Date: Thu, 2 Sep 2021 22:56:01 -0500
From: Justin Pryzby <pryzby@telsasoft.com>
To: Tomas Vondra <tomas.vondra@enterprisedb.com>
Cc: Dean Rasheed <dean.a.rasheed@gmail.com>, pgsql-hackers@postgresql.org
Subject: Re: PoC/WIP: Extended statistics on expressions
Message-ID: <20210903035601.GT26465@telsasoft.com>
References: <20201123022604.GD24052@telsasoft.com>
 <e5beb5d5-28b4-7aad-3a16-9a6abf57f897@enterprisedb.com>
 <6455e0ad-2a2e-bdb0-e1f8-fd4983d1c8aa@enterprisedb.com>
 <087086de-1384-8374-4efc-2e7915071d5c@enterprisedb.com>
 <CAEZATCU9uPo7JYdx4k0-ufXXZH8t7itodibUwCva+s+AvAKcnw@mail.gmail.com>
 <b2995773-c9a1-5d3b-fc90-4f3ab189be11@enterprisedb.com>
 <20210816013255.GS10479@telsasoft.com>
 <a2be7e05-e6b7-1467-d15a-49011e45a4b8@enterprisedb.com>
 <20210824131320.GO10479@telsasoft.com>
 <6bc0ab0c-a807-2b7f-7df8-ac90c7839013@enterprisedb.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <6bc0ab0c-a807-2b7f-7df8-ac90c7839013@enterprisedb.com>
User-Agent: Mutt/1.9.4 (2018-02-28)
Archived-At: 
 <https://www.postgresql.org/message-id/20210903035601.GT26465%40telsasoft.com>
Precedence: bulk

On Wed, Sep 01, 2021 at 06:45:29PM +0200, Tomas Vondra wrote:
> However while polishing the second patch, I realized we're allowing
> statistics on expressions referencing system attributes. So this fails;
> 
> CREATE STATISTICS s ON ctid, x FROM t;
> 
> but this passes:
> 
> CREATE STATISTICS s ON (ctid::text), x FROM t;
> 
> IMO we should reject such expressions, just like we reject direct references
> to system attributes - patch attached.

Right, same as indexes.  +1

> Furthermore, I wonder if we should reject expressions without any Vars? This
> works now:
> 
> CREATE STATISTICS s ON (11:text) FROM t;
> 
> but it seems rather silly / useless, so maybe we should reject it.

To my surprise, this is also allowed for indexes...

But (maybe this is what I was remembering) it's prohibited to have a constant
expression as a partition key.

Expressions without a var seem like a case where the user did something
deliberately silly, and dis-similar from the case of making a stats expression
on a simple column - that seemed like it could be a legitimate
mistake/confusion (it's not unreasonable to write an extra parenthesis, but
it's strange if that causes it to behave differently).

I think it's not worth too much effort to prohibit this: if they're determined,
they can still write an expresion with a var which is constant.  I'm not going
to say it's worth zero effort, though.

-- 
Justin