I18n issue with optik
Thorsten Kampe
I've written a script which uses Optik/Optparse to display the
options (which works fine). The text for the help message is localised
(with german umlauts) and when I execute the script with the localised
environment variable set, I get this traceback[1]. The interesting
thing is that the localised optparse messages from displays fine -
it's only my localisation that errors.
From my understanding, my script doesn't put out anything, it's
optik/optparse who does that. My po file is directly copied from the
optik po file (who displays fine) and modified so the po file should
be fine, too.
What can I do to troubleshoot whether the culprit is my script, optik
or gettext?
Would it make sense to post the script and the mo or po files?
Thorsten
[1]
Traceback (most recent call last):
File "script.py", line 37, in <module>
options, args = cmdlineparser.parse_args()
File "/usr/lib/python2.5/optparse.py", line 1378, in parse_args
stop = self._process_args(largs, rargs, values)
File "/usr/lib/python2.5/optparse.py", line 1418, in _process_args
self._process_long_opt(rargs, values)
File "/usr/lib/python2.5/optparse.py", line 1493, in
_process_long_opt
option.process(opt, value, values, self)
File "/usr/lib/python2.5/optparse.py", line 782, in process
self.action, self.dest, opt, value, values, parser)
File "/usr/lib/python2.5/optparse.py", line 804, in take_action
parser.print_help()
File "/usr/lib/python2.5/optparse.py", line 1648, in print_help
file.write(self.format_help().encode(encoding, "replace"))
UnicodeDecodeError: 'ascii' codec can't decode byte 0xc4 in position
264: ordinal not in range(128)
Steven Bethard
> I've written a script which uses Optik/Optparse to display the
> options (which works fine). The text for the help message is localised
> (with german umlauts) and when I execute the script with the localised
> environment variable set, I get this traceback[1]. The interesting
> thing is that the localised optparse messages from displays fine -
> it's only my localisation that errors.
>
> From my understanding, my script doesn't put out anything, it's
> optik/optparse who does that. My po file is directly copied from the
> optik po file (who displays fine) and modified so the po file should
> be fine, too.
>
> What can I do to troubleshoot whether the culprit is my script, optik
> or gettext?
>
> Would it make sense to post the script and the mo or po files?
Yes, probably. Though if you can reduce it to the simplest test case
that produces the error, it'll increase your chances of having someone
look at it.
You could also try posting to the optik list:
http://lists.sourceforge.net/lists/listinfo/optik-users
STeVe
Thorsten Kampe
> Thorsten Kampe wrote:
> > I've written a script which uses Optik/Optparse to display the
> > options (which works fine). The text for the help message is localised
> > (with german umlauts) and when I execute the script with the localised
> > environment variable set, I get this traceback[1]. The interesting
> > thing is that the localised optparse messages from displays fine -
> > it's only my localisation that errors.
> >
> > From my understanding, my script doesn't put out anything, it's
> > optik/optparse who does that. My po file is directly copied from the
> > optik po file (who displays fine) and modified so the po file should
> > be fine, too.
> >
> > What can I do to troubleshoot whether the culprit is my script, optik
> > or gettext?
> >
> > Would it make sense to post the script and the mo or po files?
>
> Yes, probably. Though if you can reduce it to the simplest test case
> that produces the error, it'll increase your chances of having someone
> look at it.
The most simple test.py is:
###
#! /usr/bin/env python
import gettext, \
os, \
sys
gettext.textdomain('optparse')
gettext.install('test')
from optparse import OptionParser, \
OptionGroup
cmdlineparser = OptionParser(description = _('THIS SOFTWARE COMES
WITHOUT WARRANTY, LIABILITY OR SUPPORT!'))
options, args = cmdlineparser.parse_args()
###
When I run LANGUAGE=de ./test.py --help I get the error.
### This is the test.de.po file
# Copyright (C) 2006 Thorsten Kampe
# Thorsten Kampe <thor...@thorstenkampe.de>, 2006
msgid ""
msgstr ""
"Project-Id-Version: Template 1.0\n"
"POT-Creation-Date: Tue Sep 7 22:20:34 2004\n"
"PO-Revision-Date: 2005-07-03 16:47+0200\n"
"Last-Translator: Thorsten Kampe <thor...@thorstenkampe.de>\n"
"Language-Team: Thorsten Kampe <thor...@thorstenkampe.de>\n"
"MIME-Version: 1.0\n"
"Content-Type: text/plain; charset=ISO-8859-15\n"
"Content-Transfer-Encoding: 8-bit\n"
"Generated-By: pygettext.py 1.5\n"
msgid "THIS SOFTWARE COMES WITHOUT WARRANTY, LIABILITY OR SUPPORT!"
msgstr "DIESES PROGRAMM HAT WEDER GEWÄHRLEISTUNG, HAFTUNG NOCH
UNTERSTÜTZUNG!"
