Allow error-recovery in top-level nodes.

dart/frog/leg/scanner/parser.dart

Index: dart/frog/leg/scanner/parser.dart
diff --git a/dart/frog/leg/scanner/parser.dart b/dart/frog/leg/scanner/parser.dart
index b276a970ce9003ee0f1544c93313e62e40485f48..506107eecbd4e21c7ad0ff4ac7c1f4e0d71ca120 100644
--- a/dart/frog/leg/scanner/parser.dart
+++ b/dart/frog/leg/scanner/parser.dart
@@ -305,13 +305,19 @@ class Parser<L extends Listener> {
     }
     token = parseIdentifier(previous);
     bool isField;
-    if (optional('(', token)) {
-      isField = false;
-    } else if (optional('=', token) || optional(';', token)) {
-      isField = true;
-    } else {
-      token = listener.unexpected(token);
-    }
+    do {

[kasperl, 2011/11/09 12:59:18]

Why do you prefer do { } while (true) over while (true) { }?

[ahe, 2011/11/09 13:07:55]

No particular reason. I was thinking about writing:

do {
  ...
} while (isField !== null);

The current version would be better with while (true) { }.

+      // Loop to allow the listener to rewrite the token stream to
+      // make the parser happy.
+      if (optional('(', token)) {
+        isField = false;
+        break;
+      } else if (optional('=', token) || optional(';', token)) {
+        isField = true;
+        break;
+      } else {
+        token = listener.unexpected(token);
+      }
+    } while (true);
     while (token !== null &&
            token.kind !== LBRACE_TOKEN &&
            token.kind !== SEMICOLON_TOKEN) {