revised patch for compare1.c and cast-qual-1.c test failures

[Gavin Romig-Koch <gavin at cygnus dot com>]

Zack Weinberg writes:
 > 1999-10-27 12:05 -0700  Zack Weinberg  <zack@bitmover.com>
 > 
 > 	* c-parse.in (primary: IDENTIFIER): When parsing an
 > 	enumeration constant, generate ((enum foo) 12) not ((int) 12).
 > 	* c-typeck.c (build_conditional_expr): If -Wsign-compare, warn
 > 	about (expr ? signed : unsigned).
 > 	(build_c_cast): Strip only one level of POINTER_TYPE nodes
 > 	to check for mismatched qualifiers.
 > 
 > 	* c-common.c (check_case_value): Use STRIP_NOPS not
 > 	STRIP_TYPE_NOPS. 
 > 	* c-decl.c (grokdeclarator, build_enumerator): Likewise.

OK, except for the follow:

 > Index: c-parse.in
 > --- c-parse.in	1999/10/26 18:13:37	1.26
 > +++ c-parse.in	1999/10/27 19:05:06
 > @@ -786,10 +786,9 @@ end ifobjc
 >  
 >  		  if (TREE_CODE ($$) == CONST_DECL)
 >  		    {
 > -		      $$ = DECL_INITIAL ($$);
 > -		      /* This is to prevent an enum whose value is 0
 > -			 from being considered a null pointer constant.  */
 > -		      $$ = build1 (NOP_EXPR, TREE_TYPE ($$), $$);
 > +		      /* Preserve the type of an enumeration constant.  */
 > +		      $$ = build1 (NOP_EXPR, TREE_TYPE ($$),
 > +				   DECL_INITIAL ($$));
 >  		      TREE_CONSTANT ($$) = 1;
 >  		    }
 >  		}

I think the old comment contained useful information that shouldn't
be dropped.


 > ===================================================================
 > Index: c-typeck.c
 > --- c-typeck.c	1999/10/27 05:32:42	1.42
 > +++ c-typeck.c	1999/10/27 19:05:08
 > @@ -3653,15 +3673,10 @@ build_c_cast (type, expr)
 >  	  && TREE_CODE (type) == POINTER_TYPE
 >  	  && TREE_CODE (otype) == POINTER_TYPE)
 >  	{
 > -	  /* Go to the innermost object being pointed to.  */
 > -	  tree in_type = type;
 > -	  tree in_otype = otype;
 > -
 > -	  while (TREE_CODE (in_type) == POINTER_TYPE)
 > -	    in_type = TREE_TYPE (in_type);
 > -	  while (TREE_CODE (in_otype) == POINTER_TYPE)
 > -	    in_otype = TREE_TYPE (in_otype);
 > -	  
 > +	  /* Strip one level of pointerness.  */
 > +	  tree in_type = TREE_TYPE (type);
 > +	  tree in_otype = TREE_TYPE (otype);
 > +
 >  	  if (TYPE_QUALS (in_otype) & ~TYPE_QUALS (in_type))
 >  	    /* There are qualifiers present in IN_OTYPE that are not
 >  	       present in IN_TYPE.  */

I think we need to be warning about loss of qualifiers at every level,
not just the first or the last.

                                             -gavin...