Refactored the CloudPolicyProvider so that it becomes initialized once and stays initialized.

chrome/browser/chromeos/login/login_browsertest.cc

Index: chrome/browser/chromeos/login/login_browsertest.cc
diff --git a/chrome/browser/chromeos/login/login_browsertest.cc b/chrome/browser/chromeos/login/login_browsertest.cc
index c4ee33d5eef533807282a0ffc44b516027344d51..f27c2fa8b5269b5df9aed876851283a6b0796649 100644
--- a/chrome/browser/chromeos/login/login_browsertest.cc
+++ b/chrome/browser/chromeos/login/login_browsertest.cc
@@ -34,6 +34,8 @@ class LoginTestBase : public CrosInProcessBrowserTest {
     mock_cryptohome_library_ = cros_mock_->mock_cryptohome_library();
     EXPECT_CALL(*mock_cryptohome_library_, IsMounted())
         .WillRepeatedly(Return(true));
+    EXPECT_CALL(*mock_cryptohome_library_, InstallAttributesIsReady())
+        .WillRepeatedly(Return(false));

[Mattias Nissler (ping if slow), 2012/03/30 08:48:46]

and this? It seems like the more reasonable choice would be to return true.

[Joao da Silva, 2012/03/30 10:09:23]

Same reason. Returning false is more straightforward, since returning true will
require a couple more expected calls; but since device policy doesn't matter for
these tests, returning false suffices.

[Mattias Nissler (ping if slow), 2012/03/30 11:46:35]

Hm. In that case, why do we enable the flags for all tests? Is there away to
just enable them for the tests that actually need them?

   }
 
   MockCryptohomeLibrary* mock_cryptohome_library_;