Dying with non-ASCII utf8 DB message, with use warnings FATAL => 'all'; leads to "Wide character in subroutine entry" error message with DB-message omitted

[nataraj-hates-MS-for-stealing-github]

Update: see #108 (comment) below, to get up to date bug description.

I've stabled a problem:
I have DB initialized with non-English (and not Latin) utf8 locale. (It is Russian, but this is not really importaint)

I am connecting to the DB via network (this one is important, peer connection works well)

I am running a query that fails with an error.

Program dies, (as I have RaiseError => 1) but does not print any error text, it is only prints Wide character in print error message.

When I reinitialized DB cluster running it with en_US.UTF8 locale, this problem have gone. I got proper error message.

Main question: is this problem known? If it is not, or known but did not fixed yet, I will explore it more, and provide more detailed but report.

[esabol]

Your report is unclear. Could you specify the steps to reproduce and provide a short Perl program which shows the problem? Thank you!

[nataraj-hates-MS-for-stealing-github]

Hm... Tried to write step by step guide how to reproduce it, and found out that I can not reproduce this from scratch on clean environment... Will need more time to revert my dev server and project I am working on to original configuration, where I met this issue, and then will try to reproduce it there, finding the key point... This might take some time

[esabol]

Understood. Thank you for your efforts. In the meantime perhaps someone else will have some insight into the issue.

[nataraj-hates-MS-for-stealing-github]

I found the cause...

If you set use warnings FATAL => 'all'; in your program, and try to run query that fails, on DB with non-latin utf8 locale for error messages, you will get Wide character in subroutine entry warning, and thus program will be terminated before real error message will be printed.

Quite a good way to shoot own leg...

But I guess this Wide character in subroutine entry warning should be fixed anyway, if it is possible, I guess...

Do you need more detailed instructions for bug reproduction?

[esabol]

Do you need more detailed instructions for bug reproduction?

Well, it wouldn't hurt....

[nataraj-hates-MS-for-stealing-github]

On clean Debian (or Debian based) installation do:

sudo dpkg-reconfigure locales

add ru_RU.UTF-8 and choose it as default locale.

Exit shell and login again.

sudo apt-get install postgresql libdbd-pg-perl

sudo su postgres
createuser -P test-user  #set password 123
createdb -O test-user test-db
echo "create table test(i integer not null)" | psql -U test-user -h 127.0.0.1 test-db

Run following script:

#!/usr/bin/perl

use strict;
use warnings FATAL => 'all';
use DBD::Pg;

my $dbh = DBI->connect("dbi:Pg:dbname=test-db;host=127.0.0.1", "test-user", "123", {
    PrintError       => 0,
    RaiseError       => 1,
    AutoCommit       => 1,
    FetchHashKeyName => 'NAME_lc',
}) or die("Can't access the database");

my $sth = $dbh->prepare( "INSERT INTO test (i) values (?)");
$sth->bind_param(1, undef);
$sth->execute();

It will give you

Wide character in subroutine entry at ./test3.pl line 17.

This a real problem, you will have hard time finding the real cause of why program fails.

Now comment out use warnings FATAL => 'all'; and run script again. It will give you

Wide character in subroutine entry at ./test3.pl line 17.
DBD::Pg::st execute failed: ОШИБКА:  значение NULL в столбце "i" отношения "test" нарушает ограничение NOT NULL
DETAIL:  Ошибочная строка содержит (null). at ./test3.pl line 17.

Two last lines is expected error message in Russian, and first line should not be there from my point of view.