Horton

Re: Procedures versus the "fastpath" API

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Michael Paquier <michael@paquier.xyz>
Cc: Joe Conway <mail@joeconway.com>, pgsql-hackers@lists.postgresql.org
Date: 2021-04-30T16:57:29Z
Lists: pgsql-hackers 
Michael Paquier <michael@paquier.xyz> writes:
> On Tue, Mar 09, 2021 at 02:33:47PM -0500, Joe Conway wrote:
>> My vote would be reject using fastpath for procedures in all relevant branches.
>> If someday someone cares enough to make it work, it is a new feature for a new
>> major release.

> FWIW, my vote would go for issuing an error if attempting to use a
> procedure in the fast path for all the branches.  The lack of
> complaint about the error you are mentioning sounds like a pretty good
> argument to fail properly on existing branches, and work on this as a
> new feature in the future if there is anybody willing to make a case
> for it.

I let this thread grow cold because I was hoping for some more votes,
but with the quarterly releases approaching, it's time to close out
the issue one way or the other.

By my count, we have three votes for forbidding procedure calls via
fastpath in all branches (me, Joe, Michael), and two for doing
something laxer (Noah, Laurenz).  The former is surely the safer
choice, so I'm going to go do that.

			regards, tom lane

Commits

  1. Doc: update libpq's documentation for PQfn().

  2. Disallow calling anything but plain functions via the fastpath API.