Implement on demand quad rasterization for PicturePiles.

cc/test/layer_tree_test_common.cc

Index: cc/test/layer_tree_test_common.cc
diff --git a/cc/test/layer_tree_test_common.cc b/cc/test/layer_tree_test_common.cc
index e9be315ff0d790709fe6ecfbe2af67aacb63e4c6..d73367007b494a11458e670888c8a98449370a55 100644
--- a/cc/test/layer_tree_test_common.cc
+++ b/cc/test/layer_tree_test_common.cc
@@ -316,10 +316,13 @@ void ThreadedTest::endTest()
 {
     // For the case where we endTest during beginTest(), set a flag to indicate that
     // the test should end the second beginTest regains control.
-    if (m_beginning)
+    if (m_beginning) {
         m_endWhenBeginReturns = true;
-    else
-        proxy()->MainThread()->PostTask(base::Bind(&ThreadedTest::realEndTest, m_mainThreadWeakPtr));
+    } else {

[danakj, 2013/03/21 02:34:42]

else if (proxy())

[Leandro Graciá Gil, 2013/03/21 16:42:48]

Done in a separate CL.

+        // Racy timeouts and explicit endTest calls might have cleaned up the tree host.

[nduca, 2013/03/21 02:22:48]

Erm,.... this is a separate patch. You should put that up separately.

[Leandro Graciá Gil, 2013/03/21 03:35:50]

Ok, but the problem was exposed by the newly introduced pixel test. If you're
fine with uploading the fix alone I can land it in a separate patch.

[danakj, 2013/03/21 04:33:39]

I was wondering about this.. does this mean the pixel test is finishing at the
time limit of the timeout? Seems like its going to time out on the bots a lot if
that's the case. Also seems incredibly slow.. I agree it could be a separate CL
(and if that's true, then maybe we need a longer time out on the pixel tests :(
?)

[Leandro Graciá Gil, 2013/03/21 16:42:48]

The new pixel test had flaky crashes, apparently because of a race condition
between timeout calling endTest and the test finishing and calling this same
method. It was easy to reproduce by running the test about 10 times. However, in
case it helps, it was very hard to reproduce when attached to gdb. I had to set
it to repeat 10000 times and go for lunch in order to get a flake in that case.

I'll be moving this into a separate CL as suggested.

+        if (proxy())
+            proxy()->MainThread()->PostTask(base::Bind(&ThreadedTest::realEndTest, m_mainThreadWeakPtr));
+    }
 }
 
 void ThreadedTest::endTestAfterDelay(int delayMilliseconds)