Обсуждение: \dO versus collations for other encodings

I've noticed that psql's \dO command for showing collations is a bit
schizophrenic about whether it shows entries for collations that are
irrelevant in the current database (because they use a different
encoding).  For example:

regression=# \dOS aa*                        List of collations  Schema   |       Name       |     Collate      |
Ctype      
 
------------+------------------+------------------+------------------pg_catalog | aa_DJ            | aa_DJ.utf8       |
aa_DJ.utf8pg_catalog| aa_DJ.utf8       | aa_DJ.utf8       | aa_DJ.utf8pg_catalog | aa_ER            | aa_ER
|aa_ERpg_catalog | aa_ER.utf8       | aa_ER.utf8       | aa_ER.utf8pg_catalog | aa_ER.utf8@saaho | aa_ER.utf8@saaho |
aa_ER.utf8@saahopg_catalog| aa_ER@saaho      | aa_ER@saaho      | aa_ER@saahopg_catalog | aa_ET            | aa_ET
     | aa_ETpg_catalog | aa_ET.utf8       | aa_ET.utf8       | aa_ET.utf8
 
(8 rows)

regression=# \dOS pg_catalog.aa*                        List of collations  Schema   |       Name       |     Collate
  |      Ctype       
 
------------+------------------+------------------+------------------pg_catalog | aa_DJ            | aa_DJ.utf8       |
aa_DJ.utf8pg_catalog| aa_DJ            | aa_DJ            | aa_DJpg_catalog | aa_DJ.iso88591   | aa_DJ.iso88591   |
aa_DJ.iso88591pg_catalog| aa_DJ.utf8       | aa_DJ.utf8       | aa_DJ.utf8pg_catalog | aa_ER            | aa_ER
  | aa_ERpg_catalog | aa_ER.utf8       | aa_ER.utf8       | aa_ER.utf8pg_catalog | aa_ER.utf8@saaho | aa_ER.utf8@saaho
|aa_ER.utf8@saahopg_catalog | aa_ER@saaho      | aa_ER@saaho      | aa_ER@saahopg_catalog | aa_ET            | aa_ET
       | aa_ETpg_catalog | aa_ET.utf8       | aa_ET.utf8       | aa_ET.utf8
 
(10 rows)

The second display is including collations that are not actually
available for use in this database.  The reason for the weird
discrepancy is that processSQLNamePattern is making use of
pg_collation_is_visible() in the first case but not the second,
and pg_collation_is_visible() rejects incompatible collations
(cf CollationGetCollid).

Given that this display doesn't include any encoding column, I'm
thinking that the intention was to show only relevant collation entries.
Which we could do by adding a WHERE clause about the encoding.
If the intention was to not restrict that way, don't we need an encoding
column?  (But I'm not actually sure how we could make that work
unsurprisingly without changes in CollationGetCollid, which would likely
break other things, so I don't really want to hear suggestions that we
should do it the other way ...)
        regards, tom lane

On fre, 2011-04-08 at 20:14 -0400, Tom Lane wrote:
> Given that this display doesn't include any encoding column, I'm
> thinking that the intention was to show only relevant collation
> entries.
> Which we could do by adding a WHERE clause about the encoding.
> If the intention was to not restrict that way, don't we need an
> encoding
> column?  (But I'm not actually sure how we could make that work
> unsurprisingly without changes in CollationGetCollid, which would
> likely
> break other things, so I don't really want to hear suggestions that we
> should do it the other way ...)

The fix you pushed looks OK.

The whole way of dealing with the wrong encodings has been schizophrenic
throughout the development of this feature.  One idea was that at some
point we could add support for creating collations in the wrong encoding
in template databases, if that turns out to be something that is
requested.  You can, of course, do that manually already.