FirstMeaningfulPaint UMA

chrome/browser/page_load_metrics/page_load_metrics_browsertest.cc

Index: chrome/browser/page_load_metrics/page_load_metrics_browsertest.cc
diff --git a/chrome/browser/page_load_metrics/page_load_metrics_browsertest.cc b/chrome/browser/page_load_metrics/page_load_metrics_browsertest.cc
index 6f6a8015d7cb5de958d574bb584c45dcfce36fff..dd2da42de6cd6901c26dfacc67b160d1693b2c6f 100644
--- a/chrome/browser/page_load_metrics/page_load_metrics_browsertest.cc
+++ b/chrome/browser/page_load_metrics/page_load_metrics_browsertest.cc
@@ -377,3 +377,49 @@ IN_PROC_BROWSER_TEST_F(MetricsWebContentsObserverBrowserTest,
   histogram_tester_.ExpectTotalCount(
       internal::kHistogramAbortClientRedirectBeforeCommit, 1);
 }
+
+IN_PROC_BROWSER_TEST_F(MetricsWebContentsObserverBrowserTest,
+                       FirstMeaningfulPaintRecorded) {
+  ASSERT_TRUE(embedded_test_server()->Start());
+
+  ui_test_utils::NavigateToURL(browser(),
+                               embedded_test_server()->GetURL("/title1.html"));
+
+  // Wait until a FMP is reported, by polling the histogram counter.

[Charlie Harrison, 2016/08/08 12:35:44]

Can this be avoided? I would slightly prefer using
content::ExecuteScriptAndExtractBool + setTimeout to use a one-shot delay that
we know will have triggered FMP.

Note: NavigateToURL should wait until the load event, which will probably (at
least in this case) be after the true FMP.

[Kunihiko Sakamoto, 2016/08/09 02:32:49]

I'm afraid that that will be a source of flakiness, unless we use a pretty long
time margin. The longer timeout we use, the slower the test will be...

btw, is it possible to delay ExecuteScriptAndExtractBool's return by using
setTimeout?

[Charlie Harrison, 2016/08/09 02:36:43]

Yeah, that was my idea. The content API call will wait for JS to post a message
back to it (codesearch for usage examples). So you can e.g. do:

ExecuteScriptAndExtractBool("setTimeout(function() {
window.domAutomationController.send(true); }, kNetworkIdleTime);");

Hopefully that should be good enough to unflake without resorting to spinning
the message loop.

[Kunihiko Sakamoto, 2016/08/09 03:31:14]

Ah, now I got your idea.
Done, thanks!

+  while (histogram_tester_.GetAllSamples(
+      internal::kHistogramFirstMeaningfulPaintStatus).empty()) {
+    base::RunLoop run_loop;
+    base::ThreadTaskRunnerHandle::Get()->PostDelayedTask(
+        FROM_HERE, run_loop.QuitClosure(),
+        base::TimeDelta::FromMilliseconds(100));
+    run_loop.Run();
+  }
+
+  NavigateToUntrackedUrl();
+  histogram_tester_.ExpectUniqueSample(
+      internal::kHistogramFirstMeaningfulPaintStatus,
+      internal::FMP_RECORDED, 1);
+  histogram_tester_.ExpectTotalCount(
+      internal::kHistogramFirstMeaningfulPaint, 1);
+  histogram_tester_.ExpectTotalCount(
+      internal::kHistogramParseStartToFirstMeaningfulPaint, 1);
+}
+
+IN_PROC_BROWSER_TEST_F(MetricsWebContentsObserverBrowserTest,
+                       FirstMeaningfulPaintNotRecorded) {
+  ASSERT_TRUE(embedded_test_server()->Start());
+
+  ui_test_utils::NavigateToURL(browser(),
+                               embedded_test_server()->GetURL("/title1.html"));
+
+  // Navigate away before a FMP is reported.
+  NavigateToUntrackedUrl();
+
+  histogram_tester_.ExpectUniqueSample(
+      internal::kHistogramFirstMeaningfulPaintStatus,
+      internal::FMP_DID_NOT_REACH_NETWORK_STABLE, 1);
+  histogram_tester_.ExpectTotalCount(
+      internal::kHistogramFirstMeaningfulPaint, 0);
+  histogram_tester_.ExpectTotalCount(
+      internal::kHistogramParseStartToFirstMeaningfulPaint, 0);
+}