…r to be more linear
…ened) Under heavy load, it was possible to spin a thread forever if the accum was changed during the calculation of the time expiry for the bucket. I think this new impl has no chance of failure or loss.
…rameter behavior. It was agreed that it's a bug to just dump the built store into an existing directory. Now you either must pass a non-existent directory or use the '-force-overwrite' flag to make sure it's wiped before running job.
… in the wrong order. Also change its own parameters to match the order of StoreDefinition's constructor. No other code needs to be changed as all callers pass the same value for preferred and required. It's probably why the bug has not been noticed until now.
Allow the NoopEngine to return data if so configured - useful for testing as you can hand it a value, which will be returned for all subsequent get() (and as part of delete()) I'm not utterly convinced this is the right way to go, but it's useful and switched off by default
…e right protocol. Ensure syntactic reconciliation is done server-side.
…, serialization, or version reconciliation yet.
Before this change there was no benefit to assigning the result to a long.