Hello, hackers.
On 18/04/2023 20:34, Tom Lane wrote (on pgsql-committers):
> I shall now retire to a safe distance and watch the buildfarm.
Unfortunately, on fresh perl (5.38.2 verified) and on ru_RU.UTF-8
locale, it breaks basic float comparison: 0 < 0.5 is no longer true.
This is the reproduction on REL_16_STABLE (but it affects master
as well), using fresh Ubuntu 24.04 container.
0. I've used lxc to get a fresh container:
$ lxc launch ubuntu-daily:noble u2404
But I don't think lxc or containerization in general matters in this
case. Also, I think any environment with fresh enough Perl would work,
Ubuntu 24.04 is just an easy example.
(obviously, install necessary dev packages)
1. Generate ru_RU.UTF-8 locale:
a. In /etc/locale.gen, uncomment the line:
# ru_RU.UTF-8 UTF-8
b. Run locale-gen as root. For me, it says:
$ sudo locale-gen
Generating locales (this might take a while)...
en_US.UTF-8... done
ru_RU.UTF-8... done
Generation complete.
2. Apply 0001-demo-of-weird-Perl-setlocale-effect-on-float-numbers.patch
(adding src/test/authentication/t/999_broken.pl)
3. Run the test
LANG=ru_RU.UTF-8 make check -C src/test/authentication
PROVE_TESTS=t/999_broken.pl PROVE_FLAGS=--verbose
The test is, basically:
use PostgreSQL::Test::Utils;
use Test::More tests => 1;
ok(0 < 0.5, "0 < 0.5");
If I comment-out the "use PostgreSQL::Test::Utils" line, the test works.
Otherwise it fails to notice that 0 is less than 0.5.
Alternatively, the test fails if I replace that "use" line with
BEGIN {
use POSIX qw(locale_h);
setlocale(LC_NUMERIC, "");
}
"BEGIN" part is essential: mere use/setlocale is fine.
Also, adding
use locale;
or even
use locale ':numeric';
fixes the test, but I doubt whether it's a good idea to add that to
Utils.pm.
Obviously, one of the reasons is that according to ru_RU.UTF-8 locale
for LC_NUMERIC, fractional part separator is ",", not ".". So one could,
technically, parse "0.5" as "0" and then unparsed ".5" tail. I think it
might even be a Perl bug, because, according to my quick browsing of man
perlfunc (setlocale) and man perllocale, this should not affect the code
outside "use locale", not in such a fundamental way. After all, we're
talking not about strtod etc, but about floating-point numbers in the
source code.
P.S. $ perl --version
This is perl 5, version 38, subversion 2 (v5.38.2) built for
x86_64-linux-gnu-thread-multi
(with 44 registered patches, see perl -V for more detail)
P.P.S. I'm replying to pgsql-hackers, even though part of previous
discussion have been on pgsql-committers. Hopefully, it's OK.
--
Anton Voloshin
Postgres Professional, The Russian Postgres Company
https://postgrespro.ru