[ http://opensource.atlassian.com/projects/hibernate/browse/HHH-1283?page=c... ] Strong Liu edited comment on HHH-1283 at 3/30/11 5:18 AM: ---------------------------------------------------------- Unfortunately, this patch will not work reliably unless the data is ordered by parent. For the HQL in the test case, "select p from Parent p join fetch p.children c", Hibernate generates: select parent0_.name as name4_0_, children1_.name as name5_1_, children1_.parent_id as parent2_4_0__, children1_.name as name0__ from Parents parent0_ inner join chlidren children1_ on parent0_.name=children1_.parent_id Notice that there is no ORDER BY clause. Unless the DB returns sorted data (e.g., due to an index or primary key), the resultset probably won't have rows for a each parent in a contiguous block. The scrolling logic reads sequentially; when it reaches a new parent, it assumes it is finished with all rows for the current parent. If there are actually more rows for the current parent later in the resultset then that parent will be returned with fewer elements than it should contain. Later, when more rows for that parent are found, Hibernate will (again) return that same parent with the same incomplete set. The testScroll() test case passes using the patch because the DB happens to return it in the right order. It fails as described above using H2, if the data generation method is changed to: {code} private void insertTestData() { Session s = openSession(); Transaction t = s.beginTransaction(); Child child_1_1 = new Child( "achild1-1"); Child child_1_2 = new Child( "ychild1-2"); Child child_1_3 = new Child( "dchild1-3"); Child child_2_1 = new Child( "bchild2-1"); Child child_2_2 = new Child( "cchild2-2"); Child child_2_3 = new Child( "zchild2-3"); s.save( child_1_1 ); s.save( child_2_1 ); s.save( child_1_2 ); s.save( child_2_2 ); s.save( child_1_3 ); s.save( child_2_3 ); s.flush(); Parent p1 = new Parent( "parent1" ); p1.addChild( child_1_1 ); p1.addChild( child_1_2 ); p1.addChild( child_1_3 ); s.save( p1 ); Parent p2 = new Parent( "parent2" ); p2.addChild( child_2_1 ); p2.addChild( child_2_2 ); p2.addChild( child_2_3 ); s.save( p2 ); t.commit(); s.close(); } {code} was (Author: stliu): Unfortunately, this patch will not work reliably unless the data is ordered by parent. For the HQL in the test case, "select p from Parent p join fetch p.children c", Hibernate generates: select parent0_.name as name4_0_, children1_.name as name5_1_, children1_.parent_id as parent2_4_0__, children1_.name as name0__ from Parents parent0_ inner join chlidren children1_ on parent0_.name=children1_.parent_id Notice that there is no ORDER BY clause. Unless the DB returns sorted data (e.g., due to an index or primary key), the resultset probably won't have rows for a each parent in a contiguous block. The scrolling logic reads sequentially; when it reaches a new parent, it assumes it is finished with all rows for the current parent. If there are actually more rows for the current parent later in the resultset then that parent will be returned with fewer elements than it should contain. Later, when more rows for that parent are found, Hibernate will (again) return that same parent with the same incomplete set. The testScroll() test case passes using the patch because the DB happens to return it in the right order. It fails as described above using H2, if the data generation method is changed to: private void insertTestData() { Session s = openSession(); Transaction t = s.beginTransaction(); Child child_1_1 = new Child( "achild1-1"); Child child_1_2 = new Child( "ychild1-2"); Child child_1_3 = new Child( "dchild1-3"); Child child_2_1 = new Child( "bchild2-1"); Child child_2_2 = new Child( "cchild2-2"); Child child_2_3 = new Child( "zchild2-3"); s.save( child_1_1 ); s.save( child_2_1 ); s.save( child_1_2 ); s.save( child_2_2 ); s.save( child_1_3 ); s.save( child_2_3 ); s.flush(); Parent p1 = new Parent( "parent1" ); p1.addChild( child_1_1 ); p1.addChild( child_1_2 ); p1.addChild( child_1_3 ); s.save( p1 ); Parent p2 = new Parent( "parent2" ); p2.addChild( child_2_1 ); p2.addChild( child_2_2 ); p2.addChild( child_2_3 ); s.save( p2 ); t.commit(); s.close(); } -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: http://opensource.atlassian.com/projects/hibernate/secure/Administrators.... - For more information on JIRA, see: http://www.atlassian.com/software/jira