On Jul 5, 2008, at 02:58, Gregory Stark wrote:
>> txt = cilower( PG_GETARG_TEXT_PP(0) );
>> str = VARDATA_ANY(txt);
>>
>> result = hash_any((unsigned char *) str, VARSIZE_ANY_EXHDR(txt));
>
> I thought your data type implemented a locale dependent collation,
> not just
> a case insensitive collation. That is, does this hash agree with your
> citext_eq on strings like "foo bar" <=> "foobar" and "fooß" <=>
> "fooss" ?
CITEXT is basically intended to replace all those queries that do
`WHERE LOWER(col) = LOWER(?)` by doing it internally. That's it. It's
locale-aware to the same extent that `LOWER()` is (and that citext 1.0
is not, since it only compares ASCII characters case-insensitively).
And I expect that it does, in fact, agree with your examples, in that
all the current tests for = and <> pass:
try=# select 'foo bar' = 'foobar'; ?column?
---------- f
try=# SELECT 'fooß' = 'fooss'; ?column?
---------- f
> You may have to use strxfrm
In the patch against CVS HEAD, it uses str_tolower() in formatting.c.
Best,
David
