From: Tom Lane <tgl@sss.pgh.pa.us>
To: Peter Geoghegan <pg@bowt.ie>
cc: Alvaro Herrera <alvherre@2ndquadrant.com>,
        PostgreSQL Hackers <pgsql-hackers@postgresql.org>
Subject: Re: Fixing findDependentObjects()'s dependency on scan order
 (regressions in DROP diagnostic messages)
In-reply-to: 
 <CAH2-WzmRtQLJdC7KSdtwxQ65Yk+_Vi89eVqKJvPOx=Xc7WLLKw@mail.gmail.com>
References: <20190205212829.GA20127@alvherre.pgsql>
 <4572.1549404528@sss.pgh.pa.us>
 <CAH2-WzmRtQLJdC7KSdtwxQ65Yk+_Vi89eVqKJvPOx=Xc7WLLKw@mail.gmail.com>
Comments: In-reply-to Peter Geoghegan <pg@bowt.ie>
	message dated "Tue, 05 Feb 2019 14:50:34 -0800"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <20443.1549423295.1@sss.pgh.pa.us>
Date: Tue, 05 Feb 2019 22:21:35 -0500
Message-ID: <20444.1549423295@sss.pgh.pa.us>
Precedence: bulk

Peter Geoghegan <pg@bowt.ie> writes:
> On Tue, Feb 5, 2019 at 2:08 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> I've got much of the code for it already (in the wreckage of my failed
>> attempts), so I'll go back and finish that up.  I was just waiting to see
>> how loudly people would howl about using object type as a condition for
>> figuring out what a pg_depend entry really means.  If we're okay with
>> that hack, I think I can make it work.

> Perhaps I've missed some subtlety, but I'm not sure that it's all that
> ugly. If splitting INTERNAL_AUTO into two new dependency types amounts
> to the same thing as what you suggest here, then what's the
> difference?

It's the same as long as we always think that the "real owner" of a
subsidiary object is of the same type as that object, eg that the
real owner of a per-partition trigger is the parent trigger, the real
owner of a per-partition index is the parent index, etc ... and that
there's only going to be one parent object of that type.

I don't immediately have a counterexample to this, which is why I feel
like this is an OK solution for now.  But I'm not sure it'll hold good
indefinitely.

Actually, the index case is already on the edge of being a problem:
both parents will be relations, and I suspect the code will have to
look at relkinds to identify which parent to consider the "real owner".

BTW, does anyone else feel like our terminology around partitions is
a dead loss?  I have not seen anybody use the word in a way that makes
a clear distinction between the parent "partitioned" object and the
child "partition" objects, at least not when it comes to subsidiary
objects like triggers.

			regards, tom lane