I’m using python’s ftplib to write a small FTP client, but some of the functions in the package don’t return string output, but print to stdout. I want to redirect stdout to an object which I’ll be able to read the output from.

I know stdout can be redirected into any regular file with:

stdout = open("file", "a")

But I prefer a method that doesn’t uses the local drive.

I’m looking for something like the BufferedReader in Java that can be used to wrap a buffer into a stream.

from cStringIO import StringIO # Python3 use: from io import StringIO
import sys

old_stdout = sys.stdout
sys.stdout = mystdout = StringIO()

# blah blah lots of code ...

sys.stdout = old_stdout

# examine mystdout.getvalue()

There is contextlib.redirect_stdout() function in Python 3.4:

import io
from contextlib import redirect_stdout

with io.StringIO() as buf, redirect_stdout(buf):
    print('redirected')
    output = buf.getvalue()

Here’s code example that shows how to implement it on older Python versions.

Just to add to Ned’s answer above: you can use this to redirect output to any object that implements a write(str) method.

This can be used to good effect to “catch” stdout output in a GUI application.

Here’s a silly example in PyQt:

import sys
from PyQt4 import QtGui

class OutputWindow(QtGui.QPlainTextEdit):
    def write(self, txt):
        self.appendPlainText(str(txt))

app = QtGui.QApplication(sys.argv)
out = OutputWindow()
sys.stdout=out
out.show()
print "hello world !"

Starting with Python 2.6 you can use anything implementing the TextIOBase API from the io module as a replacement. This solution also enables you to use sys.stdout.buffer.write() in Python 3 to write (already) encoded byte strings to stdout (see stdout in Python 3). Using StringIO wouldn’t work then, because neither sys.stdout.encoding nor sys.stdout.buffer would be available.

A solution using TextIOWrapper:

import sys
from io import TextIOWrapper, BytesIO

# setup the environment
old_stdout = sys.stdout
sys.stdout = TextIOWrapper(BytesIO(), sys.stdout.encoding)

# do something that writes to stdout or stdout.buffer

# get output
sys.stdout.seek(0)      # jump to the start
out = sys.stdout.read() # read output

# restore stdout
sys.stdout.close()
sys.stdout = old_stdout

This solution works for Python 2 >= 2.6 and Python 3.

Please note that our new sys.stdout.write() only accepts unicode strings and sys.stdout.buffer.write() only accepts byte strings. This might not be the case for old code, but is often the case for code that is built to run on Python 2 and 3 without changes, which again often makes use of sys.stdout.buffer.

You can build a slight variation that accepts unicode and byte strings for write():

class StdoutBuffer(TextIOWrapper):
    def write(self, string):
        try:
            return super(StdoutBuffer, self).write(string)
        except TypeError:
            # redirect encoded byte strings directly to buffer
            return super(StdoutBuffer, self).buffer.write(string)

You don’t have to set the encoding of the buffer the sys.stdout.encoding, but this helps when using this method for testing/comparing script output.

This method restores sys.stdout even if there’s an exception. It also gets any output before the exception.

import io
import sys

real_stdout = sys.stdout
fake_stdout = io.BytesIO()   # or perhaps io.StringIO()
try:
    sys.stdout = fake_stdout
    # do what you have to do to create some output
finally:
    sys.stdout = real_stdout
    output_string = fake_stdout.getvalue()
    fake_stdout.close()
    # do what you want with the output_string

Tested in Python 2.7.10 using io.BytesIO()

Tested in Python 3.6.4 using io.StringIO()

Bob, added for a case if you feel anything from the modified / extended code experimentation might get interesting in any sense, otherwise feel free to delete it

^{Ad informandum … a few remarks from extended experimentation during finding some viable mechanics to “grab” outputs, directed by numexpr.print_versions() directly to the <stdout> ( upon a need to clean GUI and collecting details into debugging-report )}

