From: Tom Lane <tgl@sss.pgh.pa.us>
To: Tender Wang <tndrwang@gmail.com>
cc: Richard Guo <guofenglinux@gmail.com>,
        Peter Eisentraut <peter@eisentraut.org>, exclusion@gmail.com,
        pgsql-bugs@lists.postgresql.org
Subject: Re: BUG #19000: gist index returns inconsistent result with
 gist_inet_ops
In-reply-to: 
 <CAHewXN=qY4XqMnqbHt0r-pTvS8mZ51mabExF30XQx8C9Kg0n4Q@mail.gmail.com>
References: <19000-2525470d200672ab@postgresql.org>
 <CAMbWs49vGc45oQ98bhx4dFpXOsimUX9TTnXOXG=i6+X6sAqBRw@mail.gmail.com>
 <d570c46d-1e20-4766-93de-c6fcbc18f282@eisentraut.org>
 <2107725.1753985855@sss.pgh.pa.us>
 <CAMbWs48sCRUbz0CmkPSFEH8G1b9y-erq6rqYpA0sPhb_ZmwOpQ@mail.gmail.com>
 <2189027.1754019208@sss.pgh.pa.us>
 <CAHewXN=qY4XqMnqbHt0r-pTvS8mZ51mabExF30XQx8C9Kg0n4Q@mail.gmail.com>
Comments: In-reply-to Tender Wang <tndrwang@gmail.com>
	message dated "Fri, 01 Aug 2025 12:38:00 +0800"
MIME-Version: 1.0
Content-Type: text/plain; charset="UTF-8"
Content-ID: <2197758.1754024003.1@sss.pgh.pa.us>
Content-Transfer-Encoding: 8bit
Date: Fri, 01 Aug 2025 00:53:23 -0400
Message-ID: <2197759.1754024003@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/2197759.1754024003%40sss.pgh.pa.us>
Precedence: bulk

Tender Wang <tndrwang@gmail.com> writes:
> Tom Lane <tgl@sss.pgh.pa.us> 于2025年8月1日周五 11:33写道：
>> Richard Guo <guofenglinux@gmail.com> writes:
>>> On Fri, Aug 1, 2025 at 3:17 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
>>>> We've known about this for, um, decades: btree_gist's support for
>>>> inet/cidr is fundamentally broken [1][2].

> What I debugged yesterday was, in  gbt_inet_consistent(),
> 	query = convert_network_to_scalar(dquery, INETOID, &failure);

Right --- as I mentioned in one of the linked threads, the thing that
is fundamentally wrong here is relying on convert_network_to_scalar
in the first place.  That has no charter to deliver exact results,
and it doesn't.  In the present example I think the issue is that
it doesn't consider the netmask at all; but there are many other
cases where it's not an exact representation of inet's sort order.
That's not a big problem for its intended purpose for statistical
estimates, but it's completely fatal for index infrastructure.

			regards, tom lane