Enable cloud policy by default.

chrome/browser/policy/browser_policy_connector.cc

Index: chrome/browser/policy/browser_policy_connector.cc
diff --git a/chrome/browser/policy/browser_policy_connector.cc b/chrome/browser/policy/browser_policy_connector.cc
index 5132383c0b51b850e18d45d0f18a3e3cd3ba109c..e81e9d932b899a75b36b165b4721f8043eeef965 100644
--- a/chrome/browser/policy/browser_policy_connector.cc
+++ b/chrome/browser/policy/browser_policy_connector.cc
@@ -221,7 +221,10 @@ DeviceMode BrowserPolicyConnector::GetDeviceMode() {
 
 void BrowserPolicyConnector::ScheduleServiceInitialization(
     int64 delay_milliseconds) {
-  device_management_service_->ScheduleInitialization(delay_milliseconds);
+  // Skip device initialization if the BrowserPolicyConnector was never
+  // initialized (unit tests).
+  if (device_management_service_.get())

[Joao da Silva, 2013/01/29 17:02:28]

Why not change the broken tests to call SetDeviceManagementServiceForTesting
with a MockDeviceManagementService?

[Andrew T Wilson (Slow), 2013/01/29 17:26:47]

Yeah, there are too many tests that break. I think the real fix is to pull the
call to ScheduleServiceInitialization out of UserPolicySigninService, so I'll
add a TODO here.

+    device_management_service_->ScheduleInitialization(delay_milliseconds);
 }
 
 #if defined(OS_CHROMEOS)
@@ -369,9 +372,11 @@ bool MatchDomain(const string16& domain, const string16& pattern) {
 
 // static
 bool BrowserPolicyConnector::IsNonEnterpriseUser(const std::string& username) {
-  if (username.empty()) {
-    // This means incognito user in case of ChromiumOS and
-    // no logged-in user in case of Chromium (SigninService).
+  if (username.empty() || username.find('@') == std::string::npos) {
+    // An empty username means incognito user in case of ChromiumOS and
+    // no logged-in user in case of Chromium (SigninService). Many tests use
+    // nonsense email addresses (e.g. 'test') so treat those as non-enterprise
+    // users.

[Joao da Silva, 2013/01/29 17:02:28]

Why not change those tests to use real email addresses?

[Andrew T Wilson (Slow), 2013/01/29 17:26:47]

One could. Mainly I don't want to touch a boatload of tests, and I've
rationalized that I'm being true to the spirit of the routine because "test" is
not an enterprise user :-P

     return true;
   }