[gcjx] Patch: FYI: another float->int buglet

[gcjx] Patch: FYI: another float->int buglet

[Tom Tromey]

I'm checking this in on the gcjx branch.

This fixes a float->int buglet I introduced earlier.
We don't want to convert the min/max "result" values to the floating
point type.

Tom

Index: ChangeLog
from  Tom Tromey  <tromey@redhat.com>
	* lower.cc (handle_convert): Keep non-converted copy of min/max
	for results.

Index: lower.cc
===================================================================
RCS file: /cvs/gcc/gcc/gcc/java/Attic/lower.cc,v
retrieving revision 1.1.2.21
diff -u -r1.1.2.21 lower.cc
--- lower.cc 11 Oct 2005 18:09:47 -0000 1.1.2.21
+++ lower.cc 11 Oct 2005 21:35:08 -0000
@@ -1648,8 +1648,8 @@
     }
   assert (TREE_TYPE (min_tree) == dest_tree);
 
-  min_tree = convert (TREE_TYPE (expr), min_tree);
-  max_tree = convert (TREE_TYPE (expr), max_tree);
+  tree flt_min_tree = convert (TREE_TYPE (expr), min_tree);
+  tree flt_max_tree = convert (TREE_TYPE (expr), max_tree);
 
   expr = save_expr (expr);
 
@@ -1667,13 +1667,13 @@
   // Now build:
   //   if (val <= (FROM) min) return min; else [ the above ]
   cond = build3 (COND_EXPR, dest_tree,
-		 build2 (LE_EXPR, type_jboolean, expr, min_tree),
+		 build2 (LE_EXPR, type_jboolean, expr, flt_min_tree),
 		 min_tree, cond);
 
   // Finally:
   //   if (val >= (FROM) max) return max; else [ the above ]
   return build3 (COND_EXPR, dest_tree,
-		 build2 (GE_EXPR, type_jboolean, expr, max_tree),
+		 build2 (GE_EXPR, type_jboolean, expr, flt_max_tree),
 		 max_tree, cond);
 }