postgres-xc-developers Mailing List for Postgres-XC

Following two join.sql issues are fixed by attached patch:

1. ERROR:  PlaceHolderVar found where not expected
In create_remotejoin_plan(), for queries such as :
select count(t2.*)
from int8_tbl t1 left join (select * from int8_tbl) t2 on (t1.q2 = t2.q1);
It needs all the target list vars to evaluate if it is possible to reduce
the join. And for that even if the parent targetlist has placeholder vars,
we need to recurse through it and not just abort. So the fix has used
PVC_RECURSE_PLACEHOLDERS to do this.

1. For queries such as this:
select count(t2.*)
from int8_tbl t1 left join (select * from int8_tbl) t2 on (t1.q2 = t2.q1);
After we fix issue #1 above, we get error:
ERROR: input of anonymous composite types is not implemented

The issue is: The target list in RemoteQuery node has RowExpr expressions,
not plain Var nodes. And so we deparse this to ROW(...). And when we fetch
the result, this cannot be converted to input type because it is an
anonymous record type.

The fix is to add a Result plan node over RemoteQuery node in cases where
the target list is not a plain var tlist. Result node handles the RowExpr
evaluation. The target list passed to RemoteQuery will be a derived plain
var tlist out of the passed-in tlist.  We have to do this because
RemoteQuery plan does not support target list projections.

So after the fix here's how the plan for the above query will look:

                                     QUERY
PLAN
------------------------------------------------------------------------------------
 Aggregate  (cost=0.02..0.03 rows=1 width=32)
   Output: count((ROW(int8_tbl.q1, int8_tbl.q2, int8_tbl.id3)))
   ->  Nested Loop Left Join  (cost=0.00..0.01 rows=1 width=32)
         Output: (ROW(int8_tbl.q1, int8_tbl.q2, int8_tbl.id3))
         Join Filter: (t1.q2 = int8_tbl.q1)
         ->  Data Node Scan on t1  (cost=0.00..0.00 rows=1000 width=8)
               Output: t1.q1, t1.q2, t1.id3
               Node/s: data_node_1, data_node_2
               Remote query: SELECT q1, q2, id3 FROM ONLY int8_tbl t1 WHERE
true
         ->  *Result*  (cost=0.00..0.00 rows=0 width=0)
               Output: int8_tbl.q1, ROW(int8_tbl.q1, int8_tbl.q2,
int8_tbl.id3)
               ->  Data Node Scan on int8_tbl  (cost=0.00..0.00 rows=1000
width=40)
                     Output: int8_tbl.q1, int8_tbl.q2, int8_tbl.id3
                     Node/s: data_node_1, data_node_2
                     Remote query: SELECT q1, q2, id3 FROM ONLY int8_tbl
WHERE true


Regression run is ok.