David E. Wheeler napsal(a):
> Replying to myself, but I've made some local changes (see other
> messages) and just wanted to follow up on some of my own comments.
>
> On Jul 2, 2008, at 21:38, David E. Wheeler wrote:
>
>>> 4) Operator = citext_eq is not correct. See comment
>>> http://doxygen.postgresql.org/varlena_8c.html#8621d064d14f259c594e4df3c1a64cac
>>>
>>
>> So should citextcmp() call strncmp() instead of varst_cmp()? The
>> latter is what I saw in varlena.c.
>
> I'm guessing that the answer is "no," since varstr_cmp() uses strncmp()
> internally, as appropriate to the locale. Correct?
You have to use varstr_cmp in citextcmp. Your code is correct, because for
< <= >= > operators you need collation sensible function.
You need to change only citext_cmp function to use strncmp() or call texteq
function.
>>> There must be difference between equality and collation for example
>>> in Czech language 'láska' and 'laská' are different word it means
>>> that 'láska' != 'laská'. But there is no difference in collation
>>> order. See Unicode Universal Collation Algorithm for detail.
>>
>> I'll leave the collation stuff to the functions I call (*far* from my
>> specialty), but I'll add a test for this and make sure it works as
>> expected. Um, although, with what collation should it be tested? The
>> tests I wrote assume en_US.UTF-8.
>
> I added this test and is passes:
>
> SELECT isnt( 'láska'::citext, 'laská'::citext, 'Diffrent accented
> characters should not be equivalent' );
I'm think that this test will work correctly for en_US.UTF-8 at any time. I
guess the test make sense only when Czech collation (cs_CZ.UTF-8) is selected,
but unfortunately, you cannot change collation during your test :(.
I think, Best solution for now is to keep the test and add comment about
recommended collation for this test.
Zdenek
