[C++ PATCH] Fix PR/9154 (a trunk regression)

[Giovanni Bajo]

Wolfgang Bangerth <bangerth@ices.utexas.edu> wrote:

> [Note that I'm poking wildly in the dark because I don't know anything of
> gcc internals, but got the impression that there are really weird cases
for
> this...]

I think that what I'm doing is reasonably safe. When parsing a single *TYPE*
template argument, GCC was looking for either a comma or a > as terminator
for it. With my patch, it also accepts a >>, but emits an error. In all your
examples, you are using ">>" in the middle of a expression (which is a
non-type parameter), so my code does not affect it.

> How about
>
>   template <int> struct A {
>     typedef int I;
>   };
>
>   template <typename T = A< 1>>2 >::I>
>   struct X {};

It works as expected. Added to the testcase as shown above.

Giovanni Bajo



// { dg-do compile }
// Origin: <tilps at hotmail dot com>
// c++/9154: poor error message for ">>" vs "> >" in template argument list

template <class T>
class A {};

template <int N>
void B(void) {}

template <int N = 123>>4>
struct C {};

template <int>      struct D {};
template <typename> struct E {};

E<D< 1>>2 > > E1;

const int x = 0;
E<D< 1>>x > > E2;

template <int> struct F {
  typedef int I;
};

template <typename T = F< 1>>2 >::I>
struct G {};


int main()
{
  B<256 >> 4>();
  A<A<int>> blah;  // { dg-error "" "should be '> >' within" }
}