[d8] Use ValueSerializer for postMessage (instead of ad-hoc serializer)

test/mjsunit/regress/regress-crbug-514081.js

Index: test/mjsunit/regress/regress-crbug-514081.js
diff --git a/test/mjsunit/regress/regress-crbug-514081.js b/test/mjsunit/regress/regress-crbug-514081.js
index 1acd8315cd4e7faf9188697140f0d3ce49574ec7..af31f4b9c6ceed6e74e5deea65f03d0b7b12e19b 100644
--- a/test/mjsunit/regress/regress-crbug-514081.js
+++ b/test/mjsunit/regress/regress-crbug-514081.js
@@ -4,10 +4,15 @@
 
 if (this.Worker) {
   var __v_7 = new Worker('onmessage = function() {};');
+  var e;
   try {
     var ab = new ArrayBuffer(2147483648);

[Michael Achenbach, 2017/02/07 10:17:27]

Still seeing flakes of this on windows, e.g.:
https://build.chromium.org/p/client.v8/builders/V8%20Win64/builds/15656/steps...

Either passes fast or times out. Could you make this more deterministic?
Creating a too large array buffer might interfere with other test cases using
too much RAM in the same time. Also, the exceptions seem to be deliberately
non-deterministic here. You can't end up getting both, right?

Could we mock out the memory limits here with a d8 flag passed only in this
test? 

-    // If creating the ArrayBuffer succeeded, then postMessage should fail.
-    assertThrows(function() { __v_7.postMessage(ab); });
+    try {
+      __v_7.postMessage(ab);
+    } catch (e) {
+      // postMessage failed, should be a DataCloneError message.
+      assertContains('cloned', e.message);
+    }
   } catch (e) {
     // Creating the ArrayBuffer failed.
     assertInstanceof(e, RangeError);