Date: Sat, 24 Mar 2018 14:25:02 -0700
From: Andres Freund <andres@anarazel.de>
To: Tom Lane <tgl@sss.pgh.pa.us>, Peter Eisentraut <peter_e@gmx.net>
Cc: pgsql-committers@postgresql.org, pgsql-hackers@postgresql.org
Subject: Re: pgsql: Avoid premature free of pass-by-reference CALL arguments.
Message-ID: <20180324212502.wt4serghfidge2on@alap3.anarazel.de>
References: <E1eka1V-0002Ed-A3@gemulon.postgresql.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <E1eka1V-0002Ed-A3@gemulon.postgresql.org>
Precedence: bulk


On 2018-02-10 18:37:17 +0000, Tom Lane wrote:
> Avoid premature free of pass-by-reference CALL arguments.

> src/test/regress/expected/create_procedure.out | 12 +++++++----
> src/test/regress/sql/create_procedure.sql      |  4 +++-
> 3 files changed, 33 insertions(+), 11 deletions(-)

There's a recent llvm buildfarm animal failure related to this:
https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=pogona&dt=2018-03-24%2020%3A10%3A01

*** /home/bf/build/buildfarm-pogona/HEAD/pgsql.build/../pgsql/src/test/regress/expected/create_procedure.out	2018-03-23 08:10:44.326010286 +0100
--- /home/bf/build/buildfarm-pogona/HEAD/pgsql.build/src/test/regress/results/create_procedure.out	2018-03-24 21:15:28.749352165 +0100
***************
*** 44,50 ****
  SELECT * FROM cp_test ORDER BY b COLLATE "C";
   a |   b   
  ---+-------
!  1 | 0
   1 | a
   1 | xyzzy
  (3 rows)
--- 44,50 ----
  SELECT * FROM cp_test ORDER BY b COLLATE "C";
   a |   b   
  ---+-------
!  1 | 9
   1 | a
   1 | xyzzy
  (3 rows)

With the differening output created by:

CREATE PROCEDURE ptest1(x text)
LANGUAGE SQL
AS $$
INSERT INTO cp_test VALUES (1, x);
$$;
CALL ptest1(substring(random()::text, 1, 1));  -- ok, volatile arg

At first I was gosh darned confused, this really didn't seem likely to
be an LLVM induced failure. And it turns out it isn't.  If the value
returned by random() is very small, the text representation switches to
scientific notation like 8.26204195618629e-05.

So, perhaps we should choose a different volatile function here?

Greetings,

Andres Freund