How to set sys.stdout encoding in Python 3?

Python 3.1 added io.TextIOBase.detach(), with a note in the documentation for sys.stdout:

The standard streams are in text mode by default. To write or read binary data to these, use the underlying binary buffer. For example, to write bytes to stdout, use sys.stdout.buffer.write(b'abc'). Using io.TextIOBase.detach() streams can be made binary by default. This function sets stdin and stdout to binary:

def make_streams_binary():
    sys.stdin = sys.stdin.detach()
    sys.stdout = sys.stdout.detach()

Therefore, the corresponding idiom for Python 3.1 and later is:

sys.stdout = codecs.getwriter("utf-8")(sys.stdout.detach())

sys.stdout is in text mode in Python 3. Hence you write unicode to it directly, and the idiom for Python 2 is no longer needed.

Where this would fail in Python 2:

>>> import sys
>>> sys.stdout.write(u"ûnicöde")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
UnicodeEncodeError: 'ascii' codec can't encode character u'xfb' in position 0: ordinal not in range(128)

However, it works just dandy in Python 3:

>>> import sys
>>> sys.stdout.write("Ûnicöde")
Ûnicöde7

Now if your Python doesn’t know what your stdouts encoding actually is, that’s a different problem, most likely in the build of the Python.

Setting the default output encoding in Python 2 is a well-known idiom

Eek! Is that a well-known idiom in Python 2? It looks like a dangerous mistake to me.

It’ll certainly mess up any script that tries to write binary to stdout (which you’ll need if you’re a CGI script returning an image, for example). Bytes and chars are quite different animals; it’s not a good idea to monkey-patch an interface that is specified to accept bytes with one that only takes chars.

CGI and HTTP in general explicitly work with bytes. You should only be sending bytes to sys.stdout. In Python 3 that means using sys.stdout.buffer.write to send bytes directly. Encoding page content to match its charset parameter should be handled at a higher level in your application (in cases where you are returning textual content, rather than binary). This also means print is no good for CGI any more.

(To add to the confusion, wsgiref’s CGIHandler has been broken in py3k until very recently, making it impossible to deploy WSGI to CGI that way. With PEP 3333 and Python 3.2 this is finally workable.)

I found this thread while searching for solutions to the same error,

An alternative solution to those already suggested is to set the PYTHONIOENCODING environment variable before Python starts, for my use – this is less trouble then swapping sys.stdout after Python is initialized:

PYTHONIOENCODING=utf-8:surrogateescape python3 somescript.py

With the advantage of not having to go and edit the Python code.

Using detach() causes the interpreter to print a warning when it tries to close stdout just before it exits:

Exception ignored in: <_io.TextIOWrapper mode='w' encoding='UTF-8'>
ValueError: underlying buffer has been detached

Instead, this worked fine for me:

default_out = io.TextIOWrapper(sys.stdout.buffer, encoding='utf-8')

(And, of course, writing to default_out instead of stdout.)

Other answers seem to recommend using codecs, but open works for me:

import sys
sys.stdout = open(sys.stdout.fileno(), mode='w', encoding='utf8', buffering=1)
print("日本語")
# Also works with other methods of writing to stdout:
sys.stdout.write("日本語n")
sys.stdout.buffer.write("日本語n".encode())

This works even when I run it with PYTHONIOENCODING="ascii".