[Bug c++/54740] New: Empty base optimization gets confused when a member variable derives from the same empty base.

[Bug c++/54740] New: Empty base optimization gets confused when a member variable derives from the same empty base.

[avenikov at google dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54740

             Bug #: 54740
           Summary: Empty base optimization gets confused when a member
                    variable derives from the same empty base.
    Classification: Unclassified
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: major
          Priority: P3
         Component: c++
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: avenikov@google.com


struct empty_base {};
struct another_empty_base {};

struct inner : empty_base {};
struct outer : empty_base {
    inner i_;
};


int main(int, char **)
{
    return sizeof(outer);
}

This returns 2;
If you change outer to derive from another_empty_base, it returns 1 (which is
correct). Whatever the correct result should be, they at least should be the
same.

I tried it on 4.7.2 and on 4.4.3 - same problem is present.
The optimization flags don't seem to make any difference.

Setting it to major, as this seems like a very common idiom and people expect
empty base optimization to kick in in such cases.


Andy.

P.S. Just in case: partial output from my gcc -v:
Using built-in specs.
COLLECT_GCC=/<...>/stage-4.7.2/bin/gcc
COLLECT_LTO_WRAPPER=/<...>/stage-4.7.2/bin/../libexec/gcc/x86_64-unknown-linux-gnu/4.7.2/lto-wrapper
Target: x86_64-unknown-linux-gnu
Configured with: ../gcc-4.7.2/configure --prefix=/<...>/stage-4.7.2
--with-gmp=/<...>/gmp-4.3.2/ --with-mpfr=/<...>/mpfr-3.0.0/
--with-mpc=/<...>/mpc-0.9/ --enable-languages=c,c++
Thread model: posix
gcc version 4.7.2 (GCC) 

It's a ubuntu-based system.

[Bug c++/54740] Empty base optimization gets confused when a member variable derives from the same empty base.

[redi at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54740

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |INVALID
           Severity|major                       |normal

--- Comment #1 from Jonathan Wakely <redi at gcc dot gnu.org> 2012-09-28 23:58:34 UTC ---
(In reply to comment #0)
> If you change outer to derive from another_empty_base, it returns 1 (which is
> correct). Whatever the correct result should be, they at least should be the
> same.

No they shouldn't. The standard requires this to pass:

  outer o;
  assert( static_cast<inner*>(&o) != &o.i_ );

[intro.object]/6 requires two separate objects of the same type to have
distinct addresses.

[Bug c++/54740] Empty base optimization gets confused when a member variable derives from the same empty base.

[redi at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54740

--- Comment #2 from Jonathan Wakely <redi at gcc dot gnu.org> 2012-09-29 00:01:31 UTC ---
(In reply to comment #1)
> No they shouldn't. The standard requires this to pass:
> 
>   outer o;
>   assert( static_cast<inner*>(&o) != &o.i_ );

Oops, that should be:

   assert( static_cast<empty_base*>(&o) != &o.i_ );

[Bug c++/54740] Empty base optimization gets confused when a member variable derives from the same empty base.

[avenikov at google dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54740

--- Comment #3 from Andy Venikov <avenikov at google dot com> 2012-09-29 00:09:55 UTC ---
(In reply to comment #2)
> (In reply to comment #1)
> > No they shouldn't. The standard requires this to pass:
> > 
> >   outer o;
> >   assert( static_cast<inner*>(&o) != &o.i_ );
> 
> Oops, that should be:
> 
>    assert( static_cast<empty_base*>(&o) != &o.i_ );

I see.
I was aware of the rule you mentioned, but it was hard to see how it could
apply here.
Sorry for the noise then.