From: Tom Lane <tgl@sss.pgh.pa.us>
To: Andrew Dunstan <andrew.dunstan@2ndquadrant.com>
cc: Peter Eisentraut <peter.eisentraut@2ndquadrant.com>,
        pgsql-committers@lists.postgresql.org
Subject: Re: pgsql: Prevent infinity and NaN in jsonb/plperl transform
In-reply-to: <213242a0-0ed8-ade6-822c-0270f7064799@2ndQuadrant.com>
References: <E1fDCX9-0005yu-7q@gemulon.postgresql.org>
 <21194.1525110966@sss.pgh.pa.us>
 <59e5787f-314e-6e5d-db5e-bcb6b9dbc25c@2ndquadrant.com>
 <23691.1525113718@sss.pgh.pa.us>
 <213242a0-0ed8-ade6-822c-0270f7064799@2ndQuadrant.com>
Comments: In-reply-to Andrew Dunstan <andrew.dunstan@2ndquadrant.com>
	message dated "Mon, 30 Apr 2018 15:13:05 -0400"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <28584.1525131438.1@sss.pgh.pa.us>
Date: Mon, 30 Apr 2018 19:37:18 -0400
Message-ID: <28585.1525131438@sss.pgh.pa.us>
Precedence: bulk

Andrew Dunstan <andrew.dunstan@2ndquadrant.com> writes:
> Fingers crossed we'll be OK. Looking at the perl headers on a Windows
> machine I'm doing some other stuff on, the #define is protected by an
> #ifndef. I'm always happy when my hacks can be removed :-)

Well, the early returns on that are promising, but we have another
problem: prairiedog, jacana, and probably other old BF members are
failing the new regression test cases.  It appears that what the test
script assumes will produce a NaN or Inf float doesn't work in 5.8.x
perl:

$ perl -e "print 0 + 'NaN'"
0
$ perl -e "print 0 + 'Inf'"
0

I googled a bit and found these recommendations on stackoverflow:

my $inf    = 9**9**9;
my $neginf = -9**9**9;
my $nan    = -sin(9**9**9);

These do seem to produce the desired results, at least on the
couple of Perl versions I checked, including 5.8.3.

			regards, tom lane