On 18-01-2017 12:43, Tom Lane wrote:
> I wrote:
>> running bootstrap script ... ok
>> performing post-bootstrap initialization ... 2017-01-18 09:49:45.019 EST [25919] FATAL: collation "aa_ER@saaho" for
encoding"UTF8" already exists
>> 2017-01-18 09:49:45.019 EST [25919] STATEMENT: SELECT pg_import_system_collations(if_not_exists => false, schema =>
'pg_catalog');
>
> As a stopgap so I could get some work done, I did
>
> - PG_CMD_PUTS("SELECT pg_import_system_collations(if_not_exists => false, schema => 'pg_catalog');\n\n");
> + PG_CMD_PUTS("SELECT pg_import_system_collations(if_not_exists => true, schema => 'pg_catalog');\n\n");
>
> and what I now see in pg_collation is
>
> regression=# select * from pg_collation where collname like 'aa_ER%';
> collname | collnamespace | collowner | collencoding | collcollate | collctype
> ------------------+---------------+-----------+--------------+------------------+------------------
> aa_ER | 11 | 10 | 6 | aa_ER | aa_ER
> aa_ER.utf8 | 11 | 10 | 6 | aa_ER.utf8 | aa_ER.utf8
> aa_ER.utf8@saaho | 11 | 10 | 6 | aa_ER.utf8@saaho | aa_ER.utf8@saaho
> aa_ER@saaho | 11 | 10 | 6 | aa_ER.utf8@saaho | aa_ER.utf8@saaho
> (4 rows)
>
> Maybe an appropriate fix would be to ignore collations whose names aren't
> equal to what we get for collcollate/collctype. Presumably the latter
> are getting canonicalized somehow.
>
collname 'en_US' seems to be more popular than 'en_US.utf8' (it is
shorter). We can't ignore locales without .utf8 or @something because it
would break queries with COLLATE clause.
Ignore collations at initdb time seems to fix the error. However, do
collations remain the same as 9.6?
--
Euler Taveira Timbira - http://www.timbira.com.br/
PostgreSQL: Consultoria, Desenvolvimento, Suporte 24x7 e Treinamento
