g++ template problem

[snorkelman]

I'm having a compile time problem with g++ 3.2.something, which I can
boil down to this:

struct A {
template<int N> void foo(){}
};

template<class T>
void fn( int & ) {
A a;
a.foo<0>(); //compiler reports "syntax error before semicolon"
}

main()
{
A a;
a.foo<0>();
}

Note that the template function is not being instantiated - the error
arises from the function definition. It appears that the compiler is
confused about just what foo is.

But using it in main() gives me no problem - the compiler is well aware
at that point that foo<N> is a function.

There are a couple of workarounds that coerce everything into
compiling, but in my real code things are a little hairier and I
haven't been able to get them to work. I guess what I am hoping to find
out first is what I am doing wrong. Anything obvious?

|
| markn at ieee dot org
|

 

[Alf P. Steinbach]

* (e-mail address removed):

I'm having a compile time problem with g++ 3.2.something, which I can
boil down to this:

struct A {
template<int N> void foo(){}
};

template<class T>
void fn( int & ) {
A a;
a.foo<0>(); //compiler reports "syntax error before semicolon"
}

main()
{
A a;
a.foo<0>();
}

Compiles fine with g++ 3.4.2, MSVC 7.1 and Comeau 4.3.3 online.

 

[Victor Bazarov]

I'm having a compile time problem with g++ 3.2.something, which I can
boil down to this:

struct A {
template<int N> void foo(){}
};

template<class T>
void fn( int & ) {
A a;
a.foo<0>(); //compiler reports "syntax error before semicolon"
}

main()

int main()

{
A a;
a.foo<0>();
}

Note that the template function is not being instantiated - the error
arises from the function definition. It appears that the compiler is
confused about just what foo is.

It's quite possible that the compiler is confused, but both VC++ 2003 and
Comeau (online) compile it without a problem.

But using it in main() gives me no problem - the compiler is well aware
at that point that foo<N> is a function.

There are a couple of workarounds that coerce everything into
compiling, but in my real code things are a little hairier and I
haven't been able to get them to work. I guess what I am hoping to find
out first is what I am doing wrong. Anything obvious?

Using a buggy compiler counts?

V

 

[snorkelman]

In the more complex version of my problem, I think the solution to this
is that instead of writing:

a.foo<0>()

I need to write

a.template foo<0>()

I have to admit that I'm still a little fuzzy on this requirement.
Visual C++ has never seemed to need this, but g++ frequently thinks I
need to insert that keyword to clarify things.

If I follow it correctly, the explanation in the standard implies that
when I write:

a.foo<0>()

the compiler interprets a.foo as a declaration, then it sees a less
than sign and freaks out.

I don't completely get it, but I can apply the fix with aplomb.

|
| markn at ieee dot org
|

 

[Victor Bazarov]

In the more complex version of my problem, I think the solution to
this is that instead of writing:

a.foo<0>()

I need to write

a.template foo<0>()

I have to admit that I'm still a little fuzzy on this requirement.
Visual C++ has never seemed to need this, but g++ frequently thinks I
need to insert that keyword to clarify things.

If I follow it correctly, the explanation in the standard implies that
when I write:

a.foo<0>()

the compiler interprets a.foo as a declaration, then it sees a less
than sign and freaks out.

I don't completely get it, but I can apply the fix with aplomb.

The postfix expression 'a.foo' cannot be interpreted in any other way
than the use of the member template. That means that the opening '<'
cannot be interpreted as a 'less-than' sign. The use of 'template'
keyword apparently helps the compiler in time of need to understand
that 'foo' is a member template. But according to the Standard it is
only necessary if 'a' itself is a dependent expression. It isn't.

V

 

[snorkelman]

But according to the Standard it is

only necessary if 'a' itself is a dependent expression. It isn't.

Yeah, I think the g++ 3.2 problem is probably a compiler error. My
real-world problem was using a dependent expression, and is
appropriately solved by the use of T::template blah-blah.
|
| markn at ieee dot org
|