> we exhibit an example contradicting these assumptions: it is possible for an SI [snapshot isolation] history to be non-serializable while the sub-history cont

> we exhibit an example contradicting these assumptions: it is possible for an SI [snapshot isolation] history to be non-serializable while the sub-history containing all update transactions is serializable. Do you know what this case is? It could be a truth that won't matter to many/most of us--perhaps good to know, but no reason to stop everything. I don't particularly care as DBs I work with have been Repeatable Read or Read Committed. It was demonstrated on an Oracle database and it's unclear if it would apply to others.