[ https://issues.apache.org/jira/browse/ZOOKEEPER-666?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12832405#action_12832405 ]
Martin Traverso commented on ZOOKEEPER-666: ------------------------------------------- Indeed, breaking backward-compatibility would be a big issue. One drawback I see with 2b is that the API would seem to allow calling those methods after connect() has been called. What would the semantics be in that scenario? How about a builder-based approach instead? E.g, {noformat} ZooKeeper client = ZooKeeper.builder() .timeout(...) .watcher(...) .connect(connectString); {noformat} This would make it possible to add future parameters without breaking the public API, while preserving immutability semantics for initialization parameters. > Unsafe publication in client API > -------------------------------- > > Key: ZOOKEEPER-666 > URL: https://issues.apache.org/jira/browse/ZOOKEEPER-666 > Project: Zookeeper > Issue Type: Bug > Components: java client > Affects Versions: 3.2.2 > Reporter: Martin Traverso > Fix For: 3.3.0 > > > The following code may result in a data race due to unsafe publication of a > reference to "this". The call to cnxn.start() spawns threads that have access > to the partially-constructed reference to the ZooKeeper object. > See http://www.ibm.com/developerworks/java/library/j-jtp0618.html for some > background info. > {noformat} > public ZooKeeper(String connectString, int sessionTimeout, Watcher watcher) > throws IOException > { > ..... > cnxn = new ClientCnxn(connectString, sessionTimeout, this, > watchManager); > cnxn.start(); > } > {noformat} > The obvious fix is to move the call to cnxn.start() into a separate start() > method. -- This message is automatically generated by JIRA. - You can reply to this email to add a comment to the issue online.