Re: [PATCH C/C++] Fix some diagnostics problems

[Jason Merrill <jason@redhat.com>]

On 06/08/2010 12:10 AM, Shujing Zhao wrote:
> 	* cp-tree.h (expr_list): New type.

Please call this expr_list_kind so people don't think it's a TREE_LIST. 
  Or don't bother giving it a name and just pass it as an int.

> 	(impl_conv_rhs): New type.

Why is this different in name and capitalization from impl_conv in the C 
front end?

> 	* decl.c (bad_spec): New enum.

This enum isn't telling us which bad specs we're dealing with, so the 
name is misleading.  Call it something like "bad_spec_place" or, as 
above, don't give it a name.

> +	  case EL_FUNC_CAST:
> +	    permerror (input_location, "function cast expression "
> +	    			       "list treated as compound expression");

"functional cast", not "function cast".

>        if (TREE_CODE (decl) == RESULT_DECL)
> -	init = build_x_compound_expr_from_list (init,
> -						"return value initializer");
> +	init = build_x_compound_expr_from_list (init, EL_RETURN_VAL_INIT);

I believe this code is a relic of the old named return value extension, 
and can be removed along with EL_RETURN_VAL_INIT.

While we're messing with these messages I'd also like to reword them a 
bit: rather than

"initializer expression list treated as compound expression"

let's say

"expression list treated as compound expression in initializer"

and so on.

> +		    case ICR_DEFAULT_ARGUMENT:
> +		      error ("cannot convert %qT to %qT in default argument",
> +			     rhstype, type);
> +		    case ICR_ARGPASS:
> +		      error ("cannot convert %qT to %qT in argument passing",
> +			     rhstype, type);
> +		      break;

These cases are unreachable, as they are covered by the previous if 
(fndecl) test.  Either remove the if (fndecl) message and use 
parmnum/fndecl in these cases, or just use gcc_unreachable() in these cases.

> +		      error ("cannot convert %qT to %qT in converting",
> +			     rhstype, type);

Drop "in converting".

> +	    case ICR_DEFAULT_ARGUMENT:
> +	      warning (OPT_Wmissing_format_attribute,
> +		       "default argument might be a candidate "
> +		       "for a format attribute");

"parameter %d of %qD might be [...]", parmnum, fndecl

> +	      break;
> +	    case ICR_CONVERTING:
> +	      warning (OPT_Wmissing_format_attribute,
> +		       "converting might be a candidate "
> +		       "for a format attribute");

"target of conversion might be [...]"

> +	      break;
> +	    case ICR_INIT:
> +	      warning (OPT_Wmissing_format_attribute,
> +		       "initialization might be a candidate "
> +		       "for a format attribute");

"target of initialization might be..."

> +	      break;
> +	    case ICR_ARGPASS:
> +	      warning (OPT_Wmissing_format_attribute,
> +		       "argument passing might be a candidate "
> +		       "for a format attribute");
> +	      break;

Same as ICR_DEFAULT_ARGUMENT.

> +	    case ICR_RETURN:
> +	      warning (OPT_Wmissing_format_attribute,
> +		       "return might be a candidate for a format attribute");
> +	      break;

"return type might..."

> +	    case ICR_ASSIGN:
> +	      warning (OPT_Wmissing_format_attribute,
> +		       "assignment might be a candidate "
> +		       "for a format attribute");

"left-hand side of assignment might..."

Jason