> On Nov 16, 2015, at 6:12 PM, Jinfeng Ni <[email protected]> wrote: > > The only concern I have is about JDBC data source that Drill is > pushing to. If we get the sort order fully specified in planning > time, then the same query might see different results when it goes > through the native engine (which uses a different sort order), or it > goes through Drill.
Agreed. Some work on the JDBC adapter will be needed. If the JDBC source supports the desired null direction (e.g. has NULLS FIRST/LAST syntax) we should generate it. If the column is NOT NULL, it’s not a problem. With Mondrian we ended up generating "ORDER BY ISNULL(x), x” on MySQL[1]. Julian [1] http://jira.pentaho.com/browse/MONDRIAN-226
