Details
-
Type:
Bug
-
Status: Closed
-
Priority:
Major
-
Resolution: Fixed
-
Affects Version/s: None
-
Fix Version/s: 1.4.0-incubating
-
Component/s: None
-
Labels:None
Description
A query such as :
select sum(t1.deptno) over(partition by t1.deptno)
from emp t1, emp t2
would raise exception:
org.apache.calcite.sql.validate.SqlValidatorException: Column 'EMPNO' is ambiguous.
Calcite cannot point the column in OVER-CLAUSE to the JOINed table.
Activity
- All
- Comments
- Work Log
- History
- Activity
- Transitions
I've assigned it to you. Change the status to "in progress" if you're working on it.
It struck me as strange that you needed to qualify a column name when a namespace was also passed along. Then I noticed this comment in SqlWindow.isTableSorted:
// REVIEW: jhyde, 2007/11/7: This is the only use of
// findAllColumnNames. Find a better way to detect monotonicity, then
// remove that method.
Basically, we shouldn't be using SqlMoniker. Monotonicity detection should be using column name and namespace, like the rest of the validator code. A namespace represents a table, and it may map to single or multiple strings: EMP, SCHEMA.EMP, DB.SCHEMA.EMP, T1 are all examples.
That comment basically predicted this problem, so now it's time to fix it.
I am not sure if we can remove findAllColumnNames() entirely?
Before moving into Monotonicity detection, we still need to get all the column names right? It is where findAllColumnNames() is necessary?
I put in another fix which doesn't use findAllColumnNames. See https://github.com/julianhyde/incubator-calcite/tree/754-resolve.
Fixed in http://git-wip-us.apache.org/repos/asf/incubator-calcite/commit/87acda31.
Sean Hsuan-Yi Chu, thanks for the patch! In the end, I used your test case (see http://git-wip-us.apache.org/repos/asf/incubator-calcite/commit/ebce955d) but developed a better fix.
The solution looks more concise. Thanks!
Resolved in release 1.4.0-incubating (2015-08-23)
I would like to work on this bug