making std::complex a POD

From: Toon Knapen (toon.knapen_at_[hidden])
Date: 2002-03-11 17:52:07

The std::complex discussion started out with discussing if std::complex 
should be a POD (Dave had some great discussion with an EDG guy but the 
archives are not active anymore, move to mailman I guess).

On the other hand, the discussion of creating a vector of complexes is 
getting more and more complex. It has already been commented that scientists 
fear C++ because of its complexity, so we should focus on simplicity as well 
as functionality and performance.

So getting back to the basic question: why can't complex be a POD. If we 
relax the POD requirements, changing the 'no constructor' requirement to 
'must have none or at least a no-op default constructor'. If the developer 
provides  a constructor that takes no argument, the developer indicates that 
no info is needed to get the object in a valid state (and it's up to the 
developer if this constructor indeed puts the object in a valid state).  If 
the constructor body is empty (it's a no-op), the compiler knows he can 
optimise the constructor away. 
But I guess I'm forgetting something essential, isn't it ?

toon