Track all extension frames in ProcessManager, inspect extensionoptions

chrome/browser/extensions/process_manager_browsertest.cc

Index: chrome/browser/extensions/process_manager_browsertest.cc
diff --git a/chrome/browser/extensions/process_manager_browsertest.cc b/chrome/browser/extensions/process_manager_browsertest.cc
index 980318c77b803ca623a644153d87d3799c12e521..d42af16dc59fb01318efcfcce28ba98fc5cf30e2 100644
--- a/chrome/browser/extensions/process_manager_browsertest.cc
+++ b/chrome/browser/extensions/process_manager_browsertest.cc
@@ -157,6 +157,37 @@ IN_PROC_BROWSER_TEST_F(ProcessManagerBrowserTest, HttpHostMatchingExtensionId) {
   EXPECT_TRUE(pm->GetBackgroundHostForExtension(extension->id()));
 }
 
+IN_PROC_BROWSER_TEST_F(ProcessManagerBrowserTest, NoBackgroundPage) {

[ncarter (slow), 2015/11/05 20:03:23]

I uploaded a test that works through some of the subframe cases here. It
definitely shows how some http subframes are being characterized as extension
subframes.

Please feel free to patch the test in, or graft those test cases into this test,
as you see fit. https://codereview.chromium.org/1413893004

+  ASSERT_TRUE(embedded_test_server()->InitializeAndWaitUntilReady());
+
+  ProcessManager* pm = ProcessManager::Get(profile());
+  const Extension* extension =
+      LoadExtension(test_data_dir_.AppendASCII("api_test")
+                    .AppendASCII("messaging")
+                    .AppendASCII("connect_nobackground"));
+  ASSERT_TRUE(extension);
+
+  // The extension has no background page.
+  EXPECT_EQ(0u, pm->GetRenderFrameHostsForExtension(extension->id()).size());
+
+  // Start in a non-extension process, then navigate to an extension process.
+  ui_test_utils::NavigateToURL(browser(),
+                               embedded_test_server()->GetURL("/empty.html"));
+  EXPECT_EQ(0u, pm->GetRenderFrameHostsForExtension(extension->id()).size());
+
+  const GURL extension_url = extension->url().Resolve("manifest.json");
+  ui_test_utils::NavigateToURL(browser(), extension_url);

[ncarter (slow), 2015/11/05 20:03:23]

Your current patch depends on RenderFrameHostChanged to catch this transition.
But, you ought to have been able to tell, at RenderFrameCreated time, that the
new frame will be in an extension process -- this information is in the
SiteInstance's SiteURL.

[robwu, 2015/11/12 00:33:58]

I have replaced RenderFrameHostChanged with DidNavigateAnyFrame (which also
works for non-OOPIF).

+  EXPECT_EQ(1u, pm->GetRenderFrameHostsForExtension(extension->id()).size());
+
+  ui_test_utils::NavigateToURL(browser(), GURL("about:blank"));
+  EXPECT_EQ(0u, pm->GetRenderFrameHostsForExtension(extension->id()).size());
+
+  ui_test_utils::NavigateToURLWithDisposition(
+      browser(), extension_url, NEW_FOREGROUND_TAB,
+      ui_test_utils::BROWSER_TEST_WAIT_FOR_NAVIGATION);
+  EXPECT_EQ(1u, pm->GetRenderFrameHostsForExtension(extension->id()).size());
+}
+
 // Verify correct keepalive count behavior on network request events.
 // Regression test for http://crbug.com/535716.
 IN_PROC_BROWSER_TEST_F(ProcessManagerBrowserTest, KeepaliveOnNetworkRequest) {