Feedback-ID: ie0a040ee:Fastmail
Message-ID: <03f1450c-a730-40af-890c-c269db74a8ec@eisentraut.org>
Date: Wed, 21 Jan 2026 08:52:23 +0100
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: Collation with upper and numeric comparing in unexpected way
To: Matt Magoffin <postgresql.org@msqr.us>, pgsql-general@lists.postgresql.org
References: <9EBA273E-7F3C-40F4-8156-745A4BB5B090@msqr.us>
Content-Language: en-US
From: Peter Eisentraut <peter@eisentraut.org>
In-Reply-To: <9EBA273E-7F3C-40F4-8156-745A4BB5B090@msqr.us>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Archived-At: <https://www.postgresql.org/message-id/03f1450c-a730-40af-890c-c269db74a8ec%40eisentraut.org>
Precedence: bulk

On 20.01.26 19:36, Matt Magoffin wrote:
> I am using Postgres 17 and trying to configure a collation that sorts upper case before lower case and includes numeric sorting:
> 
> CREATE COLLATION testsort (provider = icu, locale = 'und-u-kf-upper-kn’);
> 
> These comparisons are working as I expected:
> 
> SELECT 'id-45' < 'id-123' COLLATE testsort; -- true (45 before 123)
> 
> SELECT 'id' < 'ID' COLLATE testsort; -- false (upper case before lower case)
> 
> However combining them resulted in an unexpected result:
> 
> SELECT 'id-45' < 'ID-123' COLLATE testsort; -- true
> 
> I thought that last one would be false because “ID” would come before “id”. Is there a way to configure the collation to achieve that? I’m trying to match the sorting behaviour in external application code.

I suspect that this is because the effect of the numeric sorting is a 
primary difference and the case difference is only a tertiary difference.

In other words, imagine the numeric sorting pass replacing all numbers 
by hypothetical letters corresponding to the numeric order, like

'id-45' -> 'id-X'
'id-123' -> 'id-Z'
'ID-123' -> 'ID-Z'

Then you would have

'id-45' < 'ID-123' =>
'id-X' < 'ID-Z'

which would be correct.

This is just my guess from the outside.  The numeric sorting is not a 
part of the Unicode Collation Algorithm standard, it is an extension by 
ICU, so one would have to dig into the code or documentation there, but 
I didn't find anything.

I don't know if there is a way to customize this further to get the 
effect you want.  Maybe you could reach out to an ICU support forum to 
get more expert insights there.