Re: [boost] Re: safe_bool alternative?

From: Daryle Walker (darylew_at_[hidden])
Date: 2003-11-28 21:06:37

On 11/27/03 6:09 PM, "Joel de Guzman" <joel_at_[hidden]> wrote:

> David Abrahams <dave_at_[hidden]> wrote:
>> Daryle Walker <darylew_at_[hidden]> writes:
>> 
>>> All of the operator helpers need the final type to provide a core routine.
>>> AFAIK, Peter's idiom is to use a pointer-to-member(-function) as the core
>>> Boolean routine; it can't be provided by a library.
>> 
>> What am I missing?

That you two (Dave and Joel) are trying to break the letter of my message,
but are still conforming to its spirit.

Dave helped write the operator-helpers library, so he should know what I'm
talking about.  I guess my wording isn't accurate.

All the operator helpers provide some extra operators.  These extra
operators are defined in terms of other operators.  These "other" operators
are the "core routines" I am referring to.

For example, the helper for "operator /" needs the author of the final class
to provide an "operator /=".  Here, "operator /" is the extra operator and
"operator /=" is the core routine.  Helper classes can't be used for all the
operators, because the core routines that the operators call must come from
the author.

>> template <class T>
>> struct bool_testable
>> {
>>  private:
>>     struct impl { int x; };
>>     typedef int impl::*safe_bool;
>>  public:
>>     operator safe_bool() const
>>     {
>>         return !*static_cast<T const*>(this) ? 0 : &impl::x;
>>     }
>> };

As far as "Peter's idiom" is concerned, I meant that the author has to
provide the "operator SomeBuiltInSingularType" for the core Boolean
conversion routine, and a helper class can provide the related operators
like "operator !".  All you did was switch the core and derived routines!
Either way, one Boolean-related routine has to be provided by the author.

>> struct X : bool_testable<X>
>> {
>>     bool operator!() const { return true; }
>> };
>> 
>> 
>> struct Y : bool_testable<Y>
>> {
>>     bool operator!() const { return false; }
>> };
>> 
>> 
>> int main()
>> {
>>     X x;
>>     Y y;
>>     if (x) { return 1; }
>> 
>>     if (y)
>>         return 0;
>> 
>> //    if (x == true) return 1; // fails to compile
>> //    if (x == y) return 1;    // fails to compile
>> 
>>     return 0;
>> }
> 
> Indeed. A library solution is possible and better. The only thing
> I do not like with the solution above is the inheritance, which implies
> multi-inheritance when X or Y are already subclasses. Well, you can
> chain the inheritance, but... The solution that I posted a while ago
> doesn't require any inheritance at all.

-- 
Daryle Walker
Mac, Internet, and Video Game Junkie
darylew AT hotmail DOT com