> For speed no unique\foreign constraints must be checked during bulk
> insert. All this constraints will be checked at the end of bulk insertion
> during index merge\rebuild phase. Another (not perfect but possible) way
> is to disallow bulk insert into table which have such constraints.

Is that your personal opinion or result of the discussion with people
that asked this feature? What about other engines like Oracle or MS SQL?

Do we have information about how leading ETL tools handle de-duplication
during ETL process? I can imagine them relying on database handling such
tasks, since one of the main methods, at least for Java-based ETL tools,
is PreparedStatement.executeBatch(), which has to return an array of
flags for each record in batch which tells whether it was
added/updated/deleted or not...

Roman