Fix 4182: When inlining StringBase_get_length, we cannot assume that all receivers where subclasse…

runtime/vm/flow_graph_optimizer.cc

Index: runtime/vm/flow_graph_optimizer.cc
===================================================================
--- runtime/vm/flow_graph_optimizer.cc	(revision 9826)
+++ runtime/vm/flow_graph_optimizer.cc	(working copy)
@@ -298,6 +298,11 @@
   }
 
   if (recognized_kind == MethodRecognizer::kStringBaseLength) {
+    // All bets are off if we run in production mode, i.e., at one call site

[regis, 2012/07/23 21:38:57]

"if we run in production mode"
Can you do more if we run in checked mode?
The comment is not clear.

[srdjan, 2012/07/23 21:40:41]

Changed comment: it can happen in any mode. Just saying this is not
StringBase_get_length if there are multiple targets.

+    // we may have StringBase_get_length as well as ImmutableArray_get_length.
+    if (!HasOneTarget(ic_data)) {
+      return false;
+    }
     ASSERT(HasOneTarget(ic_data));
     LoadVMFieldComp* load = new LoadVMFieldComp(
         comp->InputAt(0),