2008/11/26 Heikki Linnakangas <heikki.linnakangas@enterprisedb.com>:
> Hitoshi Harada wrote:
>>
>> 2008/11/26 David Rowley <dgrowley@gmail.com>:
>>>
>>> I'm at a bit of a loss to what to do now. Should I wait for your work
>>> Heikki? Or continue validating this patch?
>>>
>>> The best thing I can think to do right now is continue and any problems I
>>> find you can add regression tests for, then if we keep your regression
>>> tests
>>> for Heikki's changes then we can validate those changes more quickly.
>>>
>>> Any thoughts? Better ideas?
>>
>> Thanks to your great tests, we now know much more about specification
>> and where to fail easily, so continuing makes sense but it may be good
>> time to take a rest and wait for Heikki's patch completing.
>
> Here's another updated patch, including all your bug fixes.
>
> There's two known issues:
> - ranking functions still don't treat peer rows correctly.
>
> - I commented out the "this function requires ORDER BY clause in the window"
> test in rank_up, because a window function shouldn't be poking into the
> WindowState struct like that. I wonder if it's really needed? In section
> 7.11, the SQL2008 spec says "if WD has no window ordering clause, then the
> window ordering is implementation-dependent, and *all rows are peers*". The
> regression test now fails because of this, but the current behavior actually
> seems correct to me.
>
Yes, I was wrong. The reason I put the error in rank() without ORDER
BY is nothing but I didn't find it. It is actually a reasonable
specification, isn't it.
This is tiny thing, but "negative transition function" can be called
"inverse transition function"? I feel the latter is more readable.
Regards,
--
Hitoshi Harada