Re: Make transformAExprIn() return a flattened bool expression directly - Mailing list pgsql-hackers
| From | Chao Li |
|---|---|
| Subject | Re: Make transformAExprIn() return a flattened bool expression directly |
| Date | |
| Msg-id | 9BEDFEA1-0F39-4D56-9A10-C419DA1B7A18@gmail.com Whole thread |
| In response to | Make transformAExprIn() return a flattened bool expression directly ("cca5507" <cca5507@qq.com>) |
| Responses |
Re: Make transformAExprIn() return a flattened bool expression directly
|
| List | pgsql-hackers |
> On Apr 24, 2026, at 14:06, cca5507 <cca5507@qq.com> wrote:
>
> Hi,
>
> Now transformAExprIn() returns a bool expression tree. Attach a patch
> to make it return a flattened bool expression directly without extra cost.
> This can reduce some work of the planner.
>
> --
> Regards,
> ChangAo Chen
> <v1-0001-Make-transformAExprIn-return-a-flattened-bool-exp.patch>
Hi ChangAo,
Thanks for the patch. In the first impression, this patch helps. It changes a deep-tree OR, like (Part1 OR (Part2 OR
Part3)), to a flat OR list, like OR [Part1, Part2, Part3].
I tried a SQL: select t1.* from t1, t2, t3 where t1.id in (1, 2, t2.id, t3.id, 3);
From the dumped query tree, I can see the difference clearly:
Before patch:
```
:quals
{BOOLEXPR
:boolop or
:args (
{BOOLEXPR
:boolop or
:args (
{SCALARARRAYOPEXPR
:opno 96
:opfuncid 65
:useOr true
:args (
{VAR
:varno 1
}
{ARRAYEXPR
:array_typeid 1007
:array_collid 0
:element_typeid 23
:elements (
{CONST
:consttype 23
:constvalue 4 [ 1 0 0 0 0 0 0 0 ]
}
{CONST
:consttype 23
:constvalue 4 [ 2 0 0 0 0 0 0 0 ]
}
{CONST
:consttype 23
:constvalue 4 [ 3 0 0 0 0 0 0 0 ]
}
)
}
)
:location 40
}
{OPEXPR
:opno 96
:opfuncid 65
:opresulttype 16
:args (
{VAR
:varno 1
}
{VAR
:varno 2
:varattno 1
}
)
:location 40
}
)
:location 40
}
{OPEXPR
:opno 96
:opfuncid 65
:opresulttype 16
:args (
{VAR
:varno 1
}
{VAR
:varno 3
}
)
:location 40
}
)
:location 40
}
}
```
After the patch:
```
:quals
{BOOLEXPR
:boolop or
:args (
{SCALARARRAYOPEXPR
:opno 96
:opfuncid 65
:args (
{VAR
:varno 1
}
{ARRAYEXPR
:array_typeid 1007
:array_collid 0
:element_typeid 23
:elements (
{CONST
:consttype 23
:constvalue 4 [ 1 0 0 0 0 0 0 0 ]
}
{CONST
:consttype 23
:constvalue 4 [ 2 0 0 0 0 0 0 0 ]
}
{CONST
:consttype 23
:constvalue 4 [ 3 0 0 0 0 0 0 0 ]
}
)
}
)
:location 40
}
{OPEXPR
:opno 96
:opfuncid 65
:args (
{VAR
:varno 1
}
{VAR
:varno 2
}
)
:location 40
}
{OPEXPR
:opno 96
:opfuncid 65
:args (
{VAR
:varno 1
}
{VAR
:varno 3
}
)
:location 40
}
)
:location 40
}
}
```
After the patch, the qual tree is flat and shorter.
However, the final execution plan is the same before and after patching, which shows the planner has been smart enough:
```
evantest=# explain select t1.* from t1, t2, t3 where t1.id in (1, 2, t2.id, t3.id, 3);
QUERY PLAN
---------------------------------------------------------------------------------------------
Nested Loop (cost=0.00..352435619.25 rows=411638852 width=4)
Join Filter: ((t1.id = ANY ('{1,2,3}'::integer[])) OR (t1.id = t2.id) OR (t1.id = t3.id))
-> Nested Loop (cost=0.00..81358.62 rows=6502500 width=8)
-> Seq Scan on t1 (cost=0.00..35.50 rows=2550 width=4)
-> Materialize (cost=0.00..48.25 rows=2550 width=4)
-> Seq Scan on t2 (cost=0.00..35.50 rows=2550 width=4)
-> Materialize (cost=0.00..48.25 rows=2550 width=4)
-> Seq Scan on t3 (cost=0.00..35.50 rows=2550 width=4)
(8 rows)
```
I am not an expert on planner, but I guess, in general, processing a flat OR list is cheaper than dealing with a tree.
So,I still agree this is an improvement for v20 unless I miss some regression case. You may add this patch to the CF
fortracking.
Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/
pgsql-hackers by date: