Re: [boost] dynamic_cast in MSland

From: Greg Colvin (gcolvin_at_[hidden])
Date: 2000-10-30 13:52:00

From: Peter Dimov <pdimov_at_[hidden]>
To: <boost_at_[hidden]>
Sent: Monday, October 30, 2000 11:42 AM
Subject: Re: [boost] dynamic_cast in MSland

> > It turns out there is a longstanding bug where Microsoft Visual C++
> > can optimize away the catch blocks for a throw from dynamic_cast:
> >
> >    http://support.microsoft.com/support/kb/articles/Q247/2/03.ASP
> >
> > The above states version 6, but I have reproduced this in 5 as well.
> 
> Hmm. The example in the article dynamic_casts an uninitialized pointer,
> which is undefined behavior, right?

Yes, but the following program fails to print "caught" if
compiled with /Og and without /EHa.

   #include <stdio.h>
   struct A { virtual void v() {} };
   struct B:A {};
   main() {
     A a;
     try { 
        dynamic_cast<B&>(a);
     } catch (...) {
        puts("caught");
     }
   }