Re: Improve the performance of Unicode Normalization Forms.

Hi Michael,

25.03.2026 09:12, Michael Paquier wrote:
> On Wed, Feb 25, 2026 at 11:21:43AM +0300, Alexander Borisov wrote:
>> Gentle ping — did you have a chance to look at this patch series?
>> If anything needs to be changed/added/removed, I’m happy to update it.
>
> I find this patch series pretty cool. I cannot take it for this
> release, unfortunately, but I'd be happy to study and potentially do
> something about this patch set when v20 opens for business.

Thank you for your time! Let's aim for v20.
I would be very glad if we can genuinely improve Unicode in PostgreSQL.

> Rather than share the files you have used for your benchmarks,
> couldn't you share a script that can generate them (dirty is fine)? I
> am guessing that this could just be a perl script that goes through
> the unicode and normalization data and builds the testing patterns you
> are looking for to prove your point. This would be much better than
> uploading things somewhere: if we don't have a reproducing set of data
> on pgsql-hackers, then we'd lose a part of the test history. That
> would not be cool. Any committer who could look at your patch will
> need these files to double-check your claims, and anything that
> reduces the review burden can speed up the evaluation process.
>
> Jeff has posted some scenarios upthread, but I am also wondering what
> you have exactly done in terms of benchmark, and reviewing benchmarks
> is part of the patch review process.

Attached rebased patches and Perl script (generate_NF_C_D_KC_KD_sql.pl)
for generating test files. At the beginning of the script, there is
a description of how to use it and what it generates.

--
Alexander Borisov