I use UTF-8 in my editor, so all strings displayed here are UTF-8 in file.

I have a python script like this:

# -*- coding: utf-8 -*-
...
parser = optparse.OptionParser(description=_('automates the dice rolling in the classic game "risk"'), usage=_("usage: %prog attacking defending"))

Then I used xgettext to get everything out and got a .pot file which can be boiled down to:

"Content-Type: text/plain; charset=CHARSET\n"
"Content-Transfer-Encoding: 8bit\n"#: auto_dice.py:16
msgid "automates the dice rolling in the classic game \"risk\""
msgstr ""

After that, I used msginit to get a de.po which I filled in like this:

"Content-Type: text/plain; charset=UTF-8\n"
"Content-Transfer-Encoding: 8bit\n"#: auto_dice.py:16
msgid "automates the dice rolling in the classic game \"risk\""
msgstr "automatisiert das Würfeln bei \"Risiko\""

Running the script, I get the following error:

  File "/usr/lib/python2.6/optparse.py", line 1664, in print_helpfile.write(self.format_help().encode(encoding, "replace"))
UnicodeDecodeError: 'ascii' codec can't decode byte 0xc3 in position 60: ordinal not in range(128)

How can I fix that?

That error means you've called encode on a bytestring, so it tries to decode it to Unicode using the system default encoding (ascii on Python 2), then re-encode it with whatever you've specified.

Generally, the way to resolve it is to call s.decode('utf-8') (or whatever encoding the strings are in) before trying to use the strings. It might also work if you just use unicode literals: u'automates...' (that depends on how strings are substituted from .po files, which I don't know about).

This sort of confusing behaviour is improved in Python 3, which won't try to convert bytes to unicode unless you specifically tell it to.