[fix](planner) Fix wrong plan of SQL with cross join and aggreagate

[Kikyou1997]

Proposed changes

Issue Number: noissue

Problem summary

SELECT col1
FROM 
    (SELECT col1 FROM t1) a
INNER JOIN 
    (SELECT sum(b3) AS 'b3u'
    FROM 
        (SELECT `c`.`col2` AS `c2`,
         sum(b.col3) AS `b3`
        FROM `t2` b
        INNER JOIN `t3` c
            ON ((`b`.`col1` = `c`.`col1`))
        GROUP BY  `c`.`col2`) dd ) b;

For such SQL type, the materialized attribute of SlotDescriptor will be updated during the optmization for countStar and cross join(see this method materializeInlineViewResultExprForCrossJoinOrCountStar), though the single node plan has been generated, and the generation of plan is dependent on the materialized attribute, so apparently there is a parodox here.

Fix:

1. Materialize all slot in the chain of sourceExprs.
2. Materialize agg info again after the creation of table ref with join operator
3. Add all the SlotDescriptor to HashJoinNode's output from child output no matter is materilalized or not
4. Once the Slot is required keep it in the output tuple instead of pruning it, no matter the slot is materialized or not

Checklist(Required)

1. Does it affect the original behavior:
   • Yes
   • No
   • I don't know
2. Has unit tests been added:
   • Yes
   • No
   • No Need
3. Has document been added or modified:
   • Yes
   • No
   • No Need
4. Does it need to update dependencies:
   • Yes
   • No
5. Are there any changes that cannot be rolled back:
   • Yes (If Yes, please explain WHY)
   • No

Further comments

If this is a relatively large or complex change, kick off the discussion at dev@doris.apache.org by explaining why you chose the solution you did and what alternatives you considered, etc...

[Kikyou1997]

@morrySnow @morrySnow

[morrySnow]

add comments

[morrySnow]

add comments