Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!think.com!zaphod.mps.ohio-state.edu!wuarchive!uunet!sbi!pivot-sts!swampthing!wfl
From: wfl@swampthing.sbi.com (W. Linke CPE)
Newsgroups: comp.lang.c++
Subject: Pure virtual destructors are illegal?
Keywords: virtual destructor
Message-ID: <183@swampthing.sbi.com>
Date: 31 May 91 23:50:54 GMT
Organization: Salomon Brothers - NJ
Lines: 44


It seems that you can't have a pure virtual destructor, but the compiler
(Sun SPARCstation cfront 2.00.02) doesn't object.  Consider the following:

	class Base {
	public:
		Base() { }
		virtual ~Base() = 0;	// bad!
		};


	class Derived : public Base {
	public:
		Derived() { }
		~Derived() { }
		};


	main()
	{
		Derived instance;
	}

This compiles successfully, but when you try to link a load module
from it the linker fails, saying it can't find " ___dt__4BaseFva ".
This actually makes sense, because a virtual destructor is different
from other virtual functions.  When the Derived 'instance' is destroyed,
both ~Derived() and ~Base() will be called, just as with a non-virtual
destructor.  What 'virtual' gives you is correct destruction even when
the object is accesed via a base class pointer, such as:

	Base *bp = new Derived;
	delete bp;	// both ~Derived() and ~Base() called

whereas if the destructor weren't virtual, only ~Base() would be called.
But by declaring ~Base() to be pure virtual, there actually is no ~Base()
routine, so the linker reports it as an undefined symbol.  It seems to
me that the compiler should either flag pure virtual destructors
as syntax errors, or generate an internal "null" ~Base() destructor
for classes derived from Base which provide their own destructor.
Do other compilers/versions catch this problem?

Bill Linke
uunet!sbi!gort!wfl