Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!mcsun!ukc!mucs!pjc
From: pjc@cs.man.ac.uk (Peter Crowther (CAG ra))
Newsgroups: comp.lang.c++
Subject: Re: PolyMorphism
Message-ID: <1167@m1.cs.man.ac.uk>
Date: 11 Apr 90 08:49:50 GMT
References: <1990Apr10.112351.24301@athena.mit.edu>
Sender: news@cs.man.ac.uk
Reply-To: pjc@cs.man.ac.uk (Peter Crowther (CAG ra))
Organization: Department of Computer Science, University of Manchester UK
Lines: 30

In article <1990Apr10.112351.24301@athena.mit.edu> mleblanc@athena.mit.edu (Marc LeBlanc) writes:
>>So the derived destructer won't be called if we construct the Derived
>>object by Base pointer. But isn't it more natural to call both? If I
>>misunderstood the C++, is there anyway you can call Derived class
>>destructer?
>
>I think this would qualify as an "undocumented feature."  If you
>declare the base destructor as virtual, you get the desired effect. (I
>tried it).  Shouldn't this be the default?

IMHO, no. Reason: I dislike languages that force me to write inefficient
code (it's what comes of writing a lot of stuff on a Sun-2 (that's *2*)
file server supporting about 10 clients). The default in C++ is to save
the space in each object that would otherwise be allocated to the
virtual function table pointer.

It also maintains compatibility with C structures if the default is not
to tag another pointer onto the end of each structure. That's also
important to me; I have interfaces from C++ to SunView, RPC, XDR, Ingres
and a bunch of C libraries. Several of these would probably screw up if
virtual destructors were used. This is a weaker argument; the
counterargument is 'Anyone trying to interface C++ and C should have
enough nouse to change the default when they need to'. I'm just lazy.

		- Peter
--
Peter Crowther, Dept. of Electrical Engineering, University of Manchester,
		Manchester M13 9PL, England.
Internet: pcrowther@r1.cs.man.ac.uk		Janet: pcrowther@uk.ac.man.cs.r1
ARPA: pcrowther%cs.man.ac.uk@nsfnet-relay.ac.uk OtherNets: No idea at all!