Details
-
Type:
Sub-task
-
Status: Closed
-
Priority:
Major
-
Resolution: Fixed
-
Affects Version/s: None
-
Fix Version/s: 1.11.0
-
Component/s: None
-
Labels:None
Description
similar to current unionMergeRule.
Activity
- All
- Comments
- Work Log
- History
- Activity
- Transitions
Ah, I see, "intersect merge" is not syntax, it is a planner rule. Makes sense.
I'm working on variants of UnionMergeRule that work on Intersect and Minus.
Note that it is not valid to merge Minus right-deep: Minus(a, Minus(b, c)) is not equivalent to Minus(a, b, c).
Yes, in Hive, we only merge intersect. We have not gone that far to merge any type of minus yet.
Fixed in http://git-wip-us.apache.org/repos/asf/calcite/commit/9f44aea5.
New instances of UnionMergeRule flatten trees of Intersect and Minus operators into nodes that have more than 2 inputs.
New rules in PruneEmptyRules act on Intersect and Minus that have empty inputs, either converting them to the empty relational expression or removing empty inputs.
RelBuilder now allows Minus to have more than 2 inputs.
Hi Julian Hyde, thanks for your patch. The merge rule can be further improved for
topOp.all != bottomOp.all
for union and intersect cases. Right now, it is covering the cases
topOp.all, bottomOp.all
are
true,true
,
false,false
. Actually, it should also cover the case
false,true
. That is, if the top level is distinct, we can always merge.
Ah yes, missed that. Good one! Fixed in http://git-wip-us.apache.org/repos/asf/calcite/commit/ab25d36c.
Resolved in release 1.11.0 (2017-01-11).
I'm familiar with INTERSECT DISTINCT and INTERSECT ALL - but what is INTERSECT MERGE?