[ https://issues.jboss.org/browse/TEIID-2555?page=com.atlassian.jira.plugin... ] Steven Hawkins commented on TEIID-2555: --------------------------------------- The initial commit is in for 8.5 alpha1. I have initially gone with the common table with a tuplebuffer set as a list abstraction for pushdown. This is different than the pushdown of just the keys, which uses parameters and references tuplebuffers on the command. These mechanisms should probably be consolidated somewhat - however the current key pushdown logic is intended to be a semi-join as duplicate rows have not been removed. This begins with using the makedep join option. For consistency, the planning decision to push the entire dependent join happens in RuleChooseDependent. For example: select pm1.g1.e1, pm1.g1.e2, pm2.g1.e2 FROM pm1.g1, pm2.g1 where (pm1.g1.e1 = pm2.g1.e1) option makedep pm1.g1(join) This will result in two pushdown queries, the first to build the independent side: SELECT g_0.e1, g_0.e2 FROM g1 AS g_0 The other is the full pushdown: WITH X__1 (e1, e2) AS (?) SELECT g_0.e1, g_0.e2, g_1.e2 FROM g1 AS g_0, X__1 AS g_1 WHERE g_0.e1 = g_1.e1 The source for full dependent join pushdown will be responsible for removing the with item defined by a tuplebuffer and turning it into a temporary table. The reference to that table must then be placed in the query as a substitute for the reference to the common table. More changes will probably be desirable to convey the original metadata ids on the with columns as otherwise all we have is the type information - however in cross source situations having native information from the other side may not be that useful. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira