Bug 920269

Reduced testcase for -g:
class A
{
  void a1 ();
  template <class T>
  void a2 ();
  template <class T, class U>
  void a2 ();
  template <class V>
  void a3 ();
};
template <class T, class U>
struct B
{
  typedef T TT;
};
template <class T, unsigned int V = 2>
class C {};
template <class T, class U>
class D {};
template <class T, class U>
struct E : public D <T, U>
{
  struct F {};
  typedef F G __attribute__ ((aligned (64)));
};
template <class V>
struct H
{
  typedef float I;
};
template <class T>
void
A::a3 ()
{
  typedef C <T, 3> J;
  typedef E <J, J> K;
  typename K::L l = K::k ();
}
template <class T, class U>
void
A::a2 ()
{
  typedef B <T, U> TraitsType;
  typedef typename TraitsType::TT CommonCVoxelType;
  typedef typename H <CommonCVoxelType>::I I;
  this->a3 <I> ();
}
template <class T>
void
A::a2 ()
{
  a2 <T, double> ();
}
void
A::a1 ()
{
  a2 <double> ();
}

I bet the typedef of a struct with aligned attribute is the problem, that had been supported really badly in older compilers.

Indeed, http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=128488 fixed this,
not sure how risky the backport would be at this point.

Setting proper component.

Based on feedback from Jakub & Jason, the fix for this problem is too risky to backport to the gcc 4.1 toolchain on Red Hat Enterprise Linux 5.

The gcc44 compiler for Red Hat Enterprise Linux 5 as well as the Developer Toolset releases for Red Hat Enterprise Linux 5 handle this code correctly.