sync: Retry soon when server returns conflict

sync/engine/backoff_delay_provider.cc

 // Copyright (c) 2012 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "sync/engine/backoff_delay_provider.h"
 
 #include "base/rand_util.h"
 #include "sync/internal_api/public/engine/polling_constants.h"
 #include "sync/internal_api/public/sessions/model_neutral_state.h"
 #include "sync/internal_api/public/util/syncer_error.h"
@@ skipping 79 matching lines @@
   // Note: If we received a MIGRATION_DONE on download updates, then commit
   // should not have taken place.  Moreover, if we receive a MIGRATION_DONE
   // on commit, it means that download updates succeeded.  Therefore, we only
   // need to check if either code is equal to SERVER_RETURN_MIGRATION_DONE,
   // and not if there were any more serious errors requiring the long retry.
   if (state.last_download_updates_result == SERVER_RETURN_MIGRATION_DONE ||
       state.commit_result == SERVER_RETURN_MIGRATION_DONE) {
     return short_initial_backoff_;
   }
 
+  // When the server tells us we have a conflict, then we should download the
+  // latest updates so we can see the conflict ourselves, resolve it locally,
+  // then try again to commit.  Running another sync cycle will do all these
+  // things.  There's no need to back off, we can do this immediately.

[tim (not reviewing), 2012/11/21 21:37:18]

I agree this solution is safe and easy and we should land it.  Can you also add
a TODO though, and perhaps reference a bug, to say that ideally we would have
dealt with protocol errors like this before dispatching to BackoffDelayProvider,
since they don't actually require backoff.

+  if (state.commit_result == SERVER_RETURN_CONFLICT) {
+    return short_initial_backoff_;
+  }
+
   return default_initial_backoff_;
 }
 
 }  // namespace syncer