Re: [boost] Re: std::complex design (Was: N-Dimensional array library

From: David Abrahams (david.abrahams_at_[hidden])
Date: 2002-03-07 10:39:22

----- Original Message -----
From: "Howard Hinnant" <hinnant_at_[hidden]>

> > I disagree. Not that there isn't a problem, but I think you've
> > misperceived its nature. The built-ins are not the problem: we have
a
> > way to create an uinitialized built-in, but I believe that
> > http://std.dkuug.dk/jtc1/sc22/wg21/docs/papers/1999/n1191.pdf
explains
> > quite nicely why double() /must/ initialize its result.
>
> Thanks for the link.  It is a very nice write up that I was not aware
> of.  And I was very interested to learn some of the history which I
was
> ignorant of.  But I think we'll just have to agree to disagree.  I
still
> think the wrong decision was made.

Just out of curiosity, what do you think the alternative could have
been?

    int f(double);        // 1
    int x = f(double());  // 2

In line 2 above, if double() produced an uninitialized value f(double())
would have produced undefined behavior. Would you have advocated that
double() produces an undefined-but-copyable value?

Maybe we should just make a magic library function which does that:

    std::uninitialized<double>()

would produce a value whose copies the compiler can optimize away.

> But it has been made, and I respect
> that.  There's no going back.  And since there is no going back, I see
> little reason to debate the decision with the benefit of 20/20
> hindsight.  Had I been in the position to contribute to this decision
> when it was made, it is quite likely I would have made the same
mistake.
>
> I believe the way forward is to create new containers and elements
that
> take advantage of the flexibility in C++ and do not zero initialize.
> It's not that much work.
>
> template <class T>
> struct cart_complex
> {
>       T real;
>       T imag;
>       cart_complex() {}
>       cart_complex(const T& r, const T& i = T())
>          : real(r), imag(i) {}
>       ...
> };
> ...
> cart_complex<double>* vec = new cart_complex<double>[1000];

This is close to what I've been advocating in that the knowledge about
skipping intialization is located in the class itself.

-Dave