[PATCH] Fix -Wswitch (ie. PR c/4475 and PR c++/3780)

[Jakub Jelinek]

On Mon, Feb 04, 2002 at 09:10:21AM -0800, Richard Henderson wrote:
> On Mon, Feb 04, 2002 at 03:42:34PM +0100, Jakub Jelinek wrote:
> > The problem is that SWITCH_COND may be already converted internally by the
> > compiler to some other type (such as int) and thus checking if it is
> > ENUMERAL_TYPE in expand_end_case doesn't work.
> 
> Where does this happen?  I don't see it.  I don't see why it 
> should happen either.

c-typeck.c (c_start_case):
          type = TYPE_MAIN_VARIANT (TREE_TYPE (exp));

          if (warn_traditional && !in_system_header
              && (type == long_integer_type_node
                  || type == long_unsigned_type_node))
            warning ("`long' switch expression not converted to `int' in ISO C");

          exp = default_conversion (exp);		<---- here
          type = TREE_TYPE (exp);

cp/semantics.c (finish_switch_cond):
      if (cond != error_mark_node)
        {
          cond = default_conversion (cond);		<---- and here
          cond = fold (build1 (CLEANUP_POINT_EXPR, TREE_TYPE (cond), cond));
        }

      type = TREE_TYPE (cond);
      index = get_unwidened (cond, NULL_TREE);

Checking what gcc-2.96-RH does, it was calling default_conversion as well,
but argument for expand_end_case was passed directly from c-parse.y (thus
the original one, not modified):
        | SWITCH '(' expr ')'
                { stmt_count++;
                  emit_line_note ($<filename>-1, $<lineno>0);
                  c_expand_start_case ($3);	<- default_conversion was done in this function
                  position_after_white_space (); }
          lineno_labeled_stmt
                { expand_end_case ($3); }

	Jakub