Cppcheck crash; I wonder how this template usage works

[Daniel Marjamäki]

Cppcheck crashes for this code:

template < class Grid >
void checkFatherLevel ( Grid & grid )
{
    for ( int level = 0 ; level <= grid . maxLevel ( ) ; ++ level )
    {
        typedef typename Grid :: LevelGridView GridView ;
        typedef typename GridView :: template Codim < 0 > :: Iterator Iterator ;
        typedef typename Iterator :: Entity Entity ;

        GridView gv = grid . levelGridView ( level ) ;
        Iterator it = gv . template begin < 0 > ( ) ;
        Iterator end = gv . template end < 0 > ( ) ;
            if ( level == 0 )
            {
            }
    }
}

I wonder do you know what the template keyword does inside the template code?

What would the proper AST be for the expression it = gv . template begin < 0 > ( ).

[Daniel Marjamäki]

The clang AST for that expression:

          | `-VarDecl 0x18bab10 <col:9, col:50> col:18 end 'Iterator':'typename GridView::template Codim<0>::Iterator' cinit
          |   `-CallExpr 0x18bac60 <col:24, col:50> '<dependent type>'
          |     `-CXXDependentScopeMemberExpr 0x18babd8 <col:24, col:46> '<dependent type>' lvalue .end
          |       `-DeclRefExpr 0x18bab78 <col:24> 'GridView':'typename Grid::LevelGridView' lvalue Var 0x1894ba8 'gv' 'GridView':'typename Grid::LevelGridView'

I guess if we remove template tokens in such code we will get a corresponding AST in Cppcheck.

Does anybody disagree about that?

[Daniel Marjamäki]

I will implement that simplification.. please feel free to comment if you think that will be wrong.

[Paul Fultz]

I wonder do you know what the template keyword does inside the template code?

This is to disambiguate the function call. Since the compiler doesnt know if begin is a variable or a function. By default, it assumes begin is a variable so < and > are treated as operators instead of template brackets.

Does anybody disagree about that?

I dont necessarily disagree, but think we should preserve this information somehow as this seems like it would be useful in createLinks.

[Daniel Marjamäki]

I dont necessarily disagree, but think we should preserve this information somehow as this seems like it would be useful in createLinks.

ok good to know.. then I will try to add a attribute instead so this information is kept..

[Robert Reif]

We already skip over the template token in the template simplifier.