stl_relops.h collision conflict

stl_relops.h collision conflict

[Olaf Petzold]

Hello,

I'm writing an ET library. Unfortunally I'm gone into trouble
with the stl internal header stl_relops.h from gcc:

template <class _Tp>
inline bool operator>(const _Tp& __x, const _Tp& __y) {
  return __y < __x;
}

and my:

template<class T1, class T2, size_t Sz>
inline
Xpr<
  XprBinOp<
    VectorReference<T1, Sz>,
    Functional::Greater<T1, T2>,
    VectorReference<T2, Sz>
  >
>
operator>(const Vector<T1, Sz>& lhs, const Vector<T2, Sz>& rhs) {
  typedef XprBinOp<
    VectorReference<T1, Sz>,
    Functional::Greater<T1, T2>,
    VectorReference<T2, Sz>
    >                                                   expr_type;
  return Xpr<expr_type>(expr_type(lhs.ref(), rhs.ref()));
}

Is there no way to force the compiler to use my implementation of
operator>() ? 

On gcc-3.0.1 the relops header has an own inner namespace but, the problems
has been occoured in prior version as well. Is there a work around ?

Thanks
Olaf

Re: stl_relops.h collision conflict

[Joe Buck]

> I'm writing an ET library. Unfortunally I'm gone into trouble
> with the stl internal header stl_relops.h from gcc:

Please be specific about versions.

> template <class _Tp>
> inline bool operator>(const _Tp& __x, const _Tp& __y) {
>   return __y < __x;
> }

2.95.x, egcs, or "2.96".

> and my:
> 
[ templated operator> that does not return a bool ]

> Is there no way to force the compiler to use my implementation of
> operator>() ? 

For older gcc's, you might try explicitly instantiating the template.

> On gcc-3.0.1 the relops header has an own inner namespace but, the problems
> has been occoured in prior version as well. Is there a work around ?

In 3.0.x or other ISO standard C++ compiler, if you avoid importing
rel_ops with a using directive, you should have no problem.

Re: stl_relops.h collision conflict

[Olaf Petzold]

Am Die, 11 Sep 2001 schrieben Sie:
> > I'm writing an ET library. Unfortunally I'm gone into trouble
> > with the stl internal header stl_relops.h from gcc:
> 
> Please be specific about versions.
> 
> > template <class _Tp>
> > inline bool operator>(const _Tp& __x, const _Tp& __y) {
> >   return __y < __x;
> > }
> 
> 2.95.x, egcs, or "2.96".
2.96 mandrake 8.0

> > On gcc-3.0.1 the relops header has an own inner namespace but, the problems
> > has been occoured in prior version as well. Is there a work around ?
> 
> In 3.0.x or other ISO standard C++ compiler, if you avoid importing
> rel_ops with a using directive, you should have no problem.
That's right, no problem on using it but, worse code. See inliner problems. I
haven't checked Kurt's patch v3 yet.

Regards
Olaf

Re: stl_relops.h collision conflict

[Joe Buck]

[ conflicts with templated operator>()]

> > In 3.0.x or other ISO standard C++ compiler, if you avoid importing
> > rel_ops with a using directive, you should have no problem.
> That's right, no problem on using it but, worse code. See inliner problems. I
> haven't checked Kurt's patch v3 yet.

According to Gerald's tests 3.0.2 should be a lot better.  You can
help ensure that it is by testing it (from snapshots or CVS using
-rgcc-3_0-branch).