How to print to stderr in Python?

sys.stderr.write() is my choice, just more readable and saying exactly what you intend to do and portable across versions.

Edit: being 'pythonic' is a third thought to me over readability and performance... with these two things in mind, with python 80% of your code will be pythonic. list comprehension being the 'big thing' that isn't used as often (readability).

For Python 2 my choice is: print >> sys.stderr, 'spam' Because you can simply print lists/dicts etc. without convert it to string. print >> sys.stderr, {'spam': 'spam'} instead of: sys.stderr.write(str('spam': 'spam'))

print >> sys.stderr is gone in Python3. http://docs.python.org/3.0/whatsnew/3.0.html says:

Old: print >>sys.stderr, "fatal error"
New: print("fatal error", file=sys.stderr)

Unfortunately, this is quite ugly. Alternatively, use

sys.stderr.write("fatal error\n")

but note that write is not a 1:1 replacement for print.

I would say that your first approach:

print >> sys.stderr, 'spam' 

is the "One . . . obvious way to do it" The others don't satisfy rule #1 ("Beautiful is better than ugly.")

I did the following using Python 3:

from sys import stderr

def print_err(*args, **kwargs):
    print(*args, file=stderr, **kwargs)

So now I'm able to add keyword arguments, for example, to avoid carriage return:

print_err("Error: end of the file reached. The word ", end='')
print_err(word, "was not found")

This will mimic the standard print function but output on stderr

def print_err(*args):
    sys.stderr.write(' '.join(map(str,args)) + '\n')

EDIT In hind-sight, I think the potential confusion with changing sys.stderr and not seeing the behaviour updated makes this answer not as good as just using a simple function as others have pointed out.

Using partial only saves you 1 line of code. The potential confusion is not worth saving 1 line of code.

original

To make it even easier, here's a version that uses 'partial', which is a big help in wrapping functions.

from __future__ import print_function
import sys
from functools import partial

error = partial(print, file=sys.stderr)

You then use it like so

error('An error occured!')

You can check that it's printing to stderr and not stdout by doing the following (over-riding code from http://coreygoldberg.blogspot.com.au/2009/05/python-redirect-or-turn-off-stdout-and.html):

# over-ride stderr to prove that this function works.
class NullDevice():
    def write(self, s):
        pass
sys.stderr = NullDevice()

# we must import print error AFTER we've removed the null device because
# it has been assigned and will not be re-evaluated.
# assume error function is in print_error.py
from print_error import error

# no message should be printed
error("You won't see this error!")

The downside to this is partial assigns the value of sys.stderr to the wrapped function at the time of creation. Which means, if you redirect stderr later it won't affect this function. If you plan to redirect stderr, then use the **kwargs method mentioned by aaguirre on this page.

Nobody's mentioned logging yet, but logging was created specifically to communicate error messages. By default it is set up to write to stderr. This script:

# foo.py
import logging
logging.basicConfig(format='%(message)s')

logging.warn('I print to stderr by default')
logging.info('For this you must change the level and add a handler.')
print('hello world')

has the following result when run on the command line:

$ python3 foo.py > bar.txt
I print to stderr by default

(and bar.txt contains the 'hello world')

The same applies to stdout:

print 'spam'
sys.stdout.write('spam\n')

As stated in the other answers, print offers a pretty interface that is often more convenient (e.g. for printing debug information), while write is faster and can also be more convenient when you have to format the output exactly in certain way. I would consider maintainability as well:

1. You may later decide to switch between stdout/stderr and a regular file.

2. print() syntax has changed in Python 3, so if you need to support both versions, write() might be better.

I am working in python 3.4.3. I am cutting out a little typing that shows how I got here:

[18:19 jsilverman@JSILVERMAN-LT7 pexpect]$ python3
>>> import sys
>>> print("testing", file=sys.stderr)
testing
>>>
[18:19 jsilverman@JSILVERMAN-LT7 pexpect]$ 

Did it work? Try redirecting stderr to a file and see what happens:

[18:22 jsilverman@JSILVERMAN-LT7 pexpect]$ python3 2> /tmp/test.txt
>>> import sys
>>> print("testing", file=sys.stderr)
>>> [18:22 jsilverman@JSILVERMAN-LT7 pexpect]$
[18:22 jsilverman@JSILVERMAN-LT7 pexpect]$ cat /tmp/test.txt
Python 3.4.3 (default, May  5 2015, 17:58:45)
[GCC 4.9.2] on cygwin
Type "help", "copyright", "credits" or "license" for more information.
testing

[18:22 jsilverman@JSILVERMAN-LT7 pexpect]$

Well, aside from the fact that the little introduction that python gives you has been slurped into stderr (where else would it go?), it works.

If you do a simple test:

import time
import sys

def run1(runs):
    x = 0
    cur = time.time()
    while x < runs:
        x += 1
        print >> sys.stderr, 'X'
    elapsed = (time.time()-cur)
    return elapsed

def run2(runs):
    x = 0
    cur = time.time()
    while x < runs:
        x += 1
        sys.stderr.write('X\n')
        sys.stderr.flush()
    elapsed = (time.time()-cur)
    return elapsed

def compare(runs):
    sum1, sum2 = 0, 0
    x = 0
    while x < runs:
        x += 1
        sum1 += run1(runs)
        sum2 += run2(runs)
    return sum1, sum2

if __name__ == '__main__':
    s1, s2 = compare(1000)
    print "Using (print >> sys.stderr, 'X'): %s" %(s1)
    print "Using (sys.stderr.write('X'),sys.stderr.flush()):%s" %(s2)
    print "Ratio: %f" %(float(s1) / float(s2))

You will find that sys.stderr.write() is consistently 1.81 times faster!