| Summary: | specialization of template class with inner template members | ||
|---|---|---|---|
| Product: | gcc | Reporter: | Boris Dorès <boris> |
| Component: | c++ | Assignee: | Not yet assigned to anyone <unassigned> |
| Status: | RESOLVED FIXED | ||
| Severity: | normal | CC: | jwakely.gcc |
| Priority: | P2 | Keywords: | rejects-valid |
| Version: | 3.4.3 | ||
| Target Milestone: | --- | ||
| Host: | i686-pc-linux-gnu | Target: | i686-pc-linux-gnu |
| Build: | i686-pc-linux-gnu | Known to work: | 4.10.0, 4.8.0, 4.9.0 |
| Known to fail: | 2.95.3, 3.0.4, 3.2.3, 3.3.2, 3.4.3, 4.0.0 | Last reconfirmed: | 2004-12-13 15:27:08 |
Should be allowed according to 14.7.3/16. All my GCC versions
fail to compile the testcase (maybe some 3.3.x does) so I can't
verify that it's a regression.
The problem is some incorrect logic in 'determine_specialization':
it tries to deduce N2 from
template<> template<class T2,int N2>
T2 Class<int,1>::function( T2 param )
to match against
template<class T1,int N1> template<class T2,int N2>
T2 Class<T1,N1>::function( T2 param );
3.3.2 also rejects this code also: pr18950.cc:11: error: template-id `function<>' for `T2 Class<int, 1>::function(T2)' does not match any template declaration pr18950.cc:11: error: syntax error before `{' token (In reply to comment #2) > 3.3.2 also rejects this code also: I'm really sorry, it seems I oversimplified our real test case (the previous problem might be related, but indeed it's not a regression). The following code is a closer version of what we have, and this time it compiles fine with g++ 3.3.1, but not with g++ 3.4.0 : template<class T1,int N1> class Class { public: template <class T2,int N2> void function( const Class<T2,N2>& ); }; template<> template<class T2,int N2> void Class<int,1>::function( const Class<T2,N2>& param ) { param;// line 12 } int main() { Class<int,1> instance; Class<char,2> param; instance.function( param ); } The output is : test-gcc34.cpp: In member function `void Class<T1, N1>::function(const Class<T2, N2>&) [with T2 = T2, int N2 = N2, T1 = int, int N1 = 1]': test-gcc34.cpp:12: error: `param' undeclared (first use this function) test-gcc34.cpp:12: error: (Each undeclared identifier is reported only once for each function it appears in.) Sorry about the misinformation in the first place. (In reply to comment #3) > (In reply to comment #2) > > Sorry about the misinformation in the first place. I split this off to PR 18962, yes that testcase is a regression, but I think the problem which causes to reject the first testcase and the second one, is different which is why I split it off to a new bug report. Fixed looong time ago. |
The following code used to compile fine with gcc 3.3 (linux x86), but doesn't anymore with gcc 3.4 (just as a hint, it seems to compile fine with Comeau) : template<class T1,int N1> class Class { public: template<class T2,int N2> T2 function( T2 param ); }; template<> template<class T2,int N2> T2 Class<int,1>::function( T2 param ) { // line 11 return param * N2; } int main() { Class<int,1> instance; return instance.function<char,4>( 12 ); } This produces the following output: test-gcc34.cpp:11: error: template-id `function<>' for `T2 Class<int, 1>::function(T2)' does not match any template declaration test-gcc34.cpp:11: error: erreur de syntaxe before `{' token It works if we add the specialized declaration of the class "Class": template<> class Class<int,1> { public: template<class T2,int N2> T2 function( T2 param ); }; but this can be painful if "Class" contains a lot more members. It also works if "function" is only templated by one argument : template<> template<class T2> T2 Class<int,1>::function( T2 param ) { // line 11 return param * 2; } Thanks.