Re: [PATCH] PR c++/26693

["Manuel López-Ibáñez" <lopezibanez at gmail dot com>]

2008/10/29 Dodji Seketeli <dodji@redhat.com>:
> +
> +      /* When a typedef statement appears in a class/struct, make sure a
> +         'typedef variant type' node is associated to that typedef
> statement.
> +        This will help to later be able to check the access of the typedef
> name.
> +        Please read the comment of the function clone_underlying_type in
> +        c-decl.c to learn more about how to represent the type of
> +        a typedef statement.  */
> +      if (decl_context == FIELD
> +         && !uses_template_parms (type))
> +       {
> +         if (DECL_IS_BUILTIN (decl))
> +           {
> +             if (TYPE_NAME (type) == 0)
> +               TYPE_NAME (type) = decl;
> +           }
> +         else
> +           {
> +             DECL_ORIGINAL_TYPE (decl) = type;
> +             type = build_variant_type_copy (type);
> +             TYPE_STUB_DECL (type) = TYPE_STUB_DECL (DECL_ORIGINAL_TYPE
> (decl));
> +             TYPE_NAME (type) = decl;
> +             TREE_TYPE (decl) = type;
> +           }
> +       }
> +
> +

Couldn't you abstract out this code and clone_underlying_type into a
function shared by C/C++? They look almost the same to me. Actually, I
don't understand why c-decl.c does not use TYPE_STUB_DECL.

Moreover, this actually means that we track typedefs, don't we? That
could improve so much our diagnostics in C++!


> @@ -8921,22 +8921,41 @@ tsubst (tree t, tree args, tsubst_flags_t complain,
> tree in_decl)
>       && TYPE_NAME (t) != TYPE_MAIN_DECL (t))
>     {
>       tree decl = TYPE_NAME (t);
> -
> +

Whitespace

> diff --git a/gcc/testsuite/g++.dg/template/typedef11.C
> b/gcc/testsuite/g++.dg/template/typedef11.C
> index db7a1d3..6e71729 100644
> --- a/gcc/testsuite/g++.dg/template/typedef11.C
> +++ b/gcc/testsuite/g++.dg/template/typedef11.C

Is this patch against trunk? I don't see this testcase in FSF's trunk.

Cheers,

Manuel.