How to duplicate sys.stdout to a log file?

Question

Edit: Since it appears that there's either no solution, or I'm doing something so non-standard that nobody knows - I'll revise my question to also ask: What is the best way to accomplish logging when a python app is making a lot of system calls?

My app has two modes. In interactive mode, I want all output to go to the screen as well as to a log file, including output from any system calls. In daemon mode, all output goes to the log. Daemon mode works great using os.dup2(). I can't find a way to "tee" all output to a log in interactive mode, without modifying each and every system call.

---

In other words, I want the functionality of the command line 'tee' for any output generated by a python app, including system call output.

To clarify:

To redirect all output I do something like this, and it works great:

# open our log file
so = se = open("%s.log" % self.name, 'w', 0)

# re-open stdout without buffering
sys.stdout = os.fdopen(sys.stdout.fileno(), 'w', 0)

# redirect stdout and stderr to the log file opened above
os.dup2(so.fileno(), sys.stdout.fileno())
os.dup2(se.fileno(), sys.stderr.fileno())

The nice thing about this is that it requires no special print calls from the rest of the code. The code also runs some shell commands, so it's nice not having to deal with each of their output individually as well.

Simply, I want to do the same, except duplicating instead of redirecting.

At first thought, I thought that simply reversing the dup2's should work. Why doesn't it? Here's my test:

import os, sys

### my broken solution:
so = se = open("a.log", 'w', 0)
sys.stdout = os.fdopen(sys.stdout.fileno(), 'w', 0)

os.dup2(sys.stdout.fileno(), so.fileno())
os.dup2(sys.stderr.fileno(), se.fileno())
###

print("foo bar")

os.spawnve("P_WAIT", "/bin/ls", ["/bin/ls"], {})
os.execve("/bin/ls", ["/bin/ls"], os.environ)

The file "a.log" should be identical to what was displayed on the screen.

Louvain

If you look at the man page (manpagez.com/man/2/dup2) the 2nd argument to dup2 is always closed (if it's open already). So in your "broken solution" it's closing so and se and then reassigning their filenos to sys.stdout. – Bartholemy

Re: your edit: this isn't uncommon, I've done similar a few times (in other langs). While Unix will allow multiple "aliases" for the same file handle, it won't "split" a file handle (copy it to multiple others). So you have to implement "tee" yourself (or just use "tee", see my crude answer). – Bartholemy

I think JohnT answer is better than the actual accepted one. You may want to change the accepted answer. – Goddord

"I'm doing something so non-standard " - you really are, people just send their logs to stderr and deal with from the command line. – Singspiel

Answer 1

Since you're comfortable spawning external processes from your code, you could use tee itself. I don't know of any Unix system calls that do exactly what tee does.

# Note this version was written circa Python 2.6, see below for
# an updated 3.3+-compatible version.
import subprocess, os, sys

# Unbuffer output (this ensures the output is in the correct order)
sys.stdout = os.fdopen(sys.stdout.fileno(), 'w', 0)

tee = subprocess.Popen(["tee", "log.txt"], stdin=subprocess.PIPE)
os.dup2(tee.stdin.fileno(), sys.stdout.fileno())
os.dup2(tee.stdin.fileno(), sys.stderr.fileno())

print "\nstdout"
print >>sys.stderr, "stderr"
os.spawnve("P_WAIT", "/bin/ls", ["/bin/ls"], {})
os.execve("/bin/ls", ["/bin/ls"], os.environ)

You could also emulate tee using the multiprocessing package (or use processing if you're using Python 2.5 or earlier).

Update

Here is a Python 3.3+-compatible version:

import subprocess, os, sys

tee = subprocess.Popen(["tee", "log.txt"], stdin=subprocess.PIPE)
# Cause tee's stdin to get a copy of our stdin/stdout (as well as that
# of any child processes we spawn)
os.dup2(tee.stdin.fileno(), sys.stdout.fileno())
os.dup2(tee.stdin.fileno(), sys.stderr.fileno())

# The flush flag is needed to guarantee these lines are written before
# the two spawned /bin/ls processes emit any output
print("\nstdout", flush=True)
print("stderr", file=sys.stderr, flush=True)

# These child processes' stdin/stdout are 
os.spawnve("P_WAIT", "/bin/ls", ["/bin/ls"], {})
os.execve("/bin/ls", ["/bin/ls"], os.environ)

Bartholemy

Well, this answer works, so I'll accept it. Still, it makes me feel dirty. – Louvain

I just posted a pure python implementation of tee (py2/3 compatible) that can run on any platform and also be used in different logging configurations. #617145 – Cerise

If Python runs on one of my machines and solution doesn't, then that's not a pythonic solution. Downvoted because of that. – Maximo

What is the seconde line (sys.stdout = os.fdopen(sys.stdout.fileno(), 'w', 0)) doing? Unbuffering sys.stdout? – Lizettelizotte

