MIME-Version: 1.0
References: <19099-e05dcfa022fe553d@postgresql.org> <CACJufxF9FcuYe8XOuWLgWK77HCUHpOc6+7+NkktFFNmzw15jKg@mail.gmail.com>
 <CAHewXN=vF5d9O4R3+iUwLqEaP7pb8iYAN_e3vEE_p5sJHofn7w@mail.gmail.com> <2960545.1761800903@sss.pgh.pa.us>
In-Reply-To: <2960545.1761800903@sss.pgh.pa.us>
From: Kirill Reshke <reshkekirill@gmail.com>
Date: Thu, 30 Oct 2025 10:31:38 +0500
Message-ID: <CALdSSPi7udsgQg3PUG=Z4+-9pRg8wT3HkDvTgYvtg30xNWQ9OA@mail.gmail.com>
Subject: Re: BUG #19099: Conditional DELETE from partitioned table with
 non-updatable partition raises internal error
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Tender Wang <tndrwang@gmail.com>, jian he <jian.universality@gmail.com>, exclusion@gmail.com, 
	pgsql-bugs@lists.postgresql.org, Amit Langote <amitlangote09@gmail.com>
Content-Type: text/plain; charset="UTF-8"
Archived-At: <https://www.postgresql.org/message-id/CALdSSPi7udsgQg3PUG%3DZ4%2B-9pRg8wT3HkDvTgYvtg30xNWQ9OA%40mail.gmail.com>
Precedence: bulk

On Thu, 30 Oct 2025 at 10:08, Tom Lane <tgl@sss.pgh.pa.us> wrote:

>
> It's surely pretty accidental (and arguably not desirable)
> if "DELETE FROM pt WHERE false" doesn't fail the same way.
>
> Now, I agree that it's not great if you instead get an
> internal error like "could not find junk ctid column".
> But that smells to me like error checks being applied in
> the wrong order rather than something fundamentally wrong.
>
> I didn't look at the proposed patch yet.
>
>                         regards, tom lane

I wrote:

> But I also wonder if Jian's fix fixed the right thing. Should we
> instead fail in the planning phase with a more user-friendly error
> message? This will be a regression though, because 'DELETE FROM
> file_fdw_table WHERE false' will no longer work...

On the second thought, I doubt anyone will get unhappy with 'DELETE
FROM file_fdw_table WHERE false' stop working

Alexander wrote:

> On 86dc9005~1 or with enable_partition_pruning = 'on', EXPLAIN outputs the
> query plan and "DELETE FROM pt WHERE false;" completes with no error.

So, behaviour was wrong both before and after  86dc9005, just in different ways.

On head, we get an error about junk columns, because without partition
pruning, we derive the result relation as 'pt', not its partition
'p1', which is correct I believe.  But with 'p1' as result relation,
we (correctly) error out in ExecInitModifyTable while with 'pt' we
don't.

So, error checks are applied, order is not wrong, but rather checks
are not full enough?  I mean, we I believe we need to execute
CheckValidResultRel against all partitions in ExecInitModifyTable, at
least when no partition pruning has been performed

-- 
Best regards,
Kirill Reshke