Re: [boost] Re: "invariants" class; Version 1.01

From: Daniel Wallin (dalwan01_at_[hidden])
Date: 2004-01-08 04:40:29

Dan W. wrote:
> I've simplified the code, following from Daniel Wallin's comments,
> 
>   Version 1.01 -- Jan 8, 2004
>     * Put this documentation in separate text file (was comments
>      in invariants.hpp;  and added use example below.
>     * Got rid of enabled_ state, and first_call/last_call, and
>      changed to using private inheritance of invariants class.
> 
> I have NOT moved away from inheritance, however, as explained in 
> invariants.txt under 'Rationale':
> 
> ------------------------------
>   It's been suggested to me that introducing inheritance and
>   virtual functions via inheritance in debug mode might not be
>   good ideas. Note however that,
>   (A) as of version 1.01 there is no longer size augmentation
>      via inheritance, since the only data member was removed.

Yes there is, there's still a virtual member function.

>   (B) also as of version 1.01 private inheritance is used.
>   (C) class invariants is templeted per derived, and therefore
>      does not conduce to unwanted sibling relationships
>   (D) the only virtual function inherited from invariants class
>      is check_invariants(), which is pure virtual in the base,
>      and no longer virtual in derived. This AFAIK should not
>      cause the compiler to generate an otherwise absent virtual
>      table.
>   (E) Using inheritance allows one to use succint class member
>      names in the invariant assertions, rather than have to
>      dereference a pointer, or reference via '.'.

 From my implementation:

    struct my_type
    {
    private:
        friend class invariant_access;
        void invariant() const
        {
            ...
        }
    };

I don't see how it's related to inheritance. Or did you mean something
else?

-- 
Daniel Wallin