NetNews Usenet Archive 1992 #19/NN_1992_19.iso/spool/comp/lang/cplus/13199

Text File | 1992-09-02 | 2.2 KB | 57 lines

Newsgroups: comp.lang.c++
Path: sparky!uunet!cs.utexas.edu!hellgate.utah.edu!fcom.cc.utah.edu!swillden
From: [e-mail] (Shawn Willden)
Subject: "Recursive" template problem
Message-ID: <[e-mail]>
Sender: [e-mail]
Organization: University of Utah Computer Center
X-Newsreader: Tin 1.1 PL3
References: <[e-mail]>
Date: Wed, 2 Sep 92 15:06:11 GMT
Lines: 44

To any of you who already saw my first post on this subject, I apologize.
However, given the resounding silence that post elicited (no one even
corrected my syntax errors!), I thought maybe I had better try again,
with a different subject (specifically with one that doesn't include
'BC' - since that was in my own kill file for quite some time).

The problem is this:  Given a template base class that contains pointers
to two of its derived classes, at least two compilers (gcc-2.2.2 and BC++
3.1) choke (the former dumps core, the latter reports an error).  The
code looks something like this:

// Forward declarations of A1<T> and A2<T>
template<class T> class A1;
template<class T> class A2;

// Now we'll do the actual class definitions.
template<class T> class A {A1<T>* a; A1<T>* b;};
template<class T> class A1 : public A<T> {};
template<class T> class A2 : public A<T> {};

int main() {A1<int> myA1;}

What happens is, the compiler:
    begins to expand A1<int>,
    realizes it needs A<int> first,
    begins to expand A<int>,
    *thinks* it needs to expand A2<int> (which it doesn't, because b is
        and A2<int>*),
    begins to expand A2<int>,
    realizes it needs A<int> first,
    notices that A<int> is partially expanded
    and pukes.

So, has anyone else seen this problem?  Come up with a work-around?  If I
get some time I'm going to try to fix gcc but my archie server doesn't
seem to find anything under BC++3.1_source_code.tar.Z ;-) that one's going
to be a little tougher.  Are there any compilers that *do* handle this?
(I only have access to gcc 2.2.2, BC++ 3.1 and some version of Oregon C++
that doesn't support templates [The man page says the -Xv switch tells
Oregon C++ to report what version it is, but it doesn't work]).

--
Shawn Willden
[e-mail]