Block scope using declarations of operators

Block scope using declarations of operators

[Richard Smith]

Gcc's C++ grammar does not accept block scope using declarations of
operators, e.g.

    namespace NS
    {
      class A {};
      bool operator==( const A &, const A & );
    }

    int main()
    {
      using NS::operator==;  // get a parse error on this line
    }


The following patch fixes this issue

diff -urpx diff_ignore gcc-3.1-clean/gcc/c-decl.c gcc-3.1/gcc/c-decl.c
--- gcc-3.1-clean/gcc/cp/parse.y	Thu Jun 27 16:19:21 2002
+++ gcc-3.1/gcc/cp/parse.y	Wed Jul 10 14:00:33 2002
@@ -567,6 +567,12 @@ namespace_using_decl:
 	        { $$ = build_nt (SCOPE_REF, global_namespace, $3); }
 	| USING global_scope namespace_qualifier identifier
 	        { $$ = build_nt (SCOPE_REF, $3, $4); }
+	| USING namespace_qualifier operator_name
+	        { $$ = build_nt (SCOPE_REF, $2, $3); }
+	| USING global_scope operator_name
+	        { $$ = build_nt (SCOPE_REF, global_namespace, $3); }
+	| USING global_scope namespace_qualifier operator_name
+	        { $$ = build_nt (SCOPE_REF, $3, $4); }
 	;

 using_directive:


--
Richard Smith

Re: Block scope using declarations of operators

[Gabriel Dos Reis]

Richard Smith <richard@ex-parrot.com> writes:

| Gcc's C++ grammar does not accept block scope using declarations of
| operators, e.g.

This is a long standing misbehaviour in GCC (already present the 2.95
series).  Although serious, I'm not sure it qualifies as a regression.

Thanks,

-- Gaby

Re: Block scope using declarations of operators

[Richard Smith]

On 10 Jul 2002, Gabriel Dos Reis wrote:

> Richard Smith <richard@ex-parrot.com> writes:
>
> | Gcc's C++ grammar does not accept block scope using declarations of
> | operators, e.g.
>
> This is a long standing misbehaviour in GCC (already present the 2.95
> series).

Yes, I'd noticed that too, although I couldn't find a PR for it.

Are there any particular reasons why it hadn't been fixed yet?  Or is it
just lack of time/enthusiasm.  Unless I'm missing something the fix ought
to be quite straight-forward.

> Although serious, I'm not sure it qualifies as a regression.

Yes, I agree.  Development for 3.2 is only in stage 2, so might it be
appropriate to apply it there?

--
Richard Smith

Re: Block scope using declarations of operators

[Gabriel Dos Reis]

Richard Smith <richard@ex-parrot.com> writes:

| On 10 Jul 2002, Gabriel Dos Reis wrote:
| 
| > Richard Smith <richard@ex-parrot.com> writes:
| >
| > | Gcc's C++ grammar does not accept block scope using declarations of
| > | operators, e.g.
| >
| > This is a long standing misbehaviour in GCC (already present the 2.95
| > series).
| 
| Yes, I'd noticed that too, although I couldn't find a PR for it.
| 
| Are there any particular reasons why it hadn't been fixed yet?  Or is it
| just lack of time/enthusiasm.  Unless I'm missing something the fix ought
| to be quite straight-forward.

I got caught by that misbehaviour in the past, but didn't put it on the
top on my TODO list.  I think it was fixed by lack of time, not lack
of enthusiasm.

| > Although serious, I'm not sure it qualifies as a regression.
| 
| Yes, I agree.  Development for 3.2 is only in stage 2, so might it be
| appropriate to apply it there?


Yes, I think 3.2 should be the right place.  After 3.1.1 release, we
might raise the issue again for 3.1.x, but it is Mark's call.

-- Gaby

Re: Block scope using declarations of operators

[Mark Mitchell]

> Yes, I think 3.2 should be the right place.  After 3.1.1 release, we
> might raise the issue again for 3.1.x, but it is Mark's call.

This is not appropriate for the 3.1 branch, but it's fine for 3.2 if
we think it's the right patch.  Martin did a lot of that work; Martin,
are you out there anywhere still?

-- 
Mark Mitchell                mark@codesourcery.com
CodeSourcery, LLC            http://www.codesourcery.com

Re: Block scope using declarations of operators

[Gabriel Dos Reis]

Mark Mitchell <mark@codesourcery.com> writes:

| > Yes, I think 3.2 should be the right place.  After 3.1.1 release, we
| > might raise the issue again for 3.1.x, but it is Mark's call.
| 
| This is not appropriate for the 3.1 branch, but it's fine for 3.2 if
| we think it's the right patch.

I have to confess that I didn't evaluate the correctness and
robustness of the patch.  I failed to make it clear that I was just
suggesting that fixing that misbehaviour should be done on 3.2.

Robustness and correctness will come later -- once I'm done with the
PPC/varargs thingy.

Thanks,

-- Gaby