fix readability regressions from previous block-scope CL

lib/src/codegen/js_codegen.dart

Index: lib/src/codegen/js_codegen.dart
diff --git a/lib/src/codegen/js_codegen.dart b/lib/src/codegen/js_codegen.dart
index 2e731517e2ac5d863f126823f623b3577a175d92..3b421f4e912f79c3d799512b4a9f32ca311efe4d 100644
--- a/lib/src/codegen/js_codegen.dart
+++ b/lib/src/codegen/js_codegen.dart
@@ -1767,9 +1767,9 @@ class JSCodegenVisitor extends GeneralizingAstVisitor with ClosureAnnotator {
   @override
   JS.Block visitBlockFunctionBody(BlockFunctionBody node) {
     var initArgs = _emitArgumentInitializers(node.parent);
-    var block = visitBlock(node.block);
-    if (initArgs != null) return new JS.Block([initArgs, block]);
-    return block;
+    var stmts = _visitList(node.block.statements) as List<JS.Statement>;
+    if (initArgs != null) stmts.insert(0, initArgs);
+    return new JS.Block(stmts);
   }
 
   @override
@@ -2932,7 +2932,8 @@ class JSCodegenVisitor extends GeneralizingAstVisitor with ClosureAnnotator {
       }
     }
 
-    body.add(_visit(node.body));
+    body.add(
+        new JS.Block(_visitList(node.body.statements) as List<JS.Statement>));

[Leaf, 2015/09/09 21:29:45]

Will JS be ok with this hoisting if either the exception parameter or the stack
trace parameter share the same name as a variable declared in the list of
statements being hoisted out?

[Jennifer Messerly, 2015/09/09 21:48:39]

JS wouldn't care, unless we end up desugaring (_catchParameter != null or
node.stackTraceParameter != null) ... but in that case, we have a duplicate
declaration.

If Dart really supports having a local variable that shadows and the exception
and/or stack trace (ugh, why?!?!) then we'll need to fix it, or make it an error
(because that seems like pretty useless behavior) :|.

Assuming that behavior is specified, we've had this bug forever though, until I
"fixed" it unintentionally last Friday. So I'd be happiest to land this and then
investigate that issue in a follow up, if that works for you. I think the fix
will be unrelated: to avoid a large readability regression, we'll need to do
some smarts to detect the conflict specifically, and that's unlike the other
places. Or, we need to change how variable declaration is generated in this case
to make it be an assignment.

[Jennifer Messerly, 2015/09/09 22:02:57]

confirmed that it's spec'd that way.

     _catchParameter = savedCatch;
     return _statement(body);
   }