There is an issue with running this code several times – Satan

How to restore stdout and stderr before the end of the program ? I tried : tee.kill(); sys.stdout = sys.__stdout__; log file is correctly closed, but print commands that follow are lost... – Aphasic

(answer to myself just above) It seems that the 2 os.dup2 commands irreversibly close sys.stdout and sys.stderr (and sys.__stdout__, sys.__stderr__) till the end of the program (unless to re-open them in a hard way, I don't know how...). – Aphasic

According to this post the line sys.stdout = os.fdopen(sys.stdout.fileno(), 'w', 0) no longer works since python 3.3 (see PEP 3116) – Deathlike

Got the error "sys:1: ResourceWarning: unclosed file <_io.BufferedWriter name=5>", so I had to add tee.stdin.close() at the end of my program. I also get "ResourceWarning: subprocess 1842 is still running", and adding sys.stdout.close(); sys.stderr.close() at the end of the program fixes it. – Gains

Can you explain how the unbuffering works? dup2 will silently close the fd before reusing it. So why does it matter to change the buffering? Does dup2 preserve the buffering across the exchange? – Stormi

Good one, I would add: (... , preexec_fn=lambda: signal.signal(signal.SIGINT, signal.SIG_IGN) Once the user hits "Ctrl+C", the "tee" process is being killed first, closing the stdout\stderr FDs, then no printouts in the main thread. This solves the problem :) – Pembrook

Quite hard to understand for a newbie :( strungling to make this work. at the moment getting error: FileNotfoundError – Scenography

Answer 2

I had this same issue before and found this snippet very useful:

class Tee(object):
    def __init__(self, name, mode):
        self.file = open(name, mode)
        self.stdout = sys.stdout
        sys.stdout = self
    def __del__(self):
        sys.stdout = self.stdout
        self.file.close()
    def write(self, data):
        self.file.write(data)
        self.stdout.write(data)
    def flush(self):
        self.file.flush()

from: http://mail.python.org/pipermail/python-list/2007-May/438106.html

Showiness

+1 for handling the sys.stdout reassignment internally so that you can end logging by deleting the Tee object – Lutist

I'd add a flush to that. E.g: 'self.file.flush()' – Clark

I disagree about logging module. Excellent for some fiddling. Logging is too big for that. – Faller

Be sure to note the revised version in this follow-up to the linked discussion in the answer. – Crowfoot

Something similar can be implemented as a contextmanager for added flexibility as illustrated in this answer. – Crowfoot

Nice idea, though i found it better to open files outside Tee class via with – Palter

That will NOT work. __del__ is not called until the end of execution. See #6105035 – Brushoff

@Brushoff I'm using it this way: log1 = Tee('a.log', 'w') ; print '123' ; log1.__del__() and log2 = Tee('b.log', 'w') ; print '456' ; log2.__del__(). (See here for the whole code). Is it the right way or it should be corrected? – Heard

@Crowfoot looks like the link to that 'revised version' is now dead. Maybe if you still have a copy of the revision, submit it as a separate answer? – Barramunda

@user5359531: Try this link instead. – Crowfoot

@Crowfoot unfortunately, that link is blocked at my workplace :( – Barramunda

@Barramunda OK, I've posted a copy for you. – Crowfoot

@Brushoff For python 3, call self.file.flush() at the end of the .write() method. See also #231251. – Tale

This solution will not log output generated by other calls than print() and sys.stdout.write(). Output generated by C code or Fortran wrapped in Python, will not be redirected to file, because they write directly to the stdout fd (1), and not the IOTextwrapper sys.stdout. @Jacob Gabrielson's method will work for that though, but look at the comments of his answer for edits of his proposal. – Gains

Note that IO.write() returns int therefore to keep the same API as stdout, the last line should return self.stdout.write(data) rather than return None – Mackie

note __del__ will be triggered by GC if not explicitly called. not advised to who not familiar with gc. contextManager is probably safer for simple usecases. – Yoohoo

Answer 3

The print statement will call the write() method of any object you assign to sys.stdout.

I would spin up a small class to write to two places at once...

import sys

class Logger(object):
    def __init__(self):
        self.terminal = sys.stdout
        self.log = open("log.dat", "a")

    def write(self, message):
        self.terminal.write(message)
        self.log.write(message)  

sys.stdout = Logger()

Now the print statement will both echo to the screen and append to your log file:

# prints "1 2" to <stdout> AND log.dat
print "%d %d" % (1,2)

This is obviously quick-and-dirty. Some notes:

• You probably ought to parametize the log filename.
• You should probably revert sys.stdout to <stdout> if you won't be logging for the duration of the program.
• You may want the ability to write to multiple log files at once, or handle different log levels, etc.

These are all straightforward enough that I'm comfortable leaving them as exercises for the reader. The key insight here is that print just calls a "file-like object" that's assigned to sys.stdout.

