C++ PATCH for c++/91521 - wrong error with operator->

[Jason Merrill]

On 8/23/19 1:37 PM, Marek Polacek wrote:
> Since r263836 we enter the "a late-specified return type" block in
> grokdeclarator when inner_declarator is null:
> 
>          /* Handle a late-specified return type.  */
>          tree late_return_type = declarator->u.function.late_return_type;
> -       if (funcdecl_p)
> +       if (funcdecl_p
> +       /* This is the case e.g. for
> +          using T = auto () -> int.  */
> +       || inner_declarator == NULL)
>            {
> 
> inner_declarator is null in this testcase too, but here we don't have
> a real trailing return type; it's just an overloaded operator ->.  That
> means that late_return_type is non-null, but is error_mark_node.  In
> that case I think it's not sensible to complain about "trailing return
> type only available" or "function with trailing return type not declared
> with auto".  In this case that made us reject valid code.
> 
> Bootstrapped/regtested on x86_64-linux, ok for trunk and 9?
> 
> 2019-08-23  Marek Polacek  <polacek@redhat.com>
> 
> 	PR c++/91521 - wrong error with operator->.
> 	* decl.c (grokdeclarator): Don't consider error_mark_node a valid
> 	trailing return type.
> 
> 	* g++.dg/parse/operator8.C: New test.
> 
> diff --git gcc/cp/decl.c gcc/cp/decl.c
> index 88aa69ce5de..ea5752a249e 100644
> --- gcc/cp/decl.c
> +++ gcc/cp/decl.c
> @@ -11546,6 +11546,7 @@ grokdeclarator (const cp_declarator *declarator,
>   		      }
>   		  }
>   		else if (late_return_type
> +			 && late_return_type != error_mark_node
>   			 && sfk != sfk_conversion)
>   		  {

What if instead of this change we check for error_mark_node inside the 
block and return without an error?

Jason