###
The localisation now produces an error in the localised optik files,
too.
Under Windows I get " File "G:\program files\python\lib\encodings
\cp1252.py", line 12, in encode
return codecs.charmap_encode(input,errors,encoding_table)"
Is there something I have to do to put the terminal in "non-ascii
output mode"?
I tried
###
#! /usr/bin/env python
# -*- coding: ISO-8859-15 -*-
print "DIESES PROGRAMM HAT WEDER GEWÄHRLEISTUNG, HAFTUNG NOCH
UNTERSTÜTZUNG!"
###
...and this worked. That means that my terminal is willing to print,
right?!
> You could also try posting to the optik list:
> http://lists.sourceforge.net/lists/listinfo/optik-users
I already did this via Gmane (although the list seems pretty dead to
me). Sourceforge seems to have a bigger problem as [1] and [2] error.
Sorry for the confusion but this Unicode magic is far from being
rational. I guess most people just don't get it...
Thorsten
[1] http://sourceforge.net/mailarchive/forum.php?forum=optik-users
[2] https://lists.sourceforge.net/lists/listinfo
Thorsten Kampe
print _('THIS SOFTWARE COMES WITHOUT WARRANTY, LIABILITY OR SUPPORT!')
into this skript, the line is printed out. So if my Skript can output
the localised text but Optparse can't it should be an optparse bug,
right?!
Thorsten
Thorsten Kampe
# used by test suite
def _get_encoding(self, file):
encoding = getattr(file, "encoding", None)
if not encoding:
encoding = sys.getdefaultencoding()
return encoding
def print_help(self, file=None):
"""print_help(file : file = stdout)
Print an extended help message, listing all options and any
help text provided with them, to 'file' (default stdout).
"""
if file is None:
file = sys.stdout
encoding = self._get_encoding(file)
file.write(self.format_help().encode(encoding, "replace"))
So this means: when the encoding of sys.stdout is US-ASCII, Optparse
sets the encoding to of the help text to ASCII, too. But that's
nonsense because the Encoding is declared in the Po (localisation)
file.
How can I set the encoding of sys.stdout to another encoding? Of
course this would be a terrible hack if the encoding of the
localisation changes or different translators use different
encodings...
Thorsten
Steven Bethard
I'm not very experienced with internationalization, but if you change::
gettext.install('test')
to::
gettext.install('test', unicode=True)
what happens?
STeVe
Steven Bethard
If print_help() is what's wrong, you should probably hack print_help()
instead of sys.stdout. You could try something like::
def print_help(self, file=None):
"""print_help(file : file = stdout)
Print an extended help message, listing all options and any
help text provided with them, to 'file' (default stdout).
"""
if file is None:
file = sys.stdout
file.write(self.format_help())
optparse.OptionParser.print_help = print_help
cmdlineparser = optparse.OptionParser(description=...)
...
That is, you could monkey-patch print_help() before you create an
OptionParser.
STeVe
Thorsten Kampe
No traceback anymore from optparse but the non-ascii umlauts are
displayed as question marks ("?").
Thorsten
Thorsten Kampe
Yes, I could do that but I'd rather know first if my code is wrong or
the optparse code.
Thorsten
Thorsten Kampe
> Yes, I could do that but I'd rather know first if my code is wrong or
> the optparse code.
It might be the bug mentioned in
http://mail.python.org/pipermail/python-dev/2006-May/065458.html
The patch although doesn't work. From my unicode-charset-codepage-
codeset-challenged point of view the encoding of sys.stdout doesn't
matter. The charset is defined in the .po/.mo file (but of course
optparse can't know if the message has been translated by gettext
("_").
Thorsten
Thorsten Kampe
If I "patch" line 1648 (the one mentioned in the traceback) of
optparse.py from
file.write(self.format_help().encode(encoding, "replace"))
to
file.write(self.format_help())
...then everything works and is displayed fine (even without the
"unicode = True" parameter to gettext.install).
But the "patch" might make other things fail, of course...
Thorsten
Thorsten Kampe
> * Thorsten Kampe (Sun, 1 Apr 2007 20:08:39 +0100)
> > * Thorsten Kampe (Sun, 1 Apr 2007 19:45:59 +0100)
> > > Yes, I could do that but I'd rather know first if my code is wrong or
> > > the optparse code.
> >
> > It might be the bug mentioned in
> > http://mail.python.org/pipermail/python-dev/2006-May/065458.html
> >
> > The patch although doesn't work. From my unicode-charset-codepage-
> > codeset-challenged point of view the encoding of sys.stdout doesn't
> > matter. The charset is defined in the .po/.mo file (but of course
> > optparse can't know if the message has been translated by gettext
> > ("_").