Eveliaevelin

Exactly what I was going to post, pretty much. +1 when you fix the problem with write not having a self argument. Also, it'd be better design to have the file that you're going to write to passed in. Hell, it might also be better design to have stdout passed in. – Iatry

@Devin, yeah this was quick and dirty, I'll make some notes for possible early improvements. – Eveliaevelin

John T's snippet brings up a good point though. When the Logger gets deleted you should set stdout back to the default. – Habile

I selected this answer too soon. It works great for "print", but not so much for external command output. – Louvain

The Logger class should also define a flush() method such as "def flush(): self.terminal.flush(); self.log.flush()" – Bahrain

You say The print statement will call the write() method of any object you assign to sys.stdout. And what about other functions sending data to stdout not using print. For instance, if I create a process using subprocess.call its output goes to the console but not to log.dat file...is there a way to fix that? – Shote

@Bahrain That's what I needed :) – Cogswell

Excelent solution, but I got error - no flush method. How to call to parent 'flush' method? Shoud it be defined in constructor? – Shape

what about errors? Those will not be written to the logfile.. (for example: traceback.print_exc()) – Scab

Answer 4

What you really want is logging module from standard library. Create a logger and attach two handlers, one would be writing to a file and the other to stdout or stderr.

See Logging to multiple destinations for details

Keeping

Logging module doesn't record exceptions and other important output to stdout, which can be useful when analyzing logs on build server (for example). – Maximo

logging module won't redirect output from system calls such as os.write(1, b'stdout') – Puisne

Answer 5

Here is another solution, which is more general than the others -- it supports splitting output (written to sys.stdout) to any number of file-like objects. There's no requirement that __stdout__ itself is included.

import sys

class multifile(object):
    def __init__(self, files):
        self._files = files
    def __getattr__(self, attr, *args):
        return self._wrap(attr, *args)
    def _wrap(self, attr, *args):
        def g(*a, **kw):
            for f in self._files:
                res = getattr(f, attr, *args)(*a, **kw)
            return res
        return g

# for a tee-like behavior, use like this:
sys.stdout = multifile([ sys.stdout, open('myfile.txt', 'w') ])

# all these forms work:
print 'abc'
print >>sys.stdout, 'line2'
sys.stdout.write('line3\n')

NOTE: This is a proof-of-concept. The implementation here is not complete, as it only wraps methods of the file-like objects (e.g. write), leaving out members/properties/setattr, etc. However, it is probably good enough for most people as it currently stands.

What I like about it, other than its generality, is that it is clean in the sense it doesn't make any direct calls to write, flush, os.dup2, etc.

Viceregent

I would have init take *files not files, but otherwise, yes, this. None of the other solutions isolate the "tee" functionality without trying to solve other problems. If you want to put a prefix on everything you output, you can wrap this class in a prefix-writer class. (If you want to put a prefix on just one stream, you wrap a stream and hand it to this class.) This one also has the advantage that multifile([]) creates a file that ignores everything (like open('/dev/null')). – Corum

Why have _wrap here at all? Couldn't you copy the code in there into __getattr__ and it work the same? – Juggins

@Corum actually multifile([]) creates a file that raises an UnboundLocalError whenever you call one of its methods. (res is returned without being assigned) – Juggins

Since it is multifile, you can also add sys.stderr. It works on Windows as well. Only drawback : in an exception, print Traceback is lost in python 2.x and cannot be logged. – Aphasic

There is small question that sys.stdout is before sys.stderror EXP：sys.stdout = multifile([ sys.stdout, open('myfile.txt', 'w') ]) sys.stderr = multifile([ sys.stderr, open('myfile.txt', 'w') ]) sys.stderr.write( "Tests 1...\n" ) sys.stdout.write( "Tests 2...\n" )Tests 2 is before Tests 1 – Cosper

Answer 6

I know this question has been answered repeatedly, but for this I've taken the main answer from John T's answer and modified it so it contains the suggested flush and followed its linked revised version. I've also added the enter and exit as mentioned in cladmi's answer for use with the with statement. In addition, the documentation mentions to flush files using os.fsync() so I've added that as well. I don't know if you really need that but its there.

import sys, os

class Logger(object):
    "Lumberjack class - duplicates sys.stdout to a log file and it's okay"
    #source: https://mcmap.net/q/37785/-how-to-duplicate-sys-stdout-to-a-log-file
    def __init__(self, filename="Red.Wood", mode="a", buff=0):
        self.stdout = sys.stdout
        self.file = open(filename, mode, buff)
        sys.stdout = self

    def __del__(self):
        self.close()

    def __enter__(self):
        pass

    def __exit__(self, *args):
        self.close()

    def write(self, message):
        self.stdout.write(message)
        self.file.write(message)

    def flush(self):
        self.stdout.flush()
        self.file.flush()
        os.fsync(self.file.fileno())

    def close(self):
        if self.stdout != None:
            sys.stdout = self.stdout
            self.stdout = None

        if self.file != None:
            self.file.close()
            self.file = None

You can then use it

with Logger('My_best_girlie_by_my.side'):
    print("we'd sing sing sing")

or

Log=Logger('Sleeps_all.night')
print('works all day')
Log.close()

