Обсуждение: Move definition of standard collations from initdb to pg_collation.dat

While working on [0], I was wondering why the collations ucs_basic and 
unicode are not in pg_collation.dat.  I traced this back through 
history, and I think this was just lost in a game of telephone.

The initial commit for pg_collation.h (414c5a2ea6) has only the default 
collation in pg_collation.h (pre .dat), with initdb handling everything 
else.  Over time, additional collations "C" and "POSIX" were moved to 
pg_collation.h, and other logic was moved from initdb to 
pg_import_system_collations().  But ucs_basic was untouched.  Commit 
0b13b2a771 rearranged the relative order of operations in initdb and 
added the current comment "We don't want to pin these", but looking at 
the email[1], I think this was more a guess about the previous intent.

I suggest we fix this now; see attached patch.


[0]: 
https://www.postgresql.org/message-id/flat/1293e382-2093-a2bf-a397-c04e8f83d3c2%40enterprisedb.com

[1]: https://www.postgresql.org/message-id/28195.1498172402%40sss.pgh.pa.us

Peter Eisentraut <peter.eisentraut@enterprisedb.com> writes:
> While working on [0], I was wondering why the collations ucs_basic and 
> unicode are not in pg_collation.dat.  I traced this back through 
> history, and I think this was just lost in a game of telephone.
> The initial commit for pg_collation.h (414c5a2ea6) has only the default 
> collation in pg_collation.h (pre .dat), with initdb handling everything 
> else.  Over time, additional collations "C" and "POSIX" were moved to 
> pg_collation.h, and other logic was moved from initdb to 
> pg_import_system_collations().  But ucs_basic was untouched.  Commit 
> 0b13b2a771 rearranged the relative order of operations in initdb and 
> added the current comment "We don't want to pin these", but looking at 
> the email[1], I think this was more a guess about the previous intent.

Yeah, I was just loath to change the previous behavior in that
patch.  I can't see any strong reason not to pin these entries.

> I suggest we fix this now; see attached patch.

While we're here, do we want to adopt some other spelling of "the
root locale" than "und", in view of recent discoveries about the
instability of that on old ICU versions?

            regards, tom lane

On 28.03.23 13:33, Tom Lane wrote:
> While we're here, do we want to adopt some other spelling of "the
> root locale" than "und", in view of recent discoveries about the
> instability of that on old ICU versions?

That issue was fixed by 3b50275b12, so we can keep using the "und" spelling.