No longer keep a lingering reference to TypedUrlDataTypeController after PSS exits

chrome/browser/sync/profile_sync_service.cc

Index: chrome/browser/sync/profile_sync_service.cc
diff --git a/chrome/browser/sync/profile_sync_service.cc b/chrome/browser/sync/profile_sync_service.cc
index 3e605140a87bf8b45e55ca13652eeb48fcf87be0..6cd304bcf4a839429c19ee126dda7928d9f9953b 100644
--- a/chrome/browser/sync/profile_sync_service.cc
+++ b/chrome/browser/sync/profile_sync_service.cc
@@ -156,6 +156,16 @@ ProfileSyncService::ProfileSyncService(ProfileSyncComponentsFactory* factory,
 ProfileSyncService::~ProfileSyncService() {
   sync_prefs_.RemoveSyncPrefObserver(this);
   Shutdown();
+#if defined(NDEBUG)
+  // Make sure there are no dangling references to the DataTypeControllers.
+  for (browser_sync::DataTypeController::TypeMap::const_iterator iter =
+           data_type_controllers_.begin(); iter != data_type_controllers_.end();
+       ++iter) {
+    DCHECK(iter->second->HasOneRef())

[Nicolas Zea, 2012/05/16 22:40:58]

This does not hold true for non-UI thread datatypes implementing the new API,
where not blocking shutdown was one of our goals. In those cases, we don't
bother to wait for them to finish stopping, since we have an disconnect-able
interface that they must go through to interact with sync.

In other words, once DTC::Stop() is called, the datatype won't be able to
interact with the PSS or anything within sync, letting us continue without
blocking shutdown. The DTC ma have running/pending tasks on the datatype's
thread that don't execute until after Stop(), but we ensure correctness.

What do you think about, at least for new DTC's, just making it clear that no
member of the DTC should rely on the profile being around after Stop is called
(either by clearing these members during Stop() or by ensuring they are POD's)?

+        << "DataTypeController will outlive ProfileSyncService: "
+        << iter->second->name();
+  }
+#endif
 }
 
 bool ProfileSyncService::AreCredentialsAvailable() {