Talmud

Many Thnaks @Talmud you solved my question (#39143917). I will put a link to your solution. – Leyba

@MohammadElNesr I just realized an issue with the code when it's used with a with statement. I have fixed it and it now correctly closes at the end of a with block. – Talmud

This worked great for me, only needed to change mode to mode="ab" and in the write function self.file.write(message.encode("utf-8")) – Skylark

Answer 7

As described elsewhere, perhaps the best solution is to use the logging module directly:

import logging

logging.basicConfig(level=logging.DEBUG, filename='mylog.log')
logging.info('this should to write to the log file')

However, there are some (rare) occasions where you really want to redirect stdout. I had this situation when I was extending django's runserver command which uses print: I didn't want to hack the django source but needed the print statements to go to a file.

This is a way of redirecting stdout and stderr away from the shell using the logging module:

import logging, sys

class LogFile(object):
    """File-like object to log text using the `logging` module."""

    def __init__(self, name=None):
        self.logger = logging.getLogger(name)

    def write(self, msg, level=logging.INFO):
        self.logger.log(level, msg)

    def flush(self):
        for handler in self.logger.handlers:
            handler.flush()

logging.basicConfig(level=logging.DEBUG, filename='mylog.log')

# Redirect stdout and stderr
sys.stdout = LogFile('stdout')
sys.stderr = LogFile('stderr')

print 'this should to write to the log file'

You should only use this LogFile implementation if you really cannot use the logging module directly.

Bahrain

As described elsewhere, with such solution you won't see exceptions and other issues that are what's most important in such case. – Undress

Answer 8

To complete John T answer: https://mcmap.net/q/37785/-how-to-duplicate-sys-stdout-to-a-log-file

I added __enter__ and __exit__ methods to use it as a context manager with the with keyword, which gives this code

class Tee(object):
    def __init__(self, name, mode):
        self.file = open(name, mode)
        self.stdout = sys.stdout
        sys.stdout = self

    def __del__(self):
        sys.stdout = self.stdout
        self.file.close()

    def write(self, data):
        self.file.write(data)
        self.stdout.write(data)

    def __enter__(self):
        pass

    def __exit__(self, _type, _value, _traceback):
        pass

It can then be used as

with Tee('outfile.log', 'w'):
    print('I am written to both stdout and outfile.log')

Schulte

I would move the __del__ functionality into __exit__ – Interdict

Indeed, I think using __del__ is a bad idea. It should be moved to a 'close' function which is called in __exit__. – Schulte

Answer 9

I wrote a tee() implementation in Python that should work for most cases, and it works on Windows also.

https://github.com/pycontribs/tendo

Also, you can use it in combination with logging module from Python if you want.

Cerise

wow, your package rocks, especially if you know how cumbersome the Windows console culture is but didn't give up to make it work! – Melodiemelodion

Answer 10

(Ah, just re-read your question and see that this doesn't quite apply.)

Here is a sample program that makes uses the python logging module. This logging module has been in all versions since 2.3. In this sample the logging is configurable by command line options.

In quite mode it will only log to a file, in normal mode it will log to both a file and the console.

import os
import sys
import logging
from optparse import OptionParser

def initialize_logging(options):
    """ Log information based upon users options"""

    logger = logging.getLogger('project')
    formatter = logging.Formatter('%(asctime)s %(levelname)s\t%(message)s')
    level = logging.__dict__.get(options.loglevel.upper(),logging.DEBUG)
    logger.setLevel(level)

    # Output logging information to screen
    if not options.quiet:
        hdlr = logging.StreamHandler(sys.stderr)
        hdlr.setFormatter(formatter)
        logger.addHandler(hdlr)

    # Output logging information to file
    logfile = os.path.join(options.logdir, "project.log")
    if options.clean and os.path.isfile(logfile):
        os.remove(logfile)
    hdlr2 = logging.FileHandler(logfile)
    hdlr2.setFormatter(formatter)
    logger.addHandler(hdlr2)

    return logger

def main(argv=None):
    if argv is None:
        argv = sys.argv[1:]

    # Setup command line options
    parser = OptionParser("usage: %prog [options]")
    parser.add_option("-l", "--logdir", dest="logdir", default=".", help="log DIRECTORY (default ./)")
    parser.add_option("-v", "--loglevel", dest="loglevel", default="debug", help="logging level (debug, info, error)")
    parser.add_option("-q", "--quiet", action="store_true", dest="quiet", help="do not log to console")
    parser.add_option("-c", "--clean", dest="clean", action="store_true", default=False, help="remove old log file")

    # Process command line options
    (options, args) = parser.parse_args(argv)

    # Setup logger format and output locations
    logger = initialize_logging(options)

    # Examples
    logger.error("This is an error message.")
    logger.info("This is an info message.")
    logger.debug("This is a debug message.")

if __name__ == "__main__":
    sys.exit(main())

