From: Tom Lane <tgl@sss.pgh.pa.us>
To: Amit Langote <amitlangote09@gmail.com>
cc: PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Subject: Re: unsupportable composite type partition keys
In-reply-to: 
 <CA+HiwqHYfc8taAAE2O46UbDRR3VQ7FEZYyzrHDi60a4D-0Vp6w@mail.gmail.com>
References: 
 <CA+HiwqFUzjfj9HEsJtYWcr1SgQ_=iCAvQ=O2Sx6aQxoDu4OiHw@mail.gmail.com>
 <5361.1576602769@sss.pgh.pa.us>
 <CA+HiwqGiGBgFFcL6YGKRqfX1-gYKmuLLG_JwU_UaSaX2J+jehQ@mail.gmail.com>
 <28160.1576676326@sss.pgh.pa.us>
 <CA+HiwqENLY2ta0goA0jcfk4cWdj2W0r6Ywjf=PQiMpJkyXygmg@mail.gmail.com>
 <26766.1576947735@sss.pgh.pa.us> <24653.1576962784@sss.pgh.pa.us>
 <CA+HiwqGug69_zhRCzENJPg_Ynpx_Eb-2-=N1u2Zw3X6T1S3wbQ@mail.gmail.com>
 <25383.1577112556@sss.pgh.pa.us> <26572.1577113220@sss.pgh.pa.us>
 <CA+HiwqHYfc8taAAE2O46UbDRR3VQ7FEZYyzrHDi60a4D-0Vp6w@mail.gmail.com>
Comments: In-reply-to Amit Langote <amitlangote09@gmail.com>
	message dated "Tue, 24 Dec 2019 10:20:28 +0900"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <3227.1577209331.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Tue, 24 Dec 2019 12:42:11 -0500
Message-ID: <3228.1577209331@sss.pgh.pa.us>
Precedence: bulk

Amit Langote <amitlangote09@gmail.com> writes:
> On Tue, Dec 24, 2019 at 12:00 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> BTW, I forgot to mention: while I think the patch to forbid pseudotypes
>> by using CheckAttributeType() can be back-patched, I'm leaning towards
>> not back-patching the other patch.  The situation where we get into
>> infinite recursion seems not very likely in practice, and it's not
>> going to cause any crash or data loss, so I think we can just say
>> "sorry that's not supported before v13".  The patch as I'm proposing
>> it seems rather invasive for a back-branch fix.

> It is indeed.

> Just to be sure, by going with "unsupported before v13", which one do yo=
u mean:

> * documenting it as so
> * giving an error in such cases, like the patch in the first email on
> this thread did
> * doing nothing really

I was thinking "do nothing in the back branches".  I don't believe we
can detect such cases reliably (at least not without complicated logic,
which'd defeat the point), so I don't think giving an error is actually
feasible, and I doubt that documenting it would be useful.  If we get
some field complaints about this, it'd be time enough to reconsider.

			regards, tom lane