Horton

Re: Collations and Replication; Next Steps

Peter Geoghegan <pg@heroku.com>

View thread View on postgr.es
From: Peter Geoghegan <pg@heroku.com>
To: Heikki Linnakangas <hlinnakangas@vmware.com>
Cc: Oleg Bartunov <obartunov@gmail.com>, Martijn van Oosterhout <kleptog@svana.org>, Tatsuo Ishii <ishii@postgresql.org>, Peter Eisentraut <peter_e@gmx.net>, Matt Kelly <mkelly@tripadvisor.com>, Pgsql Hackers <pgsql-hackers@postgresql.org>, Matt Spilich <mspilich@tripadvisor.com>
Date: 2014-09-18T18:58:59Z
Lists: pgsql-hackers 
On Thu, Sep 18, 2014 at 6:51 AM, Heikki Linnakangas
<hlinnakangas@vmware.com> wrote:
> The same it works with libxml, openssl, libreadline and all the other
> libraries you can build with.

I like the comparison with libxml. If we were to adopt ICU, it would
be as a core component that makes collation versioning work, that in
practice all packages use. It wouldn't actually be mandatory, but
almost universally available in practice.

I really think that long term, relying on the OS collations is not a
good plan. It's a big contributing factor to our reticence on the
question of when two different systems should be considered compatible
for the purposes of physical replication. Not that I'm volunteering to
work on it!
-- 
Peter Geoghegan