Drouin

Good answer. I saw some really convoluted ways of replicating logging to the console, but making a StreamHandler with stderr was the answer I've been looking for :) – Cerography

The code is nice put it does not answer the question - this outputs the log into a file and stderr, the original question was asking to duplicate the stderr into a log file. – Jordans

Answer 11

I've been using Jacob Gabrielson's accepted solution for about 1 year, but now the inevitable has happened and one of my users wants this on Windows. Looking at the other proposed answers, I think most of these answers fail at capturing the outputs of spawned processes (as bolded by the original poster); I think the only way to do this is to do os.dup2(). I think I've figured out how to answer the original poster's exact question, and without using the Unix-specific tool tee: I can now capture all outputs of my Python program, including any spawned shell commands. This works on Windows, Mac and Linux. The code is as follows:

import os, sys, threading, platform

class StreamCapture:
    def __init__(self,stream,writer,echo=True,monkeypatch=None):
        self.active = True
        self.writer = writer
        self.stream = stream
        self.fd = stream.fileno()
        self.echo = echo
        (r,w) = os.pipe()
        self.pipe_read_fd = r
        self.pipe_write_fd = w
        self.dup_fd = os.dup(self.fd)
        os.dup2(w,self.fd)
        self.monkeypatch = monkeypatch if monkeypatch is not None else platform.system()=='Windows'
        if self.monkeypatch:
            self.oldwrite = stream.write
            stream.write = lambda z: os.write(self.fd,z.encode() if type(z)==str else z)
        t = threading.Thread(target=self.printer)
        self.thread = t
        t.start()
    def printer(self):
        while True:
            data = os.read(self.pipe_read_fd,100000)
            if(len(data)==0):
                self.writer.close()
                os.close(self.dup_fd)
                os.close(self.pipe_read_fd)
                return
            self.writer.write(data)
            if self.echo:
                os.write(self.dup_fd,data)
    def close(self):
        if not self.active:
            return
        self.active = False
        self.stream.flush()
        if self.monkeypatch:
            self.stream.write = self.oldwrite
        os.dup2(self.dup_fd,self.fd)
        os.close(self.pipe_write_fd)
    def __enter__(self):
        return self
    def __exit__(self,a,b,c):
        self.close()

You use it like this (notice the hard case that other solutions, apart from Jacob Gabrielson's, fail to capture):

print("This does not get saved to the log file")
with StreamCapture(sys.stdout,open('logfile.txt','wb')):
        os.write(sys.stdout.fileno(),b"Hello, captured world!\n")
        os.system('echo Hello from the shell')     # Hard case
        print("More capturing")
print("This also does not get saved to the log file")

This is not a short-and-sweet answer, but I tried to keep it succinct, and it's as simple as I could make it. It's complicated for the following reasons:

1. Since I cannot use tee, I have to somehow perform the task of tee from within my Python process. It was not clear to me that there was a portable way of fork()ing and communicating with an os.pipe() (this states it's hard to share filedescriptors with forked processes in Windows) so I decided to use threading.

2. In Windows, sys.stdout and sys.stderr really don't appreciate when their underlying fileno() get rerouted through an os.pipe() via os.dup2(). The Python interpreter crashes immediately after the first print(...) command.

3. On Windows only, to solve the interpreter crashes, I monkeypatch sys.stdout.write = ... by setting it to a new function that simply calls to os.write(...). By default, I only do this when Windows is detected. Because I monkeypatch, I'm hoping this will reach all cached references to sys.stdout. I chose this monkeypatching approach instead of allocating a brand new stream, e.g. sys.stdout=..., because I was concerned that copies of the old sys.stdout would remain cached in various parts of the interpreter, but I guessed that sys.stdout.write was less likely to have been directly cached.

4. If you daemonize the thread that processes the output of the pipe, then that thread gets killed as soon as the main thread completes, but this does not guarantee that all outputs have been written to the log file. It's actually necessary to not daemonize those helper threads and to let them gracefully terminate themselves when the pipes are closed.

I'm actually not entirely sure that I got all the corner cases right -- threaded code that interacts with delicate OS features is scary to write. Nevertheless, it passed my tests thus far. Because it's kind of hairy, I've made a PyPI package:

pip install streamcapture

The Github is here.

Aurelioaurelius

I'm using it in unit tests to assert outputs of external non python libraries and its works perfectly, thank you. – Hypocrisy

You're welcome! – Sanfordsanfourd

Answer 12

another solution using logging module:

import logging
import sys

log = logging.getLogger('stdxxx')

class StreamLogger(object):

    def __init__(self, stream, prefix=''):
        self.stream = stream
        self.prefix = prefix
        self.data = ''

    def write(self, data):
        self.stream.write(data)
        self.stream.flush()

        self.data += data
        tmp = str(self.data)
        if '\x0a' in tmp or '\x0d' in tmp:
            tmp = tmp.rstrip('\x0a\x0d')
            log.info('%s%s' % (self.prefix, tmp))
            self.data = ''


logging.basicConfig(level=logging.INFO,
                    filename='text.log',
                    filemode='a')

sys.stdout = StreamLogger(sys.stdout, '[stdout] ')

print 'test for stdout'

Exfoliation

Answer 13

None of the answers above really seems to answer the problem posed. I know this is an old thread, but I think this problem is a lot simpler than everyone is making it:

class tee_err(object):

 def __init__(self):
    self.errout = sys.stderr

    sys.stderr = self

    self.log = 'logfile.log'
    log = open(self.log,'w')
    log.close()

 def write(self, line):

    log = open(self.log,'a')
    log.write(line)
    log.close()   

    self.errout.write(line)

Now this will repeat everything to the normal sys.stderr handler and your file. Create another class tee_out for sys.stdout.

