From: Tom Lane <tgl@sss.pgh.pa.us>
To: Alvaro Herrera <alvherre@2ndquadrant.com>
cc: Peter Geoghegan <pg@bowt.ie>,
        PostgreSQL Hackers <pgsql-hackers@postgresql.org>
Subject: Re: Fixing findDependentObjects()'s dependency on scan order
 (regressions in DROP diagnostic messages)
In-reply-to: <20190208201925.GA5901@alvherre.pgsql>
References: <20190208201925.GA5901@alvherre.pgsql>
Comments: In-reply-to Alvaro Herrera <alvherre@2ndquadrant.com>
	message dated "Fri, 08 Feb 2019 17:19:25 -0300"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <29688.1549672900.1@sss.pgh.pa.us>
Date: Fri, 08 Feb 2019 19:41:40 -0500
Message-ID: <29689.1549672900@sss.pgh.pa.us>
Precedence: bulk

Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> On 2019-Feb-08, Tom Lane wrote:
>> Also, I came across some coding in CloneFkReferencing() that looks fishy
>> as hell: that function imagines that it can delete an existing trigger
>> with nothing more than a summary CatalogTupleDelete().  I didn't do
>> anything about that here, but if it's not broken, I'd like to see an
>> explanation why not.  I added a comment complaining about the lack of
>> pg_depend cleanup, and there's also the question of whether we don't
>> need to broadcast a relcache inval for the trigger's table.

> Oops, this is new code in 0464fdf07f69 (Jan 21st).  Unless you object,
> I'll study a fix for this now, to avoid letting it appear in the minor
> next week.

+1.  The best solution would presumably be to go through the normal
object deletion mechanism; though possibly there's a reason that
won't work given you're already inside some other DDL.

			regards, tom lane