enable top-level @JS() getters

lib/src/compiler/code_generator.dart

Index: lib/src/compiler/code_generator.dart
diff --git a/lib/src/compiler/code_generator.dart b/lib/src/compiler/code_generator.dart
index d07bfb06619611d3aa7cb7238e7f19185343f0e9..59cdcca190368a0a0bc51a0879762df44e8f1b24 100644
--- a/lib/src/compiler/code_generator.dart
+++ b/lib/src/compiler/code_generator.dart
@@ -2325,6 +2325,13 @@ class CodeGenerator extends GeneralizingAstVisitor
   }
 
   JS.PropertyAccess _emitTopLevelName(Element e, {String suffix: ''}) {
+    if (e is TopLevelVariableElement &&
+        e.getter != null &&
+        findAnnotation(e.getter, isPublicJSAnnotation) != null) {
+      var annotationName = getAnnotationName(e.getter, isPublicJSAnnotation);
+      var name = js.string(annotationName ?? e.name);
+      return new JS.PropertyAccess(new JS.Identifier('self'), name);

[Jennifer Messerly, 2016/05/13 23:19:17]

Where is "self" coming from?

[Harry Terkelsen, 2016/05/13 23:34:43]

hmm, I've mostly been doing it because dart2js uses self. It is working with the
tests because 'window.self = this' but we'd probably run into issues if someone
set 'self' in a method. Is there a DDC way to refer to the 'window' object in a
way that will never get overwritten?

[Jennifer Messerly, 2016/05/13 23:51:07]

Hmmmmm ... I don't think we have a reliable way to reach "window". Using "this"
at the top level scope should do it. You could inject a temporary: `const self =
this;` at the module level to provide access.
http://www.ecma-international.org/ecma-262/6.0/#sec-global-environment-records

(That said I'm not 100% sure if our ES5 module transform won't mess that pattern
up, but conceptually I think it works :) )

[Harry Terkelsen, 2016/05/16 16:18:13]

Done.

+    }
     String name = getJSExportName(e) + suffix;
     return new JS.PropertyAccess(
         emitLibraryName(e.library), _propertyName(name));
@@ -2538,17 +2545,21 @@ class CodeGenerator extends GeneralizingAstVisitor
       return js.call('dart.#(#, #)', [name, jsTarget, args]);
     }
 
-    jsTarget = new JS.PropertyAccess(jsTarget, memberName);
-
-    if (typeArgs != null) jsTarget = new JS.Call(jsTarget, typeArgs);
-
     if (DynamicInvoke.get(node.methodName)) {
       // This is a dynamic call to a statically known target. For example:
       //     class Foo { Function bar; }
       //     new Foo().bar(); // dynamic call
-      return js.call('dart.dcall(#, #)', [jsTarget, args]);
+      if (typeArgs != null) {

[Jennifer Messerly, 2016/05/13 23:19:16]

just curious about the reason for this change?

Overall I think it's a good idea :) -- the "dynamic call to a static target"
case was a bit flaky. But just curious what particular case you hit.

[Harry Terkelsen, 2016/05/13 23:34:43]

Basically because of this line:
https://github.com/dart-lang/dev_compiler/blob/master/test/codegen/lib/html/j...

[Jennifer Messerly, 2016/05/13 23:51:06]

Thanks, makes sense :)

+        return js.call('dart.dgsend(#, #, #, #)',

[Jennifer Messerly, 2016/05/13 23:19:17]

this code looks copy+pasted, can we share it with _emitMethodCall somehow?

[Harry Terkelsen, 2016/05/13 23:34:43]

I merged it with the above

[Jennifer Messerly, 2016/05/13 23:51:06]

Looks great :)

+            [jsTarget, new JS.ArrayInitializer(typeArgs), memberName, args]);
+      } else {
+        return js.call('dart.dsend(#, #, #)', [jsTarget, memberName, args]);
+      }
     }
 
+    jsTarget = new JS.PropertyAccess(jsTarget, memberName);
+    if (typeArgs != null) jsTarget = new JS.Call(jsTarget, typeArgs);
+
     return new JS.Call(jsTarget, args);
   }