
> pointer.
>

As mentioned the integer base class can be used as value class, the only
price is the unused vtable.

> 3) unsigned_integer + unsigned_integer should return an
> unsigned_integer.  There
> is no way for the result to be negative.  This is even more glaring for
> modular_integer
>

No, what about unsigned_integer - unsigned_integer? That can be negative!

> > And in my opinion an integer is not a container, so I see no reason to
use
> > templates.
> >
>
> ???
>