Date: Fri, 10 May 2019 09:59:04 -0400
From: Alvaro Herrera <alvherre@2ndquadrant.com>
To: Amit Langote <Langote_Amit_f8@lab.ntt.co.jp>
Cc: David Rowley <david.rowley@2ndquadrant.com>,
	Bruce Momjian <bruce@momjian.us>,
	Justin Pryzby <pryzby@telsasoft.com>,
	PostgreSQL-development <pgsql-hackers@postgresql.org>
Subject: Re: pg12 release notes
Message-ID: <20190510135904.GA32055@alvherre.pgsql>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <17a07ed6-4d17-838e-2a8f-fd125c22e03e@lab.ntt.co.jp>
User-Agent: Mutt/1.9.4 (2018-02-28)
Precedence: bulk

On 2019-May-10, Amit Langote wrote:

> On 2019/05/10 12:18, David Rowley wrote:
> > On Fri, 10 May 2019 at 12:08, Bruce Momjian <bruce@momjian.us> wrote:
> >>> 17f206f Set pg_class.relhassubclass for partitioned indexes
> >>
> >> I need help with this one.  I know the system column existed in previous
> >> releases, so how is it different now?  Do we document system table
> >> changes that are implementation-behavior in the release notes?  Usually
> >> we don't.

I very much doubt that the change is relevant to the release notes.

> > This appears to be fixing something that likely should have been done
> > in PG11, where partitioned indexes were added.
> 
> That's true.  We (Michael and I) felt the need to do this change, because
> it allowed the pg_partition_tree() code (which is also new in v12) to use
> the same infrastructure for both partitioned tables and indexes; checking
> the relhassubclass flag allows to short-circuit scanning pg_inherits to
> find out that there are no children.

I'm of two minds about backpatching that fix.  In pg12 it makes sense to
have the fix, to allow the SRF functions to work correctly.  Since we
don't use that flag for partitioned indexes anywhere in the backend in
pg11, it seems pretty useless to have it there.  On the other hand, if
any user tool is inspecting catalogs, it might fail to point out
descendants for partitioned indexes, if the user asked for a report...
but frankly I doubt there are any tools that do that, or users that care
for such a report, or even that that report exist ... and even if that
report existed, I doubt that it would optimize out the read of
pg_inherits by checking relhassubclass beforehand.  

Still, it's an inconsistency in pg11.  I vote -0 to getting it
backpatched, mostly because it seems like more work than is warranted.
(I think the work consists not only of testing that the backpatched
commit works correctly, but also documenting for 11.4 release notes how
to fix existing catalogs after upgrading.)

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services