>
> If I "patch" line 1648 (the one mentioned in the traceback) of
> optparse.py from
>
> file.write(self.format_help().encode(encoding, "replace"))
> to
> file.write(self.format_help())
>
...but only in Cygwin rxvt, the standard Windows console doesn't show
the right colors.
I give up and revert back to ASCII. This whole charset mess is not
meant to solved by mere mortals.
Thorsten
Jarek Zgoda
>>> Under Windows I get " File "G:\program files\python\lib\encodings
>>> \cp1252.py", line 12, in encode
>>> return codecs.charmap_encode(input,errors,encoding_table)"
>> I'm not very experienced with internationalization, but if you change::
>>
>> gettext.install('test')
>>
>> to::
>>
>> gettext.install('test', unicode=True)
>>
>> what happens?
>
> No traceback anymore from optparse but the non-ascii umlauts are
> displayed as question marks ("?").
And this is expected behaviour of encode() with errors set to 'replace'.
I think this is the solution to your problem. I was a bit surprised I
never saw this error, but I always use the unicode=True setting to
gettext.install()...
--
Jarek Zgoda
http://jpa.berlios.de/
Leo Kislov
> I guess the culprit is this snippet from optparse.py:
>
> # used by test suite
> def _get_encoding(self, file):
> encoding = getattr(file, "encoding", None)
> if not encoding:
> encoding = sys.getdefaultencoding()
> return encoding
>
> def print_help(self, file=None):
> """print_help(file : file = stdout)
>
> Print an extended help message, listing all options and any
> help text provided with them, to 'file' (default stdout).
> """
> if file is None:
> file = sys.stdout
> encoding = self._get_encoding(file)
> file.write(self.format_help().encode(encoding, "replace"))
>
> So this means: when the encoding of sys.stdout is US-ASCII, Optparse
> sets the encoding to of the help text to ASCII, too.
.encode() method doesn't set an encoding. It encodes unicode text into
bytes according to specified encoding. That means optparse needs ascii
or unicode (at least) for help text. In other words you'd better use
unicode throughout your program.
> But that's
> nonsense because the Encoding is declared in the Po (localisation)
> file.
For backward compatibility gettext is working with bytes by default,
so the PO file encoding is not even involved. You need to use unicode
gettext.
> How can I set the encoding of sys.stdout to another encoding?
What are you going to set it to? As I understand you're going to
distribute your program to some users. How are you going to find out
the encoding of the terminal of your users?
-- Leo
Thorsten Kampe
You mean
gettext.install('test', unicode = True)
and
description = _(u'THIS SOFTWARE COMES WITHOUT WARRANTY, LIABILITY OR
SUPPORT!') ?
If I modify my code like this, I don't get any traceback anymore, but
the non-ascii umlauts are still displayed as question marks.
Thorsten
Thorsten Kampe
> Thorsten Kampe napisa?(a):
I can't see the "solution" here. Is the optparse "print_help" function
wrong? Why should there even be errors if I use "unicode = True" with
gettext.install?
I have ISO-8859-15 gettext translations and I want optparse to display
them correctly. What do I have to do?
Thorsten
Jarek Zgoda
> I can't see the "solution" here. Is the optparse "print_help" function
> wrong? Why should there even be errors if I use "unicode = True" with
> gettext.install?
>
> I have ISO-8859-15 gettext translations and I want optparse to display
> them correctly. What do I have to do?
Please, see gettext module documentation on this topic.
The solution is: always install your translation with unicode=True
setting. This assures usage of ugettext() instead of gettext() and works
properly with character sets other than ASCII. Your messages are
internally decoded to unicode objects and passed to output. Then the
displayed output will be limited only by the encoding of your terminal,
but it will not crash your program on any inconsistency, you would see
question marks.
--
Jarek Zgoda
"We read Knuth so you don't have to."
Thorsten Kampe
Actually, this is the solution.
But there's one more problem: the solution only works when the
Terminal encoding is not US-ASCII. Unfortunately (almost) all
terminals I tried are set to US-ASCII (rxvt under Cygwin, Console[1]
running bash, Poderosa[2] running bash). Only the Windows Console is
CP852 and this works.
I got the tip to set a different encoding by
sys.stdout = codecs.EncodedFile(sys.stdout, 'utf-8')
but unfortunately this does not change the encoding of any Terminal.
So my question is: how can I set a different encoding to sys.stdout
(or why can I set it without any error but nothing changes?)
Thorsten
[1] http://sourceforge.net/project/screenshots.php?group_id=43764
[2] http://en.poderosa.org/present/about_poderosa.html
Thorsten Kampe
> Thorsten Kampe napisa?(a):
>
> > I can't see the "solution" here. Is the optparse "print_help" function
> > wrong? Why should there even be errors if I use "unicode = True" with
> > gettext.install?
