Implement -Wimplicit-fallthrough (version 9)

[Jason Merrill]

On 09/21/2016 08:44 AM, Marek Polacek wrote:
> @@ -10733,12 +10758,35 @@ cp_parser_expression_statement (cp_parser* parser, tree in_statement_expr)
>       statement.  */
>    if (cp_lexer_next_token_is_not (parser->lexer, CPP_SEMICOLON))
>      {
> +      /* This might be attribute fallthrough.  */
> +      if (cp_lexer_next_token_is_keyword (parser->lexer, RID_ATTRIBUTE))
>  	{
> +	  tree attr = cp_parser_gnu_attributes_opt (parser);
> +	  if (maybe_attribute_fallthrough_p (attr))
> +	    statement = build_call_expr_internal_loc (token->location,
> +						      IFN_FALLTHROUGH,
> +						      void_type_node, 0);
> +	  else if (cp_lexer_next_token_is (parser->lexer, CPP_SEMICOLON))
> +	    {
> +	      cp_parser_error (parser, "only attribute %<fallthrough%> "
> +			       "can be applied to a null statement");
> +	      return error_mark_node;
> +	    }
> +	  else
> +	    {
> +	      cp_parser_error (parser, "expected primary-expression");
> +	      return error_mark_node;
> +	    }
> +	}

Attributes that we don't handle should be warnings, not errors.  Like we 
do for C++11 attributes in cp_parser_statement, we should first parse 
attributes, then parse the expression-statement, then give any 
appropriate warnings.

Jason