MIME-Version: 1.0
References: 
 <CAA5RZ0sbWmqdUBFo8JXMJe72pnwjxVY58htJ6pKbwnyQuRctQw@mail.gmail.com>
 <CAP53PkxqGbPw5VzpacyJb2wTofYJadCoUmxV8s2o5tHzKznwbg@mail.gmail.com>
 <CAA5RZ0t3-srmtaVNfzv0HEzXCa78eQDKYVak+3zgEccnKPZHHA@mail.gmail.com>
 <CAA5RZ0t_CJs9ye1PPBy-e0ajZ-6FgKvNEoacW9keCtF_7Y2ycQ@mail.gmail.com>
 <CAA5RZ0vEnNvp7_Ni8bWwh4GE53rg_YwNjqynQG=JpNu3YzzAWA@mail.gmail.com>
 <CAP53PkzCzeDpg6dCzVrU17LCiCmRSWfEt-5dr4y+VHnTxKr_9w@mail.gmail.com>
 <CAA5RZ0tzLhGxR3cCQtPs1=HeGWh0WDDBC1KDTgOh9x9u2gvy1Q@mail.gmail.com>
 <CAP53PkyY7AhQ8jZDwtHokyEfpuHLxxDvtY7EmcasR_n80yLiRQ@mail.gmail.com>
 <CAA5RZ0unJX_EwepME9NAuB=u8LjtRCU67FW6yMP07tFkJE5yRA@mail.gmail.com>
 <CAP53PkycUYE6d_Pyt2BcxmTaq7v8qDt6LtBObVfsUt+WNWf0ug@mail.gmail.com>
 <acYTBcq5-tck1zjh@paquier.xyz>
In-Reply-To: <acYTBcq5-tck1zjh@paquier.xyz>
From: Lukas Fittl <lukas@fittl.com>
Date: Fri, 27 Mar 2026 00:19:01 -0700
Message-ID: 
 <CAP53Pkz0XqVoHv-pfq+i6Ngs5HX1TC=G6spHWGMpusARBZ0YbQ@mail.gmail.com>
Subject: Re: Refactor query normalization into core query jumbling
To: Michael Paquier <michael@paquier.xyz>
Cc: Sami Imseih <samimseih@gmail.com>, zengman <zengman@halodbtech.com>,
	pgsql-hackers <pgsql-hackers@postgresql.org>,
 Julien Rouhaud <rjuju123@gmail.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Archived-At: 
 <https://www.postgresql.org/message-id/CAP53Pkz0XqVoHv-pfq%2Bi6Ngs5HX1TC%3DG6spHWGMpusARBZ0YbQ%40mail.gmail.com>
Precedence: bulk

On Thu, Mar 26, 2026 at 10:18=E2=80=AFPM Michael Paquier <michael@paquier.x=
yz> wrote:
> This line of arguments is stronger for the normalization of the query
> string.  Why should the core code decide what a normalized string
> should look like when it comes to the detection of the constants, if
> any?  Instead of a dollar-quoted number, we could enforce a bunch of
> things, like a '?' or a '$woozah$' at these locations.

Fair enough, though I haven't seen any extensions that do that in
practice - its reasonable to have normalization result in a query
string that's parsable again and can be passed to EXPLAIN
(GENERIC_PLAN).

> Saying that, the key point of the patch is to take a copy of the
> JumbleState locations, and recompute it for the needs of PGSS for the
> sake of the normalized query representation.  Hence, why don't we just
> do that at the end?  That should be enough to enforce the const marker
> for the JumbleState across all the loaded extensions that want to look
> at it.  This leads me to the simpler patch attached.
>
> Comments or tomatoes?  That's simpler, and I'd be OK with just that
> for v19.  That would be much better than the current state of affairs,
> where PGSS decides to enforce its own ideas in the JumbleState, ideas
> fed to anything looping into the post-parse-analyze hook after PGSS.

I'm not convinced that making the const change alone is a good idea,
without also providing some helpers to reduce the repeated code in
extensions.

What if we only put the ComputeConstantLengths (as Sami had it in v7)
in core, together with making JumbleState const?

Thanks,
Lukas

--=20
Lukas Fittl