Burkes

A similar, better answer was posted over two years before this one: stackoverflow.com/a/616686. Your method is very expensive: Each call to tee=tee_err();tee.write('');tee.write('');... opens+closes a file for each write. See https://mcmap.net/q/37812/-should-i-keep-a-file-open-or-should-i-open-and-close-often and https://mcmap.net/q/37813/-should-log-file-streams-be-opened-closed-on-each-write-or-kept-open-during-a-desktop-application-39-s-lifetime for arguments against this practice. – Gallant

Answer 14

As per a request by @user5359531 in the comments under @John T's answer, here's a copy of the referenced post to the revised version of the linked discussion in that answer:

Issue of redirecting the stdout to both file and screen
Gabriel Genellina gagsl-py2 at yahoo.com.ar
Mon May 28 12:45:51 CEST 2007

    Previous message: Issue of redirecting the stdout to both file and screen
    Next message: Formal interfaces with Python
    Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

En Mon, 28 May 2007 06:17:39 -0300, 人言落日是天涯，望极天涯不见家
<kelvin.you at gmail.com> escribió:

> I wanna print the log to both the screen and file, so I simulatered a
> 'tee'
>
> class Tee(file):
>
>     def __init__(self, name, mode):
>         file.__init__(self, name, mode)
>         self.stdout = sys.stdout
>         sys.stdout = self
>
>     def __del__(self):
>         sys.stdout = self.stdout
>         self.close()
>
>     def write(self, data):
>         file.write(self, data)
>         self.stdout.write(data)
>
> Tee('logfile', 'w')
> print >>sys.stdout, 'abcdefg'
>
> I found that it only output to the file, nothing to screen. Why?
> It seems the 'write' function was not called when I *print* something.

You create a Tee instance and it is immediately garbage collected. I'd
restore sys.stdout on Tee.close, not __del__ (you forgot to call the
inherited __del__ method, btw).
Mmm, doesn't work. I think there is an optimization somewhere: if it looks
like a real file object, it uses the original file write method, not yours.
The trick would be to use an object that does NOT inherit from file:

import sys
class TeeNoFile(object):
     def __init__(self, name, mode):
         self.file = open(name, mode)
         self.stdout = sys.stdout
         sys.stdout = self
     def close(self):
         if self.stdout is not None:
             sys.stdout = self.stdout
             self.stdout = None
         if self.file is not None:
             self.file.close()
             self.file = None
     def write(self, data):
         self.file.write(data)
         self.stdout.write(data)
     def flush(self):
         self.file.flush()
         self.stdout.flush()
     def __del__(self):
         self.close()

tee=TeeNoFile('logfile', 'w')
print 'abcdefg'
print 'another line'
tee.close()
print 'screen only'
del tee # should do nothing

--
Gabriel Genellina

Crowfoot

Answer 15

I'm writing a script to run cmd-line scripts. ( Because in some cases, there just is no viable substitute for a Linux command -- such as the case of rsync. )

What I really wanted was to use the default python logging mechanism in every case where it was possible to do so, but to still capture any error when something went wrong that was unanticipated.

This code seems to do the trick. It may not be particularly elegant or efficient ( although it doesn't use string+=string, so at least it doesn't have that particular potential bottle- neck ). I'm posting it in case it gives someone else any useful ideas.

import logging
import os, sys
import datetime

# Get name of module, use as application name
try:
  ME=os.path.split(__file__)[-1].split('.')[0]
except:
  ME='pyExec_'

LOG_IDENTIFIER="uuu___( o O )___uuu "
LOG_IDR_LENGTH=len(LOG_IDENTIFIER)

class PyExec(object):

  # Use this to capture all possible error / output to log
  class SuperTee(object):
      # Original reference: http://mail.python.org/pipermail/python-list/2007-May/442737.html
      def __init__(self, name, mode):
          self.fl = open(name, mode)
          self.fl.write('\n')
          self.stdout = sys.stdout
          self.stdout.write('\n')
          self.stderr = sys.stderr

          sys.stdout = self
          sys.stderr = self

      def __del__(self):
          self.fl.write('\n')
          self.fl.flush()
          sys.stderr = self.stderr
          sys.stdout = self.stdout
          self.fl.close()

      def write(self, data):
          # If the data to write includes the log identifier prefix, then it is already formatted
          if data[0:LOG_IDR_LENGTH]==LOG_IDENTIFIER:
            self.fl.write("%s\n" % data[LOG_IDR_LENGTH:])
            self.stdout.write(data[LOG_IDR_LENGTH:])

