Move auth-error reporting code out of SyncGlobalError and into SigninGlobalError

chrome/browser/sync/profile_sync_service_unittest.cc

Index: chrome/browser/sync/profile_sync_service_unittest.cc
diff --git a/chrome/browser/sync/profile_sync_service_unittest.cc b/chrome/browser/sync/profile_sync_service_unittest.cc
index 7c420e7b1828596d20f618755bce8ab12ebcf7f1..4e81dfa7e7eb3e1ccdc86c5e642eefdcbdc5cce1 100644
--- a/chrome/browser/sync/profile_sync_service_unittest.cc
+++ b/chrome/browser/sync/profile_sync_service_unittest.cc
@@ -250,6 +250,7 @@ TEST_F(ProfileSyncServiceTest, AbortedByShutdown) {
                                      harness_.service.get()));
 
   harness_.service->Initialize();
+  harness_.service->Shutdown();

[tim (not reviewing), 2013/01/15 04:25:42]

... what was going on here before?  Isn't there a CHECK in ~PSS that Shutdown
was called?

[Andrew T Wilson (Slow), 2013/01/15 13:24:56]

Nope. There was an implicit expectation that if you call PSS::Initialize() that
you'd then call PSS::Shutdown() before the object was freed, but there's nothing
enforcing that expectation.

Until now, that is - since PSS::Shutdown() now calls
SyncGlobalError::RemoveProvider(), freeing the PSS without calling Shutdown()
first leads to a DCHECK.

I had two ways to fix this - I could either add some kind of cleanup code to the
PSS destructor, or I could fix the tests to properly call Shutdown() and I chose
to do the right thing and fix the tests.

   harness_.service.reset();
 }