On Thu, Nov 16, 2023 at 7:26 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Laurenz Albe <laurenz.albe@cybertec.at> writes:
> > On Wed, 2023-11-15 at 09:04 +0000, PG Bug reporting form wrote:
> >> [ Microsoft decided to rename their Turkish locales ]
>
> > This seems to be a really nasty problem that Microsoft has created for us.
> > Even if the server started, you could not connect to the database if the
> > collation persisted in pg_database is no longer accepted by the operating
> > system.
>
> Yeah, that's a mess. The postgresql.conf entries could be fixed by hand,
> but if you've got "Turkish_Turkey.1254" in pg_database.datcollate or
> datctype then there's no easy way around that.
>
> I think really the right fix is a whole bunch of WTFs directed at
> Microsoft. I can understand wanting to support the Turkish_Türkiye
> spelling, but not having a backward-compatibility entry for the old
> spelling is a colossal blunder -- especially in a minor OS release.
> Aren't these the same people who prize backward compatibility at
> any cost?
For the record, the Windows manual tells us very, very clearly not to
store those "display" style names anywhere, and to use BCP 47 locale
names instead, which are based on ISO standards. I had patches to
teach initdb to use BCP 47, and I remain convinced that we must do
that, but I have failed to attract the attention of enough people who
use and understand Windows to be able to commit anything. In
particular, Juan José pointed out some problems that I just couldn't
understand.
https://www.postgresql.org/message-id/flat/CA%2BhUKGJ%3DXThErgAQRoqfCy1bKPxXVuF0%3D2zDbB%2BSxDs59pv7Fw%40mail.gmail.com