OK.
On Fri, Apr 19, 2019 at 1:43 AM Jakub Jelinek <ja...@redhat.com> wrote:
>
> Hi!
>
> On the following testcase, there is a type template argument, for which
> we create a decl (decl == parm in that case) and pushdecl it. But, if
> the pushdecl detects a redeclaration in that scope, duplicate_decls will
> ggc_free the decl passed to it and we then add that ggc_freed tree into the
> template parameter TREE_LIST.
>
> The following patch fixes that by using the result value from pushdecl.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
>
> 2019-04-19 Jakub Jelinek <ja...@redhat.com>
>
> PR c++/90138
> * pt.c (process_template_parm): Set decl to pushdecl result. If
> !is_non_type, also set parm to that.
>
> * g++.dg/template/pr90138.C: New test.
>
> --- gcc/cp/pt.c.jj 2019-04-18 12:16:47.000000000 +0200
> +++ gcc/cp/pt.c 2019-04-18 14:12:36.765494173 +0200
> @@ -4433,7 +4433,9 @@ process_template_parm (tree list, locati
> process_template_parm could fail. */
> tree reqs = finish_shorthand_constraint (parm, constr);
>
> - pushdecl (decl);
> + decl = pushdecl (decl);
> + if (!is_non_type)
> + parm = decl;
>
> /* Build the parameter node linking the parameter declaration,
> its default argument (if any), and its constraints (if any). */
> --- gcc/testsuite/g++.dg/template/pr90138.C.jj 2019-04-18 14:58:06.035564846
> +0200
> +++ gcc/testsuite/g++.dg/template/pr90138.C 2019-04-18 14:57:27.056206214
> +0200
> @@ -0,0 +1,5 @@
> +// PR c++/90138
> +
> +template <, typename T, typename typename, typename T> // { dg-error
> "expected" }
> +struct S; // { dg-error "no default" }
> +// { dg-error "two or more" "" { target *-*-* } .-2 }
>
> Jakub
