Re: POC, WIP: OR-clause support for indexes

Hi!
On 07.08.2024 04:11, Alexander Korotkov wrote:
> On Mon, Aug 5, 2024 at 11:24 PM Alena Rybakina
> <a(dot)rybakina(at)postgrespro(dot)ru> wrote:
>> Ok, thank you for your work)
>>
>> I think we can leave only the two added libraries in the first patch,
>> others are superfluous.
> Thank you.
> I also have fixed some grammar issues.

While reviewing the patch, I can't understand one part of the code where
we check the comparability of restrictinfos.

/* RestrictInfo parameters dmust match parent */
        if (subRinfo->is_pushed_down != rinfo->is_pushed_down ||
            subRinfo->is_clone != rinfo->is_clone ||
            subRinfo->security_level != rinfo->security_level ||
            !bms_equal(subRinfo->required_relids,
rinfo->required_relids) ||
            !bms_equal(subRinfo->incompatible_relids,
rinfo->incompatible_relids) ||
            !bms_equal(subRinfo->outer_relids, rinfo->outer_relids))
            return NULL;

I didn't find a place in the optimizer where required_relids,
incompatible_relids and outer_relids become different. Each
make_restrictinfo function takes arguments from
parent data.

I disabled this check and the regression tests passed. This code is
needed for security verification， may I clarify?

In the last patch I corrected the libraries - one of them was not in
alphabetical order.

--
Regards,
Alena Rybakina
Postgres Professional: http://www.postgrespro.com
The Russian Postgres Company