When cross-site navigations are cancelled, delete the request being transferred

chrome/browser/prerender/prerender_browsertest.cc

Index: chrome/browser/prerender/prerender_browsertest.cc
===================================================================
--- chrome/browser/prerender/prerender_browsertest.cc	(revision 250813)
+++ chrome/browser/prerender/prerender_browsertest.cc	(working copy)
@@ -3621,6 +3621,27 @@
   content::SetBrowserClientForTesting(original_browser_client);
 }
 
+// Checks that URLRequests for prerenders being aborted on cross-process
+// navigation from a server redirect are cleaned up, so they don't keep cache
+// entries locked.
+// See http://crbug.com/341134
+IN_PROC_BROWSER_TEST_F(PrerenderBrowserTest,
+                       PrerenderCrossProcessServerRedirectNoHang) {
+  const char kDestPath[] = "files/prerender/prerender_page.html";
+  // Force everything to be a process swap.
+  SwapProcessesContentBrowserClient test_browser_client;
+  content::ContentBrowserClient* original_browser_client =
+      content::SetBrowserClientForTesting(&test_browser_client);
+
+  PrerenderTestURL(CreateServerRedirect(kDestPath), FINAL_STATUS_OPEN_URL, 0);
+
+  content::SetBrowserClientForTesting(original_browser_client);

[davidben, 2014/02/13 21:08:47]

Is there any particular reason to revert to the original ContentBrowserClient
before navigating? I assume it doesn't matter much either way, but it seems a
little strange.

[mmenke, 2014/02/13 21:13:56]

I did that because I wanted to test a "normal" navigation, and a navigation with
a weird browser client isn't very normal.  The content browser test I added
*also* use it to do a weird navigation, of course.

I did go back and forth on this, though, and I agree it's also weird.  I've gone
ahead and switched.

+
+  ui_test_utils::NavigateToURL(
+      browser(),
+      test_server()->GetURL(kDestPath));
+}
+
 // Checks that prerenders are aborted on cross-process navigation from
 // a client redirect.
 IN_PROC_BROWSER_TEST_F(PrerenderBrowserTest,