Re: Collider and Mover code

[David Philippi]

On Saturday 25 January 2003 20:27, Gervase Lam wrote:
> > - templates parameters should be named by using typename instead of
>  I've changed it from Collider to ColliderFunctor.

Nice touch but I meant something different :-)
  template<class Functor> --> template<typename Functor>
was what I had in mind. I hate the usage of class most in expressions like    
  template<template <class> foo>
Using typenam here makes things much easier to read in my opinion.

> I've made the change.  However, it would be "interesting" if pingu_height
> were to change to a double in the far future, for example.  But I think
> the likelihood of that happening is practically zero.

As long as we keep using window systems using pixels I don't think this will 
happen. Although, just yesterday I've read about a project which tries to 
replace X with a completely new system based on vector graphics instead of 
pixels as the default... ;-)

> "Passing function objects by value instead of by reference has the
> advantage that you can pass constant and temporary expressions."

Note that they are saying by reference. You may very well pass constant or 
temporary objects as const &.

> STL book.  This needs to be done using references so that seq keeps its
> value.

Notice that they are here using a non-const reference so you can't use const 
values or temporaries anymore.

> In this case, mover is taking in falling_down_pingu and falling_up_pingu
> by reference.  This means that in the future, (non-const) colliders could
> receive feedback information.

Mover will never take any value by reference unless you change the signature 
to Collider & collider. If you want to allow changeable colliders as well as 
temporaries you'll need to methods:

virtual void update(const Vector& move, const Collider& collider) const =0
virtual void update(const Vector& move, Collider& collder) [ =0 ]

I guess that the non-const variant is less likely so it could be usefull to 
provide an default implementation which consists of an assert(false).

> Letting the person using linear_mover decide what the type of the
> parameter is going to be provides more flexibility.  The person is not
> forced make a value be passable by reference.  If required, the person can
> use a temporary.

That's what I suggested, but this can only be achieved by providing two 
methods.

>  I think people can see this more easily.  It is also flexible in that the
> person can ask for the Function Object to be passed by reference if
> needed, or not.  The person has a choice.

Why do you think the person should ask for the object to be passed by 
reference? It's declared as by value so it will never be something else.

> I've changed the counter i to an int.  I assume that the expression (i <
> move_length) would promote i to a float before doing the comparison.

Well, you're partially right. It would be promoted to an float for the 
comparison so using float is better but I think changing move_length to int 
does even more good. We are operating with integer steps and so we're going 
to loose the precision of float anyway - why not right at the beginning?

> I'll wait for the answers before posting the updates for this.  I also
> need to sort out a strange compilation problem.

Are those compiler errors or linker errors? They look more like linker errors, 
so you should make sure that the .o of upright_pingu_collider comes before 
bomber.o in the linking stage (or rather the colliders.a must be before the 
actions.a).

Bye David