|Welcome to the Monastery|
DBIx::Log4perl sees the data before it is sent to the database by DBD::Pg so you cannot rely on what you see in its output as DBD::Pg can change the data.
In theory, yes, but practically Log4perl was logging as expected a line before the SQL statement and the output of DBIx makes sense for the problem I described. If I encode the data to WIN1252 before passing it to DBI the logged output of DBIx change as well, therefore my strong guess that DBIx is logging what gets send to the database.
Are you binding the data as parameters when it is inserted?
No, the application simply creates a SQL string by concatenating different UTF-8 Perlstrings together, pushes that with a "do" and the whole string formed by SQL command and values gets encoded to UTF-8.
What is your postgres client charset set to
It's automatically detected as WIN1252 when the problem occurs, which makes perfectly sense on Windows and the Linux I tested. Another Ubuntu server automatically detects it as UTF-8 instead, because it has an UTF-8 locale. If I manually change it to UTF-8 everything works as expected, the database server properly encodes UTF-8 bytes to WIN1252 characters in the target database.
and what is the encoding of the data you pass to DBD::Pg when it fails
Valid UTF-8 Perlstrings with UTF-8 flag turned on.
and do you have pg_enable_utf8 on?
No, because from my understanding it is deprecated and only used in reading from the database, not writing to.
In reply to Re^4: DBD::Pg encodes Perlstring to UTF-8 bytes instead of WIN1252 regardless client encoding