Add a test for timers in child isolates.

tools/dom/src/Timer.dart

Index: tools/dom/src/Timer.dart
diff --git a/tools/dom/src/Timer.dart b/tools/dom/src/Timer.dart
index 43a208afe0a43f0a43360e1361ec6cc656f1d36b..9fa22617829bb7960851264f2c22d8ae0dee7ae9 100644
--- a/tools/dom/src/Timer.dart
+++ b/tools/dom/src/Timer.dart
@@ -37,6 +37,8 @@ class _PureIsolateTimer implements Timer {
   final ReceivePort _port = new ReceivePort();
   SendPort _sendPort; // Effectively final.
 
+  static SendPort _SEND_PORT;
+
   _PureIsolateTimer(int milliSeconds, callback, repeating) {
     _sendPort = _port.toSendPort();
     _port.receive((msg, replyTo) {
@@ -45,9 +47,22 @@ class _PureIsolateTimer implements Timer {
       callback(this);
       if (!repeating) _cancel();
     });
-    _HELPER_ISOLATE_PORT.then((port) {
-      port.send([_NEW_TIMER, milliSeconds, repeating], _sendPort);
-    });
+
+    registerTimer() {
+      _SEND_PORT.send([_NEW_TIMER, milliSeconds, repeating], _sendPort);
+    }
+
+    // Tricky part: Future.then is always delayed might prevent
+    // new timer registration if many events come to this isolate.
+    // Therefore, let's resort to then only if really necessary.

[vsm, 2013/03/29 14:37:39]

I'm confused by this.  What was the problem you were hitting in the old code? 
Why wasn't the new time registering?

It doesn't seem like this change should fix anyway (beyond reducing a potential
delay).

[Anton Muhin, 2013/03/29 16:09:15]

(whatever outcome of our discuss be, I need to modify cancel code to behave
similarly).

There was an interesting corner case I found.

Imagine you spawn a timer which pings you every 10 ms and from the handler of
this timer you'd like to spawn another timer.  Alas, timer creation requires
.then which invoke Timer.run which invokes new Timer.

So I'd like to by pass timer creation while I am trying to create a new timer.

Does that sound reasonable?

I've also updates the text.

On 2013/03/29 14:37:39, vsm wrote:

[vsm, 2013/03/29 17:34:36]

I think I'm starting to get this - was the old code hitting an infinite loop on
_HELPER_ISOLATE_PORT.then?

Not entirely clear to me whether this is a bug fix or an optimization from your
comment.

On 2013/03/29 16:09:15, Anton Muhin wrote:

[Anton Muhin, 2013/04/01 14:52:57]

I've updated the comment.  May you, please, tell me if it's sufficient enough?

On 2013/03/29 17:34:36, vsm wrote:

+    if (_SEND_PORT != null) {
+      registerTimer();
+    } else {
+      _HELPER_ISOLATE_PORT.then((port) {
+        _SEND_PORT = port;
+        registerTimer();
+      });
+    }
   }
 
   void cancel() {
@@ -73,12 +88,13 @@ _helperIsolateMain() {
     if (cmd == _NEW_TIMER) {
       final duration = new Duration(milliseconds: msg[1]);
       bool periodic = msg[2];
-      final callback = () { replyTo.send(_TIMER_PING); };
+      ping() { replyTo.send(_TIMER_PING); }
       _TIMER_REGISTRY[replyTo] = periodic ?
-          new Timer.periodic(duration, callback) :
-          new Timer(duration, callback);
+          new Timer.periodic(duration, (_) { ping(); }) :
+          new Timer(duration, ping);
     } else if (cmd == _CANCEL_TIMER) {
-      _TIMER_REGISTRY.remove(replyTo).cancel();
+      Timer timer = _TIMER_REGISTRY.remove(replyTo);
+      if (timer != null) timer.cancel();
     }
   });
 }