# THIS WORKS AS HELL: as Bob Stein proposed years ago:
#  py2 SURPRISEDaBIT:
#
import io
import sys
#
real_stdout = sys.stdout                        #           PUSH <stdout> ( store to REAL_ )
fake_stdout = io.BytesIO()                      #           .DEF FAKE_
try:                                            # FUSED .TRY:
    sys.stdout.flush()                          #           .flush() before
    sys.stdout = fake_stdout                    #           .SET <stdout> to use FAKE_
    # ----------------------------------------- #           +    do what you gotta do to create some output
    print 123456789                             #           + 
    import  numexpr                             #           + 
    QuantFX.numexpr.__version__                 #           + [3] via fake_stdout re-assignment, as was bufferred + "late" deferred .get_value()-read into print, to finally reach -> real_stdout
    QuantFX.numexpr.print_versions()            #           + [4] via fake_stdout re-assignment, as was bufferred + "late" deferred .get_value()-read into print, to finally reach -> real_stdout
    _ = os.system( 'echo os.system() redir-ed' )#           + [1] via real_stdout                                 + "late" deferred .get_value()-read into print, to finally reach -> real_stdout, if not ( _ = )-caught from RET-d "byteswritten" / avoided from being injected int fake_stdout
    _ = os.write(  sys.stderr.fileno(),         #           + [2] via      stderr                                 + "late" deferred .get_value()-read into print, to finally reach -> real_stdout, if not ( _ = )-caught from RET-d "byteswritten" / avoided from being injected int fake_stdout
                       b'os.write()  redir-ed' )#  *OTHERWISE, if via fake_stdout, EXC <_io.BytesIO object at 0x02C0BB10> Traceback (most recent call last):
    # ----------------------------------------- #           ?                              io.UnsupportedOperation: fileno
    #'''                                                    ? YET:        <_io.BytesIO object at 0x02C0BB10> has a .fileno() method listed
    #>>> 'fileno' in dir( sys.stdout )       -> True        ? HAS IT ADVERTISED,
    #>>> pass;            sys.stdout.fileno  -> <built-in method fileno of _io.BytesIO object at 0x02C0BB10>
    #>>> pass;            sys.stdout.fileno()-> Traceback (most recent call last):
    #                                             File "<stdin>", line 1, in <module>
    #                                           io.UnsupportedOperation: fileno
    #                                                       ? BUT REFUSES TO USE IT
    #'''
finally:                                        # == FINALLY:
    sys.stdout.flush()                          #           .flush() before ret'd back REAL_
    sys.stdout = real_stdout                    #           .SET <stdout> to use POP'd REAL_
    sys.stdout.flush()                          #           .flush() after  ret'd back REAL_
    out_string = fake_stdout.getvalue()         #           .GET string           from FAKE_
    fake_stdout.close()                         #                <FD>.close()
    # +++++++++++++++++++++++++++++++++++++     # do what you want with the out_string
    #
    print "\n{0:}\n{1:}{0:}".format( 60 * "/\\",# "LATE" deferred print the out_string at the very end reached -> real_stdout
                                     out_string #                   
                                     )
'''
PASS'd:::::
...
os.system() redir-ed
os.write()  redir-ed
/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\
123456789
'2.5'
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Numexpr version:   2.5
NumPy version:     1.10.4
Python version:    2.7.13 |Anaconda 4.0.0 (32-bit)| (default, May 11 2017, 14:07:41) [MSC v.1500 32 bit (Intel)]
AMD/Intel CPU?     True
VML available?     True
VML/MKL version:   Intel(R) Math Kernel Library Version 11.3.1 Product Build 20151021 for 32-bit applications
Number of threads used by default: 4 (out of 4 detected cores)
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\
>>>

EXC'd :::::
...
os.system() redir-ed
/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\
123456789
'2.5'
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Numexpr version:   2.5
NumPy version:     1.10.4
Python version:    2.7.13 |Anaconda 4.0.0 (32-bit)| (default, May 11 2017, 14:07:41) [MSC v.1500 32 bit (Intel)]
AMD/Intel CPU?     True
VML available?     True
VML/MKL version:   Intel(R) Math Kernel Library Version 11.3.1 Product Build 20151021 for 32-bit applications
Number of threads used by default: 4 (out of 4 detected cores)
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\

Traceback (most recent call last):
  File "<stdin>", line 9, in <module>
io.UnsupportedOperation: fileno
'''

A context manager for python3:

import sys
from io import StringIO


class RedirectedStdout:
    def __init__(self):
        self._stdout = None
        self._string_io = None

    def __enter__(self):
        self._stdout = sys.stdout
        sys.stdout = self._string_io = StringIO()
        return self

    def __exit__(self, type, value, traceback):
        sys.stdout = self._stdout

    def __str__(self):
        return self._string_io.getvalue()

use like this:

>>> with RedirectedStdout() as out:
>>>     print('asdf')
>>>     s = str(out)
>>>     print('bsdf')
>>> print(s, out)
'asdf\n' 'asdf\nbsdf\n'

In Python3.6, the StringIO and cStringIO modules are gone, you should use io.StringIO instead.So you should do this like the first answer:

import sys
from io import StringIO

old_stdout = sys.stdout
old_stderr = sys.stderr
my_stdout = sys.stdout = StringIO()
my_stderr = sys.stderr = StringIO()

# blah blah lots of code ...

sys.stdout = self.old_stdout
sys.stderr = self.old_stderr

// if you want to see the value of redirect output, be sure the std output is turn back
print(my_stdout.getvalue())
print(my_stderr.getvalue())

my_stdout.close()
my_stderr.close()

Use pipe() and write to the appropriate file descriptor.

https://docs.python.org/library/os.html#file-descriptor-operations

Here’s another take on this. contextlib.redirect_stdout with io.StringIO() as documented is great, but it’s still a bit verbose for every day use. Here’s how to make it a one-liner by subclassing contextlib.redirect_stdout:

import sys
import io
from contextlib import redirect_stdout

class capture(redirect_stdout):

    def __init__(self):
        self.f = io.StringIO()
        self._new_target = self.f
        self._old_targets = []  # verbatim from parent class

    def __enter__(self):
        self._old_targets.append(getattr(sys, self._stream))  # verbatim from parent class
        setattr(sys, self._stream, self._new_target)  # verbatim from parent class
        return self  # instead of self._new_target in the parent class

    def __repr__(self):
        return self.f.getvalue()  

Since __enter__ returns self, you have the context manager object available after the with block exits. Moreover, thanks to the __repr__ method, the string representation of the context manager object is, in fact, stdout. So now you have,

with capture() as message:
    print('Hello World!')
print(str(message)=='Hello World!\n')  # returns True