Re: Crash report for some ICU-52 (debian8) COLLATE and work_mem values

On Thu, Aug 17, 2017 at 6:13 PM, Peter Eisentraut
<peter.eisentraut@2ndquadrant.com> wrote:
> On 8/14/17 23:55, Peter Geoghegan wrote:
>> pg_import_system_collations() should simply use uloc_countAvailable()
>> + uloc_getAvailable, rather than ucol_countAvailable() +
>> ucol_getAvailable().
>
> It's not clear to me that this is better.  Why do we need to use a
> function that is clearly not the preferred API for this ("col" vs "loc")
> just to get more entries?

My argument for doing this is very simple: ICU/CLDR/BCP 47 provides
stability guarantees for locales, not collations [1]. For example, as
we discussed, de_BE didn't actually go away -- it just stopped being a
distinct collation within ICU, for reasons that are implementation
defined.

I admit that there are arguments against this, but by far the most
important consideration should be the stability of the names used for
pg_collation entries created during initdb.

> If we go down this route, then we'll be on
> the hook forever to keep adding more and more predefined entries by
> whatever means necessary.

Why? Locales have a mapping to various ISO codes (Country, language,
etc). It's not like new ones come along very often.

[1] https://tools.ietf.org/html/bcp47#section-3.4
-- 
Peter Geoghegan