[Bug c++/43522] New: 4.4 regression: Attempts to instantiate unnecessary code

[Bug c++/43522] New: 4.4 regression: Attempts to instantiate unnecessary code

[georgeh at rentec dot com]

Code:
---------------
template <bool x> struct tester;
template <> struct tester<true> { enum { value = 1 }; };

template <class T>
struct allow_double_instantiations {
  static const bool value = false;
};

template <int Power>
class FixedPoint {
  static const bool allow_double_instantiations = false;

  typedef FixedPoint<Power> Self;

public:
  FixedPoint(): mValue(0) {}

  //Default copy c'tor, assignment op valid & useful.

  Self &operator=(double const &d) {
    mValue = sizeof(tester<Self::allow_double_instantiations>);
    return *this;
  }

  int mValue;
};

void foo(FixedPoint<4> &);

int main() {
  FixedPoint<4> x;
  foo(x);
  return 0;
}

template <>
FixedPoint<4> &FixedPoint<4>::operator=(double const &d) {
  mValue = int(d);
  return *this;
}

void foo(FixedPoint<4> &t) {
  t = 13.5;
}
-------------

gcc 4.3.2 and earlier versions compile this without complaint; 4.4.3 dies with 
test_fixedpoint.C: In member function `FixedPoint<Power>&
FixedPoint<Power>::operator=(const double&)':
test_fixedpoint.C:21: error: invalid application of `sizeof' to incomplete type
`tester<false>'

By my reading of the standard, the invalid definition of operator= should not
be attempted to be instantiated until line 43, where the valid template
specialization has been seen, and previous versions of g++ agree with me. Note
that in the original code, line 21 was a BOOST_STATIC_ASSERT and produced the
same result.


-- 
           Summary: 4.4 regression: Attempts to instantiate unnecessary code
           Product: gcc
           Version: 4.4.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: georgeh at rentec dot com
GCC target triplet: i686-suse-linux


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43522

[Bug c++/43522] 4.4 regression: Attempts to instantiate unnecessary code

[georgeh at rentec dot com]

------- Comment #1 from georgeh at rentec dot com  2010-03-25 18:07 -------
g++ -v output -- apologies: 

Using built-in specs.
Target: i686-suse-linux
Configured with: ../../gcc-4.4.3/configure --enable-languages=c,c++,fortran --p
refix=/usr/local/products/gcc/4.4.3-32 --with-gnu-as --with-as=/usr/local/produ
cts/gcc/binutils-2.19-32/bin/as --with-gnu-ld --with-ld=/usr/local/products/gcc
/binutils-2.19-32/bin/ld --enable-threads=posix --enable-shared --enable-__cxa_
atexit --with-gmp=/usr/local/products/gcc/gmp-4.2.4-32 --with-mpfr=/usr/local/p
roducts/gcc/mpfr-2.3.2-32 --enable-libstdcxx-allocator=pool i686-suse-linux
Thread model: posix
gcc version 4.4.3 (GCC) 


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43522

[Bug c++/43522] 4.4 regression: Attempts to instantiate unnecessary code

[pinskia at gcc dot gnu dot org]

------- Comment #2 from pinskia at gcc dot gnu dot org  2010-03-25 18:08 -------
So you are saying the standard thinks
FixedPoint<Power>::allow_double_instantiations is dependent.  


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43522

[Bug c++/43522] 4.4 regression: Attempts to instantiate unnecessary code

[pinskia at gcc dot gnu dot org]

------- Comment #3 from pinskia at gcc dot gnu dot org  2010-03-25 18:20 -------
I think that name is not dependent because or DR 224, see PR 29607.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43522

[Bug c++/43522] [4.4 regression] Attempts to instantiate unnecessary code

[bangerth at gmail dot com]

------- Comment #4 from bangerth at gmail dot com  2010-03-25 18:37 -------
(In reply to comment #2)
> So you are saying the standard thinks
> FixedPoint<Power>::allow_double_instantiations is dependent.  

That is the correct question. We get the error message during
template parsing, not during template instantiation (otherwise
the error message would have "[with Power=4]"). So apparently
the compiler thinks that
  Self::allow_double_instantiations
is non-dependent and so tries to instantiate the "tester" template
I think this is wrong: if you write it as Self::allow_double_instantiations
it should be dependent, whereas if you write it without the Self:: it 
should be non-dependent.

Can someone double check whether this also happens with mainline?

W.


-- 

bangerth at gmail dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jason at redhat dot com,
                   |                            |bangerth at gmail dot com
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2010-03-25 18:37:17
               date|                            |
            Summary|4.4 regression: Attempts to |[4.4 regression] Attempts to
                   |instantiate unnecessary code|instantiate unnecessary code


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43522

[Bug c++/43522] [4.4 regression] Attempts to instantiate unnecessary code

[pinskia at gcc dot gnu dot org]

------- Comment #5 from pinskia at gcc dot gnu dot org  2010-03-25 18:47 -------
>it should be non-dependent.

Except DR 224 changed that.  See also PR 9634 comment #3 which shows that it is
not dependent at all.

And the trunk has the same behavior as 4.4.

So this code is invalid as DR 224 changed the wording such that name is non
dependent.  Sorry to say that but it is just a weird case. 


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |INVALID


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43522