Re: Support for RANGE ... PRECEDING windows in OVER

Craig Ringer

Thursday, June 20, 2013 7:24 PM

Hi all

Since 8.4, PostgreSQL has had extremely useful window function support -
but support for "RANGE PRECEDING / FOLLOWING" windows was dropped late
in 8.4's development in order to get the rest of the feature in, per
http://archives.postgresql.org/pgsql-hackers/2010-02/msg00540.php.

It looks like there was discussion of requiring a new opclass to be
declared for types or otherwise extending opclasses to provide the
information required for RANGE ... PRECEDING / FOLLOWING (
http://www.postgresql.org/message-id/20100211201444.GA28270@svana.org )
. I can't find any sign that it went anywhere beyond some broad
discussion:
http://www.postgresql.org/message-id/13993.1265920013@sss.pgh.pa.us at
the time.

I've missed this feature more than once, and am curious about whether
any more recent changes may have made it cleaner to tackle this, or
whether consensus can be formed on adding the new entries to btree's
opclass to avoid the undesirable explicit lookups of the '+' and '-'
oprators.

Some question seems to remain open about how ranges over
timestamps/intervals should work, but this wasn't elaborated on.

There's been interest in this, eg:

http://pgsql.hackers.free-usenet.eu/[HACKERS]-range-intervals-in-window-function-frames_T66085695_S1

http://grokbase.com/t/postgresql/pgsql-general/105a89gm2n/postgresql-9-0-support-for-range-value-preceding-window-functions