Task manager should support Idle Wakeups on Windows

chrome/browser/task_management/task_manager_browsertest.cc

Index: chrome/browser/task_management/task_manager_browsertest.cc
diff --git a/chrome/browser/task_management/task_manager_browsertest.cc b/chrome/browser/task_management/task_manager_browsertest.cc
index 3b34761669eab8635e33dea2a535a1238608eefb..698dc2e3537618e68a237e376583e615aee9eb18 100644
--- a/chrome/browser/task_management/task_manager_browsertest.cc
+++ b/chrome/browser/task_management/task_manager_browsertest.cc
@@ -850,6 +850,37 @@ IN_PROC_BROWSER_TEST_F(TaskManagerBrowserTest, JSHeapMemory) {
   ASSERT_NO_FATAL_FAILURE(WaitForTaskManagerRows(1, MatchTab("title1.html")));
 }
 
+// Checks that task manager counts idle wakeups.
+IN_PROC_BROWSER_TEST_F(TaskManagerBrowserTest,
+                       IdleWakeups) {

[ncarter (slow), 2016/07/28 21:37:27]

I think this test is a good start. By its nature it might turn out to be flaky
and have {platform|trybot|configuration}-dependent behavior, so I expect some
iteration here.

[stanisc, 2016/08/01 22:34:27]

OK. I've implemented changes suggested below.

+  ShowTaskManager();
+  model()->ToggleColumnVisibility(ColumnSpecifier::IDLE_WAKEUPS);
+
+  ui_test_utils::NavigateToURL(browser(), GetTestURL());
+
+  std::string test_js =
+    "function myWait(n) {\n"

[ncarter (slow), 2016/07/28 21:37:27]

I would probably make this an infinite setTimeout chain. TaskManager only
samples every 1 sec so running for 1000 1ms intervals might not quite cut it --
we might be thrown awry, say, by something like an unlucky gc pause on the 999th
iteration.

We can rely on WaitForTaskManagerStatToExceed not running forever (it has a test
timeout, unless you run it with a debugger attached); and it should still be
possible to cleanly close a tab that's waking up every 1ms.

[stanisc, 2016/08/01 22:34:27]

Done.

+    "  if (n < 1000) {\n"
+    "    setTimeout(function() { myWait(n + 1); }, 1)\n"
+    "  } else {\n"
+    "    window.domAutomationController.send(\"okay\");\n"
+    "  }\n"
+    "}\n"
+    "myWait(0);\n";
+
+  std::string ok;
+  ASSERT_TRUE(content::ExecuteScriptAndExtractString(
+      browser()->tab_strip_model()->GetActiveWebContents(), test_js, &ok));

[ncarter (slow), 2016/07/28 21:37:27]

ExecuteScriptAndExtractString will block until
window.domAutomationController.send() completes, so we don't start looking at
the TaskManager stats until after the js has finished running.

If you move the window.domAutomationController.send('okay') so that it's after
the myWait(0) call, they'll run concurrently.

[stanisc, 2016/08/01 22:34:27]

Done.

+  ASSERT_EQ("okay", ok);
+
+  // The script above should trigger a lot of idle wakeups - up to 1000 per
+  // second. Let's make sure we get at least 100 (in case the test runs slow).
+  ASSERT_NO_FATAL_FAILURE(WaitForTaskManagerStatToExceed(
+      MatchTab("title1.html"), ColumnSpecifier::IDLE_WAKEUPS, 100));
+  ASSERT_NO_FATAL_FAILURE(WaitForTaskManagerRows(1, MatchAnyTab()));
+  ASSERT_NO_FATAL_FAILURE(WaitForTaskManagerRows(1, MatchTab("title1.html")));
+}
+
 // Checks that task manager counts utility process JS heap size.
 IN_PROC_BROWSER_TEST_F(TaskManagerUtilityProcessBrowserTest,
                        UtilityJSHeapMemory) {