Noone else posted anything, so I'll give it a quick go...

It seems you're assuming Perl's internal format is UTF-8.

To be sure, it would help if I knew exactly what you were getting from the database. The "PV = " lines printed by following would provide that info.

use Devel::Peek qw( Dump );
Dump($db_field);  # Output sent to STDERR
[download]

You should get one of the following four output combinations for chr(259) and chr(238).

259: PV = 0x18e914c "\304\203"\0 [UTF8 "\x{103}"]
238: PV = 0x18e914c "\356"\0

259: PV = 0x18e914c "\304\203"\0 [UTF8 "\x{103}"]
238: PV = 0x18ecffc "\303\256"\0 [UTF8 "\x{ee}"]

259: PV = 0x18d73fc "?"\0
238: PV = 0x18d7444 "\356"\0

259: PV = 0x18d736c "\304\203"\0
238: PV = 0x18ecffc "\303\256"\0

259: PV = 0x18d74dc "\304\203"\0     <- Lack of [UTF8 ...]
238: PV = 0x18d74dc "\356"\0
[download]

• 259: PV = 0x18e914c "\304\203"\0 [UTF8 "\x{103}"]
  238: PV = 0x18e914c "\356"\0
  [download]

  or

  259: PV = 0x18e914c "\304\203"\0 [UTF8 "\x{103}"]
  238: PV = 0x18ecffc "\303\256"\0 [UTF8 "\x{ee}"]
  [download]

  The data is in Perl's internal text format (called "UTF8", no dash). It can either be in iso-latin-1 or UTF-8. iso-latin-1 is preferred since it's faster, but it can only be used if every character in the string can be represented by iso-latin-1.

  Solution: You need to convert from Perl's internal format to a string of bytes.

  use Encode qw( encode );
  print(encode('UTF-8', $text));
  [download]

  or

  binmode(STDOUT, ':encoding(UTF-8)');
  print($text);
  [download]

• 259: PV = 0x18d73fc "?"\0
  238: PV = 0x18d7444 "\356"\0
  [download]

  The data is in iso-latin-1. This can't be the case, since "ă" is working.

• 259: PV = 0x18d736c "\304\203"\0
  238: PV = 0x18ecffc "\303\256"\0
  [download]

  The data is in UTF-8 already. Just print it out without re-encoding it.

• 259: PV = 0x18d74dc "\304\203"\0     <- Lack of [UTF8 ...]
  238: PV = 0x18d74dc "\356"\0
  [download]

  That means the data in your database is bad, or your database is returning bad data.

  The problem should be fixed at the source, but I think it's possible to fix it at this point too.

Update: Added second pair of output combinations, and fixed s/utf8/utf-8/