AW: PostgreSQL 10 initdb with locale de-x-icu fails on Windows Server 2012

Hello Thomas,

> Windows 10 built in ICU support
I thought all ICU stuff was built in PostgreSQL binaries and DLLs, but your explanation seems to be the cause of this
problem.

> ancient Windows
This sounds like using default-collation "de" for initdb and createdb has the same effect as "de-x-icu" with PostgreSQL
10up to 14?
 
Should we then use default-collation "de" instead of "de-x-icu" or "German_Germany.1252" or could this cause problems
laterwhen upgrading to PostgreSQL 15 or 16 and using "de-x-icu" as default-collation?
 

Regards, Stefan.

-----Ursprüngliche Nachricht-----
Von: Thomas Munro <thomas.munro@gmail.com> 
Gesendet: Dienstag, 5. Juli 2022 10:19
An: ROSESOFT S. Hoffmann <sh@rosesoft.de>
Cc: Kyotaro Horiguchi <horikyota.ntt@gmail.com>; pgsql-bugs@lists.postgresql.org
Betreff: Re: PostgreSQL 10 initdb with locale de-x-icu fails on Windows Server 2012

On Tue, Jul 5, 2022 at 7:22 PM Thomas Munro <thomas.munro@gmail.com> wrote:
> On Mon, Jul 4, 2022 at 11:53 PM ROSESOFT S. Hoffmann <sh@rosesoft.de> wrote:
> > On Windows Server 2016/2019/2022 and Windows 10/11 the ICU-Options work as expected!
> > Why does the problem only occurs on Windows Server 2012?
>
> Windows 10 and later contain a copy of ICU inside them.  That's why 
> strings like "de", "de-CH" etc work there: PostgreSQL is using the 
> normal Windows collation API (it's a bit weird that we call it "libc"
> mode, when it's not exactly libc, but what we mean by that is the OS's 
> native collations...).  So I think it is probably Windows' internal 
> copy of ICU that receives that name "de-..." and ignores the trailing 
> part that it doesn't understand.

Oh, it might also be that ancient Windows can also understand "de-DE"
or "de" too, but not "de-x-icu"?  That "de-x-icu" name is an internal PostgreSQL convention for collation identifiers
andshouldn't be passed to initdb.  If that's the case it might just be that the ICU-based versions of Windows are more
tolerant,and interpret "de-x-icu" as "de", ignoring the rest, but the ancient versions reject it?
 

(I have no access to old Windows, I just read that even Vista could understand de-DE type names, formally known as
BCP47,in some contexts... it just wasn't using ICU yet).