TITLE: compile-time structure decision

(Newsgroup: comp.lang.c++.moderated, 10 Jan 2000)

ALEXANDRESCU: "Andrei Alexandrescu"

>> By the way, now that the rant is on, structure variations are
>> easier to implement with the help of either private inheritance, 
>> inner classes with total specialization, and/or this little structure:
>>
>> template <bool, typename T, typename U>
>> struct Choice
>> {
>>     typedef T Type;
>> };
>>
>> template <typename T, typename U>
>> struct Choice<false, T, U>
>> {
>>     typedef U Type;
>> };

ABRAHAMS: Dave Abrahams <abrahams@mediaone.net>

> Care to elaborate? It's not clear at all how you intend to use this.

ALEXANDRESCU: "Andrei Alexandrescu" <andrewalex@hotmail.com>

Sorry. The intent of Choice is to choose a type or another depending on a
boolean parameter. This allows you to tweak the structure depending on a
compile-time expression, like this:

enum ThreadingModel { singleThreaded, multiThreadedClassLevel,
multiThreadedObjectLevel };

template <ThreadingModel threading>
class Something
{
    typedef Choice<threading == singleThreaded,
        Null,
        Choice<threading == multiThreadedClassLevel,
            ClassLock<Something>, Lock>::Type>::Type
    ThreadingController;

    ThreadingController lock_;
    ...
};

ALEXANDRESCU:

>> It just happens that tweaking behavior is a lot harder 
>> because of this curious little hole in the template system. 
>> I dream I could convince some influent people about this :o).

ABRAHAMS:

> When you say "hole" you're not referring to your Choice template
> above, are you? I presume you mean the problem originally raised...

ALEXANDRESCU: 

I referred to the hole being the lack of PTS at the level of member
functions.

[snip]

[ PTS = partial template specialization  --adc ]




_______________________________________________
cpptips mailing list
http://cpptips.hyperformix.com