158
votes

I have many "can't encode" and "can't decode" problems with Python when I run my applications from the console. But in the Eclipse PyDev IDE, the default character encoding is set to UTF-8, and I'm fine.

I searched around for setting the default encoding, and people say that Python deletes the sys.setdefaultencoding function on startup, and we can not use it.

So what's the best solution for it?

12
See the blog post The Illusive setdefaultencoding.djc
The best solution is to learn to use encode and decode correctly instead of using hacks. This was certainly possible with python2 at the cost of always remembering to do so / consistently using your own interface. My experience suggests that this becomes highly problematic when you are writing code that you want to work with both python2 and python3.Att Righ

12 Answers

172
votes

Here is a simpler method (hack) that gives you back the setdefaultencoding() function that was deleted from sys:

import sys
# sys.setdefaultencoding() does not exist, here!
reload(sys)  # Reload does the trick!
sys.setdefaultencoding('UTF8')

(Note for Python 3.4+: reload() is in the importlib library.)

This is not a safe thing to do, though: this is obviously a hack, since sys.setdefaultencoding() is purposely removed from sys when Python starts. Reenabling it and changing the default encoding can break code that relies on ASCII being the default (this code can be third-party, which would generally make fixing it impossible or dangerous).

82
votes

If you get this error when you try to pipe/redirect output of your script

UnicodeEncodeError: 'ascii' codec can't encode characters in position 0-5: ordinal not in range(128)

Just export PYTHONIOENCODING in console and then run your code.

export PYTHONIOENCODING=utf8

53
votes

A) To control sys.getdefaultencoding() output:

python -c 'import sys; print(sys.getdefaultencoding())'

ascii

Then

echo "import sys; sys.setdefaultencoding('utf-16-be')" > sitecustomize.py

and

PYTHONPATH=".:$PYTHONPATH" python -c 'import sys; print(sys.getdefaultencoding())'

utf-16-be

You could put your sitecustomize.py higher in your PYTHONPATH.

Also you might like to try reload(sys).setdefaultencoding by @EOL

B) To control stdin.encoding and stdout.encoding you want to set PYTHONIOENCODING:

python -c 'import sys; print(sys.stdin.encoding, sys.stdout.encoding)'

ascii ascii

Then

PYTHONIOENCODING="utf-16-be" python -c 'import sys; 
print(sys.stdin.encoding, sys.stdout.encoding)'

utf-16-be utf-16-be

Finally: you can use A) or B) or both!

18
votes

Starting with PyDev 3.4.1, the default encoding is not being changed anymore. See this ticket for details.

For earlier versions a solution is to make sure PyDev does not run with UTF-8 as the default encoding. Under Eclipse, run dialog settings ("run configurations", if I remember correctly); you can choose the default encoding on the common tab. Change it to US-ASCII if you want to have these errors 'early' (in other words: in your PyDev environment). Also see an original blog post for this workaround.

13
votes

Regarding python2 (and python2 only), some of the former answers rely on using the following hack:

import sys
reload(sys)  # Reload is a hack
sys.setdefaultencoding('UTF8')

It is discouraged to use it (check this or this)

In my case, it come with a side-effect: I'm using ipython notebooks, and once I run the code the ´print´ function no longer works. I guess there would be solution to it, but still I think using the hack should not be the correct option.

After trying many options, the one that worked for me was using the same code in the sitecustomize.py, where that piece of code is meant to be. After evaluating that module, the setdefaultencoding function is removed from sys.

So the solution is to append to file /usr/lib/python2.7/sitecustomize.py the code:

import sys
sys.setdefaultencoding('UTF8')

When I use virtualenvwrapper the file I edit is ~/.virtualenvs/venv-name/lib/python2.7/sitecustomize.py.

And when I use with python notebooks and conda, it is ~/anaconda2/lib/python2.7/sitecustomize.py

8
votes

There is an insightful blog post about it.

See https://anonbadger.wordpress.com/2015/06/16/why-sys-setdefaultencoding-will-break-code/.

I paraphrase its content below.

In python 2 which was not as strongly typed regarding the encoding of strings you could perform operations on differently encoded strings, and succeed. E.g. the following would return True.

u'Toshio' == 'Toshio'

That would hold for every (normal, unprefixed) string that was encoded in sys.getdefaultencoding(), which defaulted to ascii, but not others.

The default encoding was meant to be changed system-wide in site.py, but not somewhere else. The hacks (also presented here) to set it in user modules were just that: hacks, not the solution.

Python 3 did changed the system encoding to default to utf-8 (when LC_CTYPE is unicode-aware), but the fundamental problem was solved with the requirement to explicitly encode "byte"strings whenever they are used with unicode strings.

5
votes

Here is the approach I used to produce code that was compatible with both python2 and python3 and always produced utf8 output. I found this answer elsewhere, but I can't remember the source.

This approach works by replacing sys.stdout with something that isn't quite file-like (but still only using things in the standard library). This may well cause problems for your underlying libraries, but in the simple case where you have good control over how sys.stdout out is used through your framework this can be a reasonable approach.

sys.stdout = io.open(sys.stdout.fileno(), 'w', encoding='utf8')
4
votes

First: reload(sys) and setting some random default encoding just regarding the need of an output terminal stream is bad practice. reload often changes things in sys which have been put in place depending on the environment - e.g. sys.stdin/stdout streams, sys.excepthook, etc.

