Re: BUG #6669: unique index w/ multiple columns and NULLs

On Mon, Jun 04, 2012 at 10:29:22AM -0500, Kevin Grittner wrote:
> Bruce Momjian <bruce@momjian.us> wrote:
>
> > COUNT(*) can't skip nulls because there is no specified column,
> > but why does COUNT(col) skip nulls --- again, inconsistent.
>
> I disagree -- one is counting rows, the other is counting rows with
> a value in that column.  I guess one could criticize the syntax for
> specifying that as non-obvious, but it seems pretty reasonable to
> me.

I get your point about COUNT(*) really counting rows, not values, but
why doesn't GROUP BY then skip nulls?

    WITH null_test (col1, col2) AS
    (
       SELECT 1, null
       UNION ALL
       SELECT null, null
    )
    SELECT COUNT(*), col2 FROM null_test group by col2
    UNION ALL
    SELECT COUNT(col1), col2 FROM null_test group by col2;

     count | col2
    -------+------
         2 |
         1 |
    (2 rows)

Since col2 is null in both places, why it is processed?  Looks like
GROUP BY is selecting the NULL rows, then COUNT is processing them based
on its rules.

I think the original complaint is that NULL != NULL in a WHERE clause,
but GROUP BY is able to group them together just fine.

Anyway, just thoughts on the topic.

--
  Bruce Momjian  <bruce@momjian.us>        http://momjian.us
  EnterpriseDB                             http://enterprisedb.com

  + It's impossible for everything to be true. +