Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!utgpu!water!watmath!clyde!rutgers!sri-spam!ames!ucbcad!ucbvax!decvax!decwrl!labrea!uluru!calder
From: calder@uluru.UUCP
Newsgroups: comp.lang.c++
Subject: Possible cfront v1.2 bug.
Message-ID: <1728@uluru.STANFORD.EDU>
Date: Thu, 4-Jun-87 19:17:06 EDT
Article-I.D.: uluru.1728
Posted: Thu Jun  4 19:17:06 1987
Date-Received: Sat, 6-Jun-87 11:34:22 EDT
Reply-To: calder@uluru.UUCP (Paul Calder)
Distribution: world
Organization: Stanford University (Centre for Integrated Systems)
Lines: 48
Keywords: Version 1.2, member functions

I have had a few problems with pointers to member functions in 
version 1.2 of the C++ translator.  Specifically, there seem to
be bugs in using local variables of type "pointer-to-member-function",
especially if the class concerned has virtual member functions.

Here is a simple example.

	struct S {
	    virtual void f() { }
	} s;

	void (S::*p)() = &S::f;	    // works fine if p is global
	
	int main() {
	    (s.*p)();
	}

But now try making p local to main.

	struct S {
	    virtual void f() { }
	} s;
	
	int main() {
	    void (S::*p)() = &S::f; // "syntax error" if p is local
	    (s.*p)();
	}

OK then, why not make a global typedef and a local variable?

	struct S {
	    virtual void f() { }
	} s;
	
	typedef void (S::*pf)();
	
	int main() {
	    pf p = &S::f;
	    (s.*p)();
	}			    // "bus error (or something nasty...)"

Note that if S::f is not virtual, all is well. (Also note that this used to
work fine under version 1.1)

Is this a bug, or am I missing something?

Paul Calder (calder@uluru.stanford.edu)
Stanford University, Centre for Integrated Systems, Computer Systems Lab.