Re: [boost] Serious problem with gcc 4 on Mac OS X, when using boost:shared_ptr

From: Ananda Tallur (anand_at_[hidden])
Date: 2006-03-26 17:52:27

Peter Dimov Wrote:

> Ananda Tallur wrote:
> > Hello,
> >
> > Originally, I had discovered a problem when using
> > boost::unit_test_framework, for which I have already posted.
> > But, by further analysing the problem, I could come out to a simple
> > program, using only boost shared pointers, and which fails :
> > -> on Mac OS X 10.4
> > -> when compiled with gcc 4.0.1 (coming with XCode 2.2.1), or gcc 4.1
> > (delivered by darwin ports)
> > -> with optimisations : at least -O, or -O1, or -O2,etc... It works
> > correctly with -O0
> > -> using either boost 1.33.1, 1.33.0, or 1.32.0
> >
> > Below is the sample test program :
> >
> > --- macosx_gcc4problem.cpp ---
> 
> [...]
> 
> This issue has already been reported, along with a fix, by Howard Hinnant:
> 
> http://lists.boost.org/Archives/boost/2006/03/102088.php
> http://lists.boost.org/Archives/boost/2006/03/102122.php
> http://lists.boost.org/Archives/boost/2006/03/102134.php
> 
> and the upcoming 1.34 release will no longer suffer from it.
> 
> Thank you very much for taking the time to simplify the failing example. 
> I've reformatted it a bit as a test case. If you verify that it still fails 
> for you, I'll include it in the regression tests.
> 
> 
> 
> #include <boost/shared_ptr.hpp>
> #include <boost/detail/lightweight_test.hpp>
> 
> struct A
> {
>     int value;
> };
> 
> struct B
> {
>     B( boost::shared_ptr<A> val ): a( val ) {}
> 
>     boost::shared_ptr<A> a;
> };
> 
> struct C
> {
>     C( B val ): b( val ) {}
> 
>     B b;
> };
> 
> struct D
> {
>     D( C val ): c( val ) {}
> 
>     C c;
> };
> 
> D create()
> {
>     boost::shared_ptr <A> a( new A );
> 
>     a->value = 12345;
> 
>     return D( C( B( a ) ) );
> }
> 
> int main()
> {
>     D d = create();
> 
>     BOOST_TEST( d.c.b.a->value == 12345 );
> 
>     return boost::report_errors();
> }

Hello Peter,
Thank you very much, this is very good news to know that the problem has
already been fixed !

I have checked the reformatted test program just above, and confirm
that :
- It does not work for me (gcc 4.0.1 on Mac OS X, ppc G5), when compiled
with optimization.

- Once patched the file sp_counted_base_gcc_ppc.hpp including
modifications from cvs, the problem disappears, and the test passes with
success, with or without optimization.

So the test program does its job, 
Thank you again
Anand