On Sun, Nov 25, 2012 at 7:05 PM, Peter Geoghegan <peter@2ndquadrant.com> wrote:
> Well, you can make your class copy-constructable by providing a
> constructor (and a copy-assignment operator) whose only argument is,
> say, an int. In additional to that, you could potentially define a
> conversion operator, which will make the class implicitly cast back
> into an int. That is kind of a big distinction, because it doesn't
> have to go both ways, and in fact it usually doesn't - plenty of
> working C++ programmers don't know what a conversion operator is, but
> they could all tell you how to get this behaviour:
>
> MyClass foo = 5; // actually calls copy constructor - equivalent to
> MyClass foo(5);
> foo = 4; // This calls copy assignment operator
I remember this sort of thing un-fondly from my C++ days, but it
doesn't make me like our current behavior any better. As in C++, we
seem to have created a system where the only way to get even locally
sensible behavior is to throw large piles of hackery at the problem.
Getting the behavior you want globally cannot be obtained at any
price.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
