Re: Why is this code accepted?

Re: Why is this code accepted?

[Mike Stump]

> To: gcc@gcc.gnu.org
> From: nbecker@fred.net
> Date: 16 Apr 2001 16:27:42 -0400

> This code compiles without complaint both on gcc-2.96 and on a
> recent gcc-3 snapshot, but doesn't link.  I suspect the
> initialization of the static "a" is invalid (I don't have a good
> reference handy), but isn't gcc behaviour wrong?

There was a g++ feature that allowed this to work, a while ago.
Someone broke that feature, and didn't remove the feature from all
places in the compiler that supported the feature.  As I recall, the
standard doesn't support this feature.

Re: Why is this code accepted?

[Alexandre Oliva]

On Apr 16, 2001, Mike Stump <mrs@windriver.com> wrote:

>> To: gcc@gcc.gnu.org
>> From: nbecker@fred.net
>> Date: 16 Apr 2001 16:27:42 -0400

>> This code compiles without complaint both on gcc-2.96 and on a
>> recent gcc-3 snapshot, but doesn't link.  I suspect the
>> initialization of the static "a" is invalid (I don't have a good
>> reference handy), but isn't gcc behaviour wrong?

> There was a g++ feature that allowed this to work, a while ago.
> Someone broke that feature, and didn't remove the feature from all
> places in the compiler that supported the feature.  As I recall, the
> standard doesn't support this feature.

Anyway, even in the few cases in which it does allow a static const
data member to be initialized inside the class body, it still requires
the data member to be defined outside the class body, which would
probably have got you around the problem.

-- 
Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer                  aoliva@{cygnus.com, redhat.com}
CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist    *Please* write to mailing lists, not to me

Re: Why is this code accepted?

[nbecker]

If it doesn't work shouldn't gcc reject it?

Re: Why is this code accepted?

[Alexandre Oliva]

On Apr 17, 2001, nbecker@fred.net wrote:

> If it doesn't work shouldn't gcc reject it?

Nope.  How would GCC tell that the static data member isn't going to
be defined in another translation unit?  That's why this rule doesn't
require a diagnostic from the compiler.

-- 
Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer                  aoliva@{cygnus.com, redhat.com}
CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist    *Please* write to mailing lists, not to me

Re: Why is this code accepted?

[Mike Stump]

> Cc: Mike Stump <mrs@windriver.com>, gcc@gcc.gnu.org
> From: Alexandre Oliva <aoliva@redhat.com>
> Date: 17 Apr 2001 18:51:45 -0300

> On Apr 17, 2001, nbecker@fred.net wrote:

> > If it doesn't work shouldn't gcc reject it?

Oh, wait, I know what's wrong...

g++  -pedantic-errors -ansi -S t.cc
t.cc:13: ISO C++ forbids initialization of member constant `a' of non-integral 
   type `const double[]'

It works as designed, and if we add:

const double X::a[2];

to the program to meet the logical requirements of the extension, and
don't ask for the error, then it works as expected.  I was thinking
of a language without the clause:

4 If a  static data member is of  const integral or   const  enumeration
  type,  its  declaration  in  the  class  definition  can  specify  a
  constant-initializer which shall be an  integral  constant  expression
  (5.19).   In  that  case,  the  member can appear in integral constant
  expressions within its scope.  The member shall still be defined in  a  |
  namespace  scope  if it is used in the program and the namespace scope  |
  definition shall not contain an  initializer.

in it, sorry for the confusion.  This wording first appeared in the 95
(95-0029) version of the working paper, so it is new...  :-)

So, no, no extension was broken, and no, no bug is present.

Re: Why is this code accepted?

[Mike Stump]

> To: Alexandre Oliva <aoliva@redhat.com>
> Cc: Mike Stump <mrs@windriver.com>, gcc@gcc.gnu.org
> From: nbecker@fred.net
> Date: 17 Apr 2001 07:56:56 -0400

> If it doesn't work shouldn't gcc reject it?

Yes, like I said, I believe that the standard mandates that a
diagnostic is produced for that code.

Why is this code accepted?

[nbecker]

This code compiles without complaint both on gcc-2.96 and on a recent gcc-3 snapshot,
but doesn't link.  I suspect the initialization of the static "a" is
invalid (I don't have a good reference handy), but isn't gcc behaviour
wrong?  If the initialization is invalid, why doesn't the compiler
complain (it doesn't even with -Wall)?  If not, why doesn't it link?
--------------------

cd /home/nbecker/bitexact/bitexact/
/opt/gcc/bin/g++ -o bug bug.cc
/tmp/ccZtF0aa.o: In function `X::X()   ':
/tmp/ccZtF0aa.o(.gnu.linkonce.t._ZN1XC1Ev+0xa): undefined reference to `X::a    '
/tmp/ccZtF0aa.o(.gnu.linkonce.t._ZN1XC1Ev+0xf): undefined reference to `X::a    '
collect2: ld returned 1 exit status

--------------------
#include <vector>
using namespace std;

class Y {
  vector<double> c;
public:
  Y (const double* a, const double* b) :
    c (a, b)
  {}
};

class X {
  static const double a[] = {1, 2};
  Y y;
public:
  X () :
    y (&a[0], &a[2])
  {}

};

main () {
  vector<X> x (2);
}
-------------------