> >
> > I have ISO-8859-15 gettext translations and I want optparse to display
> > them correctly. What do I have to do?
>
> Please, see gettext module documentation on this topic.
>
> The solution is: always install your translation with unicode=True
> setting. This assures usage of ugettext() instead of gettext() and works
> properly with character sets other than ASCII. Your messages are
> internally decoded to unicode objects and passed to output. Then the
> displayed output will be limited only by the encoding of your terminal,
You are right. My problem is that all the terminals I use are set to
US-ASCII (rxvt under Cygwin, Console[1] running bash, Poderosa[2]
running bash). Even those who actually support non-ASCII characters.
I got the tip to set a different encoding by
sys.stdout = codecs.EncodedFile(sys.stdout, 'utf-8')
but unfortunately this does not change the encoding.
paul
[snipp]
> I got the tip to set a different encoding by
> sys.stdout = codecs.EncodedFile(sys.stdout, 'utf-8')
>
> but unfortunately this does not change the encoding of any Terminal.
> So my question is: how can I set a different encoding to sys.stdout
> (or why can I set it without any error but nothing changes?)
display anything else; it might not even have the right font, so how are
you supposed to fix that? The .encode(encoding, "replace") ensures safe
downgrades though.
cheers
Paul
Thorsten Kampe
> Thorsten Kampe schrieb:
> [snipp]
> > I got the tip to set a different encoding by
> > sys.stdout = codecs.EncodedFile(sys.stdout, 'utf-8')
> >
> > but unfortunately this does not change the encoding of any Terminal.
> > So my question is: how can I set a different encoding to sys.stdout
> > (or why can I set it without any error but nothing changes?)
> AFAIK you can't. If the terminal is limited to ascii it won't be able to
> display anything else; it might not even have the right font, so how are
> you supposed to fix that?
Actually rxvt, Poderosa and console have the ability to display non-
ASCII characters. I use the dejavu fonts that support non-ASCII, too.
But the problem is even simpler: I can't even set the standard Windows
console ("cmd") to Windows 1252 in Python. Although directly executing
"chcp 1252" works.
Thorsten
Damjan
> Actually rxvt, Poderosa and console have the ability to display non-
> ASCII characters. I use the dejavu fonts that support non-ASCII, too.
>
> But the problem is even simpler: I can't even set the standard Windows
> console ("cmd") to Windows 1252 in Python. Although directly executing
> "chcp 1252" works.
Maybe try to use http://sourceforge.net/projects/console it's claimed to be
muc better than the sucky CDM (I don't have windows to try it).
--
damjan
Thorsten Kampe
It is definitely. But it just runs bash or cmd.exe so its capabilities
(encoding) are defined by Windows or Cygwin.
Thorsten
Thorsten Kampe
> * Steven Bethard (Sun, 01 Apr 2007 10:21:40 -0600)
> > Thorsten Kampe wrote:
> > I'm not very experienced with internationalization, but if you change::
> >
> > gettext.install('test')
> >
> > to::
> >
> > gettext.install('test', unicode=True)
> >
> > what happens?
>
> Actually, this is the solution.
>
> But there's one more problem: the solution only works when the
> Terminal encoding is not US-ASCII. Unfortunately (almost) all
> terminals I tried are set to US-ASCII (rxvt under Cygwin, Console[1]
> running bash, Poderosa[2] running bash). Only the Windows Console is
> CP852 and this works.
>
> I got the tip to set a different encoding by
> sys.stdout = codecs.EncodedFile(sys.stdout, 'utf-8')
>
> but unfortunately this does not change the encoding of any Terminal.
> So my question is: how can I set a different encoding to sys.stdout
> (or why can I set it without any error but nothing changes?)
I solved it (finally after two days with the help of a lot of people):
You have to set this...
sys.stdout = codecs.EncodedFile(sys.stdout, 'iso-8859-15')
sys.stdout.encoding = 'iso-8859-15'
...both of these and exactly in this order and not vice versa. It
doesn't have to be 'iso-8859-15', Windows-1252 is fine, too (but UTF-8
doesn't work). Now we have a new problem: the native Windows consoles
don't print the right characters. So you wrap this in a query:
if sys.platform in ['cygwin', 'linux2']:
sys.stdout = codecs.EncodedFile(sys.stdout, 'iso-8859-15')
sys.stdout.encoding = 'iso-8859-15'
This would be a problem if there's more than one translation (for
instance one with polish characters that aren't contained in iso-8859-
15). One could work around this with
if sys.platform in ['cygwin', 'linux2']:
sys.stdout = codecs.EncodedFile(sys.stdout,
locale.getpreferredencoding())
sys.stdout.encoding = locale.getpreferredencoding()
Funny (more or less): two days work to print "U" and "A" with double
points above.
Thanks to all who have helped to clear my confusion and to bring me
into the right direction for the solution.
Thorsten