Add cross origin to Blink-driven preconnect

chrome/browser/net/predictor_browsertest.cc

Index: chrome/browser/net/predictor_browsertest.cc
diff --git a/chrome/browser/net/predictor_browsertest.cc b/chrome/browser/net/predictor_browsertest.cc
index c4ebc559fd76ea783fa357a164c99dbea33be3f7..7e7bebcbb01732ead5d73bf530ce7b1479434589 100644
--- a/chrome/browser/net/predictor_browsertest.cc
+++ b/chrome/browser/net/predictor_browsertest.cc
@@ -383,5 +383,29 @@ IN_PROC_BROWSER_TEST_F(PredictorBrowserTest, PreconnectAndUse) {
   EXPECT_EQ(1u, connection_listener_->GetReadSocketCount());
 }
 
+// Tests that preconnect warms up a crossorigin socket connection to a test
+// server,
+// and that that socket is later used when fetching an anonymous resource.

[mmenke, 2015/07/09 15:55:56]

nit: Fix line wrap (Weird to wrap at a comma, and leave all that whitespace)

[Yoav Weiss, 2015/07/09 16:19:39]

fixed

+// Note: This test uses a data URI to serve the preconnect hint, to make sure
+// that the network stack doesn't just re-use its connection to the test server.
+IN_PROC_BROWSER_TEST_F(PredictorBrowserTest, PreconnectAndUseAnonymous) {
+  GURL preconnect_url = embedded_test_server()->base_url();
+  // First navigation to content with a preconnect hint.
+  std::string preconnect_content = "<link rel=\"preconnect\" href=\"" +
+                                   preconnect_url.spec() + "\" crossorigin>";
+  NavigateToDataURLWithContent(preconnect_content);
+  connection_listener_->WaitUntilFirstConnectionAccepted();
+  EXPECT_EQ(1u, connection_listener_->GetAcceptedSocketCount());
+  EXPECT_EQ(0u, connection_listener_->GetReadSocketCount());
+
+  // Second navigation to content with an img.
+  std::string font_content = "<script>var font = new FontFace('FontA', 'url(" +
+                             preconnect_url.spec() +
+                             "test.woff2)');font.load();</script>";

[mmenke, 2015/07/09 15:55:56]

Why would this test fail without the crossorigin on the preconnect?  Why does
the previous test not need it?

[Yoav Weiss, 2015/07/09 16:19:39]

Since this is a request for a font resource, which is fetch as an anonymous
resource, and therefore must be fetched using a connection from the pm pool.
For the previous test, it was for an image without a crossorigin attribute, so
it was a resource allowed to use credentials, and therefore must be fetched
using a connection from the non-pm pool.

[mmenke, 2015/07/09 16:31:30]

So how does one request make it to the PM pool, and the other make it to the
non-PM pool?  I guess there's magic in blink that give them different
first_party_for_cookies values?

If there's a crossorigin attribute for images, may make sense to just use an
image set with it here in this test, to make it more like the first
test...Though I suppose then we could silently regress if both cross origin
attributes broke at once.  Unclear how likely that is to happen and still have
font loading correct know its an anonymous request, though.

+  NavigateToDataURLWithContent(font_content);
+  connection_listener_->WaitUntilFirstConnectionRead();
+  EXPECT_EQ(1u, connection_listener_->GetAcceptedSocketCount());
+  EXPECT_EQ(1u, connection_listener_->GetReadSocketCount());
+}
 }  // namespace chrome_browser_net

[mmenke, 2015/07/09 15:55:56]

nit:  Blank line before end of namespace.

[Yoav Weiss, 2015/07/09 16:19:39]

added