DomainKey-Signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma;
	h=message-id:date:from:to:subject:references:mime-version
	:content-type:content-disposition:in-reply-to:user-agent;
	b=dESpWkXr6nj9NpSy60t4kduIr9L+1jO8wQ7TyCYhf18swKNWT+G/khP8kfkShpEY5j
	tAva7u5WN1DK1ppTCncEeErhPRl7KujYvX/+sDzKaK27UNkGZdMXX0qFSsFLE1YGqIHg
	Wutz36eYAwRsoPlp1D04tfoTBcHhsCkXMjUTk=
Message-ID: <4c52f6b3.02b28f0a.02e2.ffff9a9e@mx.google.com>
Date: Fri, 30 Jul 2010 09:58:32 -0600
From: Joshua Tolley <eggyknap@gmail.com>
To: pgsql-sql@postgresql.org
Subject: Re: grouping subsets
References: <i28ubc$ie9$1@dough.gmane.org>
	<20100722100226.GE10348@a-kretschmer.de>
	<4C48230D.7030304@archonet.com>
	<m3mxtju510.fsf@passepartout.tim-landscheidt.de>
MIME-Version: 1.0
Content-Type: multipart/signed; micalg=pgp-sha1;
	protocol="application/pgp-signature"; boundary="5/uDoXvLw7AC5HRs"
Content-Disposition: inline
In-Reply-To: <m3mxtju510.fsf@passepartout.tim-landscheidt.de>
User-Agent: Mutt/1.5.18 (2008-05-17)


--5/uDoXvLw7AC5HRs
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Thu, Jul 22, 2010 at 11:31:23AM +0000, Tim Landscheidt wrote:
> Richard Huxton <dev@archonet.com> wrote:
>=20
> >>> What I want to get is the values grouped by "subset", where a subset =
is a set of rows with identical column until the colum changes.
> >>> Is there a way to get
>=20
> >>> | 2 | B |
> >>> | 4 | C |
> >>> | 4 | B |
> >>> | 3 | D |
>=20
> >>> by SQL only?
>=20
> >> I think, the problem is that there are 2 identical groups. I think, you
> >> can write a pl/pgsql-proc, selecting all ordered by the date-field and
> >> walking through the result to do the grouping, checking if the 2nd
> >> column is different from the previous.
>=20
> >> With plain SQL it's maybe possible too, but i don't know how ...
>=20
> > It should be do-able in 8.4 onwards, look into windowing
> > functions. In particular the lag() function:
>=20
> > SELECT
> >     mycode,
> >     mydate,
> >     lag(mycode) OVER (ORDER BY mydate) AS prev_code
> > FROM
> >     mytable
> > ORDER BY mydate;
>=20
> > It should be possible to use that as a subquery with an
> > outer query that compares mycode=3Dprev_code to get a run
> > length.
>=20
> Hmmm. Can the outer query be done without using "WITH
> RECURSIVE"?

How about this:

select
    a, b, c, d, sum
=66rom (
    select
        a, b, c, d, new_partition,
        sum(e) over (partition by partition_num)=20
    from (
        select
            a, b, c, d, e,
            case when
                lag(a, 1, null) over (order by d) is null or
                lag(a, 1, null) over (order by d) !=3D a or
                lag(b, 1, null) over (order by d) is null or
                lag(b, 1, null) over (order by d) !=3D b or
                lag(c, 1, null) over (order by d) is null or
                lag(c, 1, null) over (order by d) !=3D c
                    then nextval('a')
                else currval('a')
            end as partition_num,
            case when
                lag(a, 1, null) over (order by d) is null or
                lag(a, 1, null) over (order by d) !=3D a or
                lag(b, 1, null) over (order by d) is null or
                lag(b, 1, null) over (order by d) !=3D b or
                lag(c, 1, null) over (order by d) is null or
                lag(c, 1, null) over (order by d) !=3D c
                    then 'T'::boolean
                else 'f'::boolean
            end as new_partition
        from foo
    ) bar
) baz
where
    new_partition;

Here's my test table:
5432 josh@postgres# select * from foo;
 a | b | c |           d            | e=20
---+---+---+------------------------+---
 1 | 9 | 1 | 2007-01-01 00:00:05-07 | 8
 9 | 2 | 1 | 2007-01-01 00:00:10-07 | 4
 9 | 2 | 1 | 2007-01-01 00:00:15-07 | 6
 9 | 2 | 1 | 2007-01-01 00:00:20-07 | 2
 6 | 5 | 7 | 2007-01-01 00:00:25-07 | 3
 4 | 9 | 0 | 2007-01-01 00:00:30-07 | 0
 4 | 9 | 0 | 2007-01-01 00:00:35-07 | 7
 5 | 2 | 7 | 2007-01-01 00:01:25-07 | 7
 5 | 2 | 7 | 2007-01-01 00:01:30-07 | 7
 5 | 2 | 7 | 2007-01-01 00:01:35-07 | 9
 5 | 2 | 7 | 2007-01-01 00:01:40-07 | 2
 5 | 2 | 7 | 2007-01-01 00:01:45-07 | 5
 5 | 2 | 7 | 2007-01-01 00:01:50-07 | 8
 5 | 2 | 7 | 2007-01-01 00:01:55-07 | 5
 5 | 2 | 7 | 2007-01-01 00:02:00-07 | 9
 7 | 8 | 8 | 2007-01-01 00:02:05-07 | 7
 7 | 8 | 8 | 2007-01-01 00:02:10-07 | 8
 9 | 3 | 0 | 2007-01-01 00:02:15-07 | 0
 9 | 3 | 0 | 2007-01-01 00:02:20-07 | 8
 9 | 2 | 1 | 2007-01-01 00:02:25-07 | 3
 9 | 2 | 1 | 2007-01-01 00:02:30-07 | 3
(21 rows)

=2E..and these results...
 a | b | c |           d            | sum=20
---+---+---+------------------------+-----
 1 | 9 | 1 | 2007-01-01 00:00:05-07 |   8
 9 | 2 | 1 | 2007-01-01 00:00:10-07 |  12
 6 | 5 | 7 | 2007-01-01 00:00:25-07 |   3
 4 | 9 | 0 | 2007-01-01 00:00:30-07 |   7
 5 | 2 | 7 | 2007-01-01 00:01:25-07 |  52
 7 | 8 | 8 | 2007-01-01 00:02:05-07 |  15
 9 | 3 | 0 | 2007-01-01 00:02:15-07 |   8
 9 | 2 | 1 | 2007-01-01 00:02:25-07 |   6
(8 rows)

--
Joshua Tolley / eggyknap
End Point Corporation
http://www.endpoint.com

--5/uDoXvLw7AC5HRs
Content-Type: application/pgp-signature; name="signature.asc"
Content-Description: Digital signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)

iEYEARECAAYFAkxS9qgACgkQRiRfCGf1UMPVqwCgqIzV/T5sh5RTVWvbJbbclJ/a
848AnRiBWqga0ByBqYDxP5ovZ5JdLSRm
=CSMB
-----END PGP SIGNATURE-----

--5/uDoXvLw7AC5HRs--