Re: [HACKERS] path toward faster partition pruning

On 2018/02/22 17:41, David Rowley wrote:
> On 21 February 2018 at 23:44, Amit Langote
> <Langote_Amit_f8(at)lab(dot)ntt(dot)co(dot)jp> wrote:
>> Please find attached updated patches.
>
> Thanks for updating the code.
>
> The question I have now is around NULL handling in
> partkey_datum_from_expr(). I've not managed to find a way to get a
> NULL Const in there as it seems all the clauses I try get removed
> somewhere earlier in planning. Do you know for a fact that a NULL
> Const is impossible to get there?

We only ever call partkey_datum_from_expr() for an OpExpr's arg and if you
have a NULL Const in there, eval_const_expressions() would've folded the
OpExpr's and subsequently any AND'd OpExpr's into a constant-false qual.

create table p (a int) partition by list (a);
create table p1 partition of p for values in (1);
create table p2 partition of p for values in (2);

explain select * from p where a = null and a = 1;
QUERY PLAN
-------------------------------------------
Result (cost=0.00..0.00 rows=0 width=40)
One-Time Filter: false

explain select * from p where (a = null and a = 1) or a = 2;
QUERY PLAN
----------------------------------------------------------
Append (cost=0.00..41.94 rows=13 width=4)
-> Seq Scan on p2 (cost=0.00..41.88 rows=13 width=4)
Filter: (a = 2)
(3 rows)

> I'm having to add some NULL handling there for the run-time pruning
> patch but wondered if it was also required for your patch.

Hmm, not sure why. Can you explain a bit more?

Thanks,
Amit