org.hibernate.Session.clear() considered Harmful?

Here's another reason that I just ran into: caching previous results when calling a stored procedure multiple times within same transaction. Simplified code as follows.

//Begin transaction
SessionFactory sf = HibernateSessionFactory.getFactory();
Session dbSession = sf.getCurrentSession();
dbSession.beginTransaction();

//First call to stored procedure
Query query = dbSession.getNamedQuery("RR_CUST_OPP_DATA");
query.setString("custName", "A");
List<ShipSummaryRow> shipSummaryRows = query.list();

//Second call to stored procedure
Query query = dbSession.getNamedQuery("RR_CUST_OPP_DATA");
query.setString("custName", "B");
List<ShipSummaryRow> shipSummaryRows = query.list();

//Commit both    
dbSession.getTransaction().commit();

Without a clear() after the first call, the resultset rows of the first call are replicated into the resultset of the second call. I'm using Oracle 11gR2.

Key to replicating this bug is to make both calls within the same transaction. Since I'm using open session in view pattern, both calls automatically happen within the same transaction (as the original code calls the proc within a loop storing the results of each). Hence I call it a bug; else could be considered a feature but even then clear() is not called out in code samples stating it should be called. session.flush() did nothing. Mapping file as below. As a result I've added clear() to the end of all my procedure calls. Have not yet tested with my custom SQL calls. This is trivial stuff; surprised the bug exists.

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE hibernate-mapping PUBLIC "-//Hibernate/Hibernate Mapping DTD 3.0//EN" "http://hibernate.sourceforge.net/hibernate-mapping-3.0.dtd">
<hibernate-mapping>
    <class name="com.jfx.rr.model.ShipSummaryRow">
        <id name="id" type="integer"/>
        <property name="shipQtrString" not-null="true" type="string"/>
        <property name="shipAmount" not-null="true" type="double"/>
    </class>
    <sql-query callable="true" name="RR_CUST_OPP_DATA">
        <return class="com.jfx.rr.model.ShipSummaryRow">
            <return-property column="SHIPPED_ID" name="id"/>
            <return-property column="SHIP_QTR" name="shipQtrString"/>
            <return-property column="SHIPPED_AMOUNT" name="shipAmount"/>
        </return>
        { call RR_DASHBOARD_REPORTS_PKG.RR_CUST_OPP_DATA(?, :custName) }
    </sql-query>
</hibernate-mapping>

Ad. A: Looks like you are aware what clear() does. The reason to call it explicitly is to remove all managed entities from L1 cache, so that it does not grow infinitely when processing large data sets in one transaction.

It discards all the changes that were made to managed entites not explicitly persisted. This means that you can safely modify an entity, update it explicitly and clear the session. This is the right design. Obviously if no changes are made (long, but read only session), clear() is always safe.

You can also use stateless sessions.

Ad. B: No, it exists for the reasons above: to make sure L1 (session cache) does not grow too much. Of course maintaining it manually is a poor idea and an indication that another tool should be used for large data sets, but sometimes it is a must.

Note that in JPA specification there is also clear() and flush() method. In this case you should always call flush() first to push changes into the database (explicit update) prior to calling clear().

Ad. C: It's actually a good idea to warn the user (maybe by issuing warning message rather than throwing an exception) when he/she clears the session with dirty changes. Also I don't think a framework code should call clear() unconditionally, unless it is sure that the user code it runs flushes or does not make any changes.