[Bug c++/16777] New: C++ parse error in 3.3, OK in 3.2 and 3.4

[Bug c++/16777] New: C++ parse error in 3.3, OK in 3.2 and 3.4

[lindahl at pathscale dot com]

Reading specs from /usr/lib/gcc-lib/x86_64-redhat-linux/3.3.2/specs
Configured with: ../configure --prefix=/usr --mandir=/usr/share/man
--infodir=/usr/share/info --enable-shared --enable-threads=posix
--disable-checking --with-system-zlib --enable-__cxa_atexit
--host=x86_64-redhat-linux
Thread model: posix
gcc version 3.3.2 20031022 (Red Hat Linux 3.3.2-1)

This is OK in gcc 3.2, broken in gcc 3.3 (e.g. 3.3.2 from redhat), ok in gcc 3.4.0
 
g++ -c test4a.cc
test4a.cc: In member function `AttributeT<char*, int>*
   AttrManager::NewUserAttribute(const char*, const VcbSearchSet&)':
test4a.cc:40: error: `NewAttribute<char*, int>' undeclared (first use this
   function)
test4a.cc:40: error: (Each undeclared identifier is reported only once for each
   function it appears in.)
-----test4a.cc-----------------------------------------------------------------
 
#define NULL 0x00000000
class AttrManager;
 
static AttrManager* attr_manager;
 
template <class T, class Obj>
class AttributeT {
public:
  int x;
};
 
class VcbSearchSet {
 
  public:
    VcbSearchSet(void*);
};
 
 
class AttrManager {
public:
  AttrManager();
  ~AttrManager();
 
  template <class T, class Obj>
  AttributeT<T, Obj>* NewAttribute(const char* name, int vm,
    const char* help, const VcbSearchSet& vcbs,
    T (*default_value_function) (),
    int (*check_obj_and_value_function) (Obj*,T)
  )
  {
    return NULL;
  }
 
  AttributeT<char*, int>* AttrManager::NewUserAttribute(const char* name,
  const VcbSearchSet& vcbs)
  {
    int avm_user = 1;
    return attr_manager->
      AttrManager::NewAttribute<char*, int>(name, avm_user,
      "user defined", vcbs, NULL, NULL);
  }
 
};
 
----------------------------------------------------------------------
funky attempts to cast the NULL don't help: (and shouldn't be needed)
----------------------------------------------------------------------
This shows a workaround, remove the AttrManager:: in the call
----------------------------------------------------------------------
g++ -c test4b.cc
 
-----test4b.cc-----------------------------------------------------------------
#define NULL 0x00000000
class AttrManager;
 
static AttrManager* attr_manager;
 
template <class T, class Obj>
class AttributeT {
public:
  int x;
};
 
class VcbSearchSet {
 
  public:
    VcbSearchSet(void*);
};
 
 
class AttrManager {
public:
  AttrManager();
  ~AttrManager();
 
  template <class T, class Obj>
  AttributeT<T, Obj>* NewAttribute(const char* name, int vm,
    const char* help, const VcbSearchSet& vcbs,
    T (*default_value_function) (),
    int (*check_obj_and_value_function) (Obj*,T)
  )
  {
    return NULL;
  }
 
  AttributeT<char*, int>* AttrManager::NewUserAttribute(const char* name,
  const VcbSearchSet& vcbs)
  {
    int avm_user = 1;
    return attr_manager->
      NewAttribute<char*, int>(name, avm_user,
      "user defined", vcbs, (char*(*)())NULL, (int(*)(int*,char*))NULL);
  }
 
};
 
----------------------------------------------------------------------

-- 
           Summary: C++ parse error in 3.3, OK in 3.2 and 3.4
           Product: gcc
           Version: 3.3.2
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c++
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: lindahl at pathscale dot com
                CC: gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16777

[Bug c++/16777] [3.3 Regression] C++ parse error in 3.3, OK in 3.2 and 3.4

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-07-27 03:27 -------
Confirmed, I think this is just the parser not being able to handle template right, in previous versions 
of gcc before 3.3, it used to add the template keyword for you sometimes but in 3.3 we removed that 
extension which broke a couple of things, this example included.  Now in 3.4.0 we fixed this by 
rewriting the parser so it would find the places where you do not need (and cannot, this is an example 
of where you cannot use the template keyword) to use the template keyword.