Solving the encode problem on stdout

The best solution I know for solving the encode problem of print'ing unicode strings and beyond-ascii str's (e.g. from literals) on sys.stdout is: to take care of a sys.stdout (file-like object) which is capable and optionally tolerant regarding the needs:

  • When sys.stdout.encoding is None for some reason, or non-existing, or erroneously false or "less" than what the stdout terminal or stream really is capable of, then try to provide a correct .encoding attribute. At last by replacing sys.stdout & sys.stderr by a translating file-like object.

  • When the terminal / stream still cannot encode all occurring unicode chars, and when you don't want to break print's just because of that, you can introduce an encode-with-replace behavior in the translating file-like object.

Here an example:

#!/usr/bin/env python
# encoding: utf-8
import sys

class SmartStdout:
    def __init__(self, encoding=None, org_stdout=None):
        if org_stdout is None:
            org_stdout = getattr(sys.stdout, 'org_stdout', sys.stdout)
        self.org_stdout = org_stdout
        self.encoding = encoding or \
                        getattr(org_stdout, 'encoding', None) or 'utf-8'
    def write(self, s):
        self.org_stdout.write(s.encode(self.encoding, 'backslashreplace'))
    def __getattr__(self, name):
        return getattr(self.org_stdout, name)

if __name__ == '__main__':
    if sys.stdout.isatty():
        sys.stdout = sys.stderr = SmartStdout()

    us = u'aouäöüфżß²'
    print us
    sys.stdout.flush()

Using beyond-ascii plain string literals in Python 2 / 2 + 3 code

The only good reason to change the global default encoding (to UTF-8 only) I think is regarding an application source code decision - and not because of I/O stream encodings issues: For writing beyond-ascii string literals into code without being forced to always use u'string' style unicode escaping. This can be done rather consistently (despite what anonbadger's article says) by taking care of a Python 2 or Python 2 + 3 source code basis which uses ascii or UTF-8 plain string literals consistently - as far as those strings potentially undergo silent unicode conversion and move between modules or potentially go to stdout. For that, prefer "# encoding: utf-8" or ascii (no declaration). Change or drop libraries which still rely in a very dumb way fatally on ascii default encoding errors beyond chr #127 (which is rare today).

And do like this at application start (and/or via sitecustomize.py) in addition to the SmartStdout scheme above - without using reload(sys):

...
def set_defaultencoding_globally(encoding='utf-8'):
    assert sys.getdefaultencoding() in ('ascii', 'mbcs', encoding)
    import imp
    _sys_org = imp.load_dynamic('_sys_org', 'sys')
    _sys_org.setdefaultencoding(encoding)

if __name__ == '__main__':
    sys.stdout = sys.stderr = SmartStdout()
    set_defaultencoding_globally('utf-8') 
    s = 'aouäöüфżß²'
    print s

This way string literals and most operations (except character iteration) work comfortable without thinking about unicode conversion as if there would be Python3 only. File I/O of course always need special care regarding encodings - as it is in Python3.

Note: plains strings then are implicitely converted from utf-8 to unicode in SmartStdout before being converted to the output stream enconding.

2
votes

This is a quick hack for anyone who is (1) On a Windows platform (2) running Python 2.7 and (3) annoyed because a nice piece of software (i.e., not written by you so not immediately a candidate for encode/decode printing maneuvers) won't display the "pretty unicode characters" in the IDLE environment (Pythonwin prints unicode fine), For example, the neat First Order Logic symbols that Stephan Boyer uses in the output from his pedagogic prover at First Order Logic Prover.

I didn't like the idea of forcing a sys reload and I couldn't get the system to cooperate with setting environment variables like PYTHONIOENCODING (tried direct Windows environment variable and also dropping that in a sitecustomize.py in site-packages as a one liner ='utf-8').

So, if you are willing to hack your way to success, go to your IDLE directory, typically: "C:\Python27\Lib\idlelib" Locate the file IOBinding.py. Make a copy of that file and store it somewhere else so you can revert to original behavior when you choose. Open the file in the idlelib with an editor (e.g., IDLE). Go to this code area:

# Encoding for file names
filesystemencoding = sys.getfilesystemencoding()

encoding = "ascii"
if sys.platform == 'win32':
    # On Windows, we could use "mbcs". However, to give the user
    # a portable encoding name, we need to find the code page 
    try:
        # --> 6/5/17 hack to force IDLE to display utf-8 rather than cp1252
        # --> encoding = locale.getdefaultlocale()[1]
        encoding = 'utf-8'
        codecs.lookup(encoding)
    except LookupError:
        pass

In other words, comment out the original code line following the 'try' that was making the encoding variable equal to locale.getdefaultlocale (because that will give you cp1252 which you don't want) and instead brute force it to 'utf-8' (by adding the line 'encoding = 'utf-8' as shown).

I believe this only affects IDLE display to stdout and not the encoding used for file names etc. (that is obtained in the filesystemencoding prior). If you have a problem with any other code you run in IDLE later, just replace the IOBinding.py file with the original unmodified file.

1
votes

This fixed the issue for me.

import os
os.environ["PYTHONIOENCODING"] = "utf-8"
1
votes

You could change the encoding of your entire operating system. On Ubuntu you can do this with

sudo apt install locales 
sudo locale-gen en_US en_US.UTF-8    
sudo dpkg-reconfigure locales
0
votes

set default encoding of OS to be UTF-8. Eg., on ubuntu edit file /etc/default/locale and set:

LANG=en_US.UTF-8
LANGUAGE=en_US.UTF-8
LC_ALL=en_US.UTF-8