          # Otherwise, we can give it a timestamp
          else:

            timestamp=str(datetime.datetime.now())
            if 'Traceback' == data[0:9]:
              data='%s: %s' % (timestamp, data)
              self.fl.write(data)
            else:
              self.fl.write(data)

            self.stdout.write(data)


  def __init__(self, aName, aCmd, logFileName='', outFileName=''):

    # Using name for 'logger' (context?), which is separate from the module or the function
    baseFormatter=logging.Formatter("%(asctime)s \t %(levelname)s \t %(name)s:%(module)s:%(lineno)d \t %(message)s")
    errorFormatter=logging.Formatter(LOG_IDENTIFIER + "%(asctime)s \t %(levelname)s \t %(name)s:%(module)s:%(lineno)d \t %(message)s")

    if logFileName:
      # open passed filename as append
      fl=logging.FileHandler("%s.log" % aName)
    else:
      # otherwise, use log filename as a one-time use file
      fl=logging.FileHandler("%s.log" % aName, 'w')

    fl.setLevel(logging.DEBUG)
    fl.setFormatter(baseFormatter)

    # This will capture stdout and CRITICAL and beyond errors

    if outFileName:
      teeFile=PyExec.SuperTee("%s_out.log" % aName)
    else:
      teeFile=PyExec.SuperTee("%s_out.log" % aName, 'w')

    fl_out=logging.StreamHandler( teeFile )
    fl_out.setLevel(logging.CRITICAL)
    fl_out.setFormatter(errorFormatter)

    # Set up logging
    self.log=logging.getLogger('pyExec_main')
    log=self.log

    log.addHandler(fl)
    log.addHandler(fl_out)

    print "Test print statement."

    log.setLevel(logging.DEBUG)

    log.info("Starting %s", ME)
    log.critical("Critical.")

    # Caught exception
    try:
      raise Exception('Exception test.')
    except Exception,e:
      log.exception(str(e))

    # Uncaught exception
    a=2/0


PyExec('test_pyExec',None)

Obviously, if you're not as subject to whimsy as I am, replace LOG_IDENTIFIER with another string that you're not like to ever see someone write to a log.

Domain

Answer 16

You can also add stderr as well, based on shx2's answer above using class multifile :

class Log(object):

    def __init__(self, path_log, mode="w", encoding="utf-8"):
        h = open(path_log, mode, encoding=encoding)
        sys.stdout = multifile([ sys.stdout, h ])
        sys.stderr = multifile([ sys.stderr, h ])

    def __enter__(self):
        """ Necessary if called by with (or with... as) """
        return self     # only necessary if "as"

    def __exit__(self, type, value, tb):
        """ Necessary if call by with """
        pass

    def __del__(self):
        if sys is not None:
            # restoring
            sys.stdout = sys.__stdout__
            sys.stderr = sys.__stderr__

log = Log("test.txt")
print("line 1")
print("line 2", file=sys.stderr)
del log
print("line 3 only on screen")

Aphasic

The only drawback is in python 2.x when an exception is raised : the message seems to be lost. – Aphasic

Answer 17

Here's a context manager that temporarily duplicates stdout to a file. It's an improvement in my view because it resets sys.stdout & closes the file even when exceptions occur, and the syntax is indicative of an invisible change in the background. Expaneded on John T's solution.

class DuplicateStdout:
    def __init__(self, path):
        self.stdout = sys.stdout
        self.path = path
        self.f = None
    
    def write(self, s):
        self.stdout.write(s)
        self.f.write(s)

    def __enter__(self):
        self.f = open(self.path, "w")
        sys.stdout = self
    
    def __exit__(self, *args):
        sys.stdout = self.stdout
        self.f.close()

Example usage:

with DuplicateStdout("foo.log"):
    print("Hey") # also in foo.log

print("There") # not in foo.log

Crowded

Answer 18

I wrote a full replacement for sys.stderr and just duplicated the code renaming stderr to stdout to make it also available to replace sys.stdout.