This is almost an example of where we cannot fix 3.3.x at all.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |gdr at gcc dot gnu dot org
           Severity|normal                      |minor
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
           Keywords|                            |rejects-valid
   Last reconfirmed|0000-00-00 00:00:00         |2004-07-27 03:27:56
               date|                            |
            Summary|C++ parse error in 3.3, OK  |[3.3 Regression] C++ parse
                   |in 3.2 and 3.4              |error in 3.3, OK in 3.2 and
                   |                            |3.4
   Target Milestone|---                         |3.3.5


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16777

[Bug c++/16777] [3.3 Regression] Trouble with qualified declaration in class

[bangerth at dealii dot org]

------- Additional Comments From bangerth at dealii dot org  2004-07-27 03:38 -------
That's actually a strange piece of code:  
-------------------  
struct X {  
    template <typename> void foo();  
   
    void X::bar() {  
      this->X::foo<int>();  
    }  
};  
-------------------  
gcc 3.2, 3.4 and mainline actually accept the odd declaration using the   
qualification X:: of the member function, whereas 3.3 somehow manages  
to ICE (this is with checking enabled, of course):  
  
g/x> /home/bangerth/bin/gcc-3.3.4-pre/bin/c++ -c x.cc  
x.cc: In member function `void X::bar()':  
x.cc:6: internal compiler error: tree check: expected identifier_node, have   
   baselink in cxx_scope_find_binding_for_name, at cp/decl.c:2384  
Please submit a full bug report,  
with preprocessed source if appropriate.  
  
So there are two bugs: that 3.3.x ICEs, and that the other compilers  
actually accept the code. I'd bet that it is invalid; btw, icc8 says  
  
g/x> icc -Xc -ansi -c x.cc  
x.cc(4): warning #470: qualified name is not allowed in member declaration  
      void X::bar()  
           ^  
  
W.  

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[3.3 Regression] C++ parse  |[3.3 Regression] Trouble
                   |error in 3.3, OK in 3.2 and |with qualified declaration
                   |3.4                         |in class


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16777

[Bug c++/16777] [3.3 Regression] Trouble with qualified declaration in class

[lindahl at pathscale dot com]

------- Additional Comments From lindahl at pathscale dot com  2004-07-27 03:43 -------
I _think_ that Sun's C++ guru approved this code, but if you like I can send a
question back through channels -- I'm not a C++ language laywer. Sun's compiler
does accept it, apparently. This is a snippet from an ISV code.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16777

[Bug c++/16777] [3.3 Regression] Trouble with qualified declaration in class

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-07-27 03:56 -------
Note after removing the "qualified declaration", gcc 3.3.3 rejects the code so that is still a bug.
Now the other issue in this bug is a different bug all together.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16777

[Bug c++/16777] [3.3 Regression] Trouble with qualified declaration in class

[bangerth at dealii dot org]

------- Additional Comments From bangerth at dealii dot org  2004-07-27 04:05 -------
gcc3.3 required you to write 
  this->X::template foo<int> (); 
That's fixed, though, in 3.4 and beyond and immaterial to the bug here. 
Note, however, that using above syntax also gets us rid of the ICE. 
 
W. 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16777

[Bug c++/16777] [3.3 Regression] Trouble with qualified declaration in class

[bangerth at dealii dot org]

------- Additional Comments From bangerth at dealii dot org  2004-07-27 13:17 -------
I have separated this PR into two new ones about the different 
parts of what's going wrong here: PR 16781 and PR 16782. This 
PR is therefore obsolete. 
 
W. 

*** This bug has been marked as a duplicate of 16781 ***

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16777