Re: std::complex design (Was: N-Dimensional array library submission...)

From: rwgk (rwgk_at_[hidden])
Date: 2002-03-06 14:31:44

In light of the upcoming Curacao meeting David Abrahams and
myself are discussion a proposal for the revision of
std::complex. Attached is one facet from our discussion. We
would like to hear comments if there are different
opinions.

David Abrahams wrote:
> I need you to explain why we need a complex<> with a
> predictable POD-like layout and a way to avoid
> initializing the members (anything else important about
> complex<>?)

Regarding David's last point:

It seems to me that this is more a question of how
containers work rather than one of the design of
std::complex<T>.

E.g.:

  std::vector<T>(1000000); // elements are default constructed -> T()

B. Stroustrup writes (section 6.2.8 of The C++ Prog. Lang.):

"The value of an explicit use of the constructor for a
built-in type is 0 converted to that type."

Do we want std::complex<T> to behave differently than
built-in types?

Here is a more general solution that will work for any type
that has a trivial destructor:

From files misc.h and shared_plain.h at:
http://cvs.sourceforge.net/cgi-
bin/viewcvs.cgi/cctbx/cctbx/cctbx/array_family/?
only_with_tag=array_family_integration

  struct no_initialization_flag {};

  template <typename ElementType>
  class shared_plain {
      // ...
      shared_plain(const size_type& sz, no_initialization_flag)
        : m_is_weak_ref(false),
          m_handle(new handle_type(sz * element_size()))
      {
        CCTBX_ARRAY_FAMILY_STATIC_ASSERT_HAS_TRIVIAL_DESTRUCTOR
        m_handle->size = m_handle->capacity;
      }
      // ...
  };

IMO both std::vector and std::valarray should have a
similar additional constructor.
This suggestion requires that type traits are part of the
standard. Is this likely to happen any time soon?

Thanks,
        Ralf