To do this I create the same object type as the current stderr and stdout, and forward all methods to the original system stderr and stdout:

import os
import sys
import logging

class StdErrReplament(object):
    """
        How to redirect stdout and stderr to logger in Python
        https://mcmap.net/q/37814/-how-to-redirect-stdout-and-stderr-to-logger-in-python

        Set a Read-Only Attribute in Python?
        https://mcmap.net/q/37815/-set-a-read-only-attribute-in-python
    """
    is_active = False

    @classmethod
    def lock(cls, logger):
        """
            Attach this singleton logger to the `sys.stderr` permanently.
        """
        global _stderr_singleton
        global _stderr_default
        global _stderr_default_class_type

        # On Sublime Text, `sys.__stderr__` is set to None, because they already replaced `sys.stderr`
        # by some `_LogWriter()` class, then just save the current one over there.
        if not sys.__stderr__:
            sys.__stderr__ = sys.stderr

        try:
            _stderr_default
            _stderr_default_class_type

        except NameError:
            _stderr_default = sys.stderr
            _stderr_default_class_type = type( _stderr_default )

        # Recreate the sys.stderr logger when it was reset by `unlock()`
        if not cls.is_active:
            cls.is_active = True
            _stderr_write = _stderr_default.write

            logger_call = logger.debug
            clean_formatter = logger.clean_formatter

            global _sys_stderr_write
            global _sys_stderr_write_hidden

            if sys.version_info <= (3,2):
                logger.file_handler.terminator = '\n'

            # Always recreate/override the internal write function used by `_sys_stderr_write`
            def _sys_stderr_write_hidden(*args, **kwargs):
                """
                    Suppress newline in Python logging module
                    https://mcmap.net/q/37816/-suppress-newline-in-python-logging-module
                """

                try:
                    _stderr_write( *args, **kwargs )
                    file_handler = logger.file_handler

                    formatter = file_handler.formatter
                    terminator = file_handler.terminator

                    file_handler.formatter = clean_formatter
                    file_handler.terminator = ""

                    kwargs['extra'] = {'_duplicated_from_file': True}
                    logger_call( *args, **kwargs )

                    file_handler.formatter = formatter
                    file_handler.terminator = terminator

                except Exception:
                    logger.exception( "Could not write to the file_handler: %s(%s)", file_handler, logger )
                    cls.unlock()

            # Only create one `_sys_stderr_write` function pointer ever
            try:
                _sys_stderr_write

            except NameError:

                def _sys_stderr_write(*args, **kwargs):
                    """
                        Hides the actual function pointer. This allow the external function pointer to
                        be cached while the internal written can be exchanged between the standard
                        `sys.stderr.write` and our custom wrapper around it.
                    """
                    _sys_stderr_write_hidden( *args, **kwargs )

        try:
            # Only create one singleton instance ever
            _stderr_singleton

        except NameError:

            class StdErrReplamentHidden(_stderr_default_class_type):
                """
                    Which special methods bypasses __getattribute__ in Python?
                    https://mcmap.net/q/37817/-which-special-methods-bypasses-__getattribute__-in-python
                """

                if hasattr( _stderr_default, "__abstractmethods__" ):
                    __abstractmethods__ = _stderr_default.__abstractmethods__

                if hasattr( _stderr_default, "__base__" ):
                    __base__ = _stderr_default.__base__

                if hasattr( _stderr_default, "__bases__" ):
                    __bases__ = _stderr_default.__bases__

                if hasattr( _stderr_default, "__basicsize__" ):
                    __basicsize__ = _stderr_default.__basicsize__

                if hasattr( _stderr_default, "__call__" ):
                    __call__ = _stderr_default.__call__

                if hasattr( _stderr_default, "__class__" ):
                    __class__ = _stderr_default.__class__

                if hasattr( _stderr_default, "__delattr__" ):
                    __delattr__ = _stderr_default.__delattr__

                if hasattr( _stderr_default, "__dict__" ):
                    __dict__ = _stderr_default.__dict__

                if hasattr( _stderr_default, "__dictoffset__" ):
                    __dictoffset__ = _stderr_default.__dictoffset__

                if hasattr( _stderr_default, "__dir__" ):
                    __dir__ = _stderr_default.__dir__

                if hasattr( _stderr_default, "__doc__" ):
                    __doc__ = _stderr_default.__doc__

                if hasattr( _stderr_default, "__eq__" ):
                    __eq__ = _stderr_default.__eq__

                if hasattr( _stderr_default, "__flags__" ):
                    __flags__ = _stderr_default.__flags__

                if hasattr( _stderr_default, "__format__" ):
                    __format__ = _stderr_default.__format__

                if hasattr( _stderr_default, "__ge__" ):
                    __ge__ = _stderr_default.__ge__

                if hasattr( _stderr_default, "__getattribute__" ):
                    __getattribute__ = _stderr_default.__getattribute__

                if hasattr( _stderr_default, "__gt__" ):
                    __gt__ = _stderr_default.__gt__

                if hasattr( _stderr_default, "__hash__" ):
                    __hash__ = _stderr_default.__hash__

                if hasattr( _stderr_default, "__init__" ):
                    __init__ = _stderr_default.__init__

                if hasattr( _stderr_default, "__init_subclass__" ):
                    __init_subclass__ = _stderr_default.__init_subclass__

                if hasattr( _stderr_default, "__instancecheck__" ):
                    __instancecheck__ = _stderr_default.__instancecheck__

                if hasattr( _stderr_default, "__itemsize__" ):
                    __itemsize__ = _stderr_default.__itemsize__

                if hasattr( _stderr_default, "__le__" ):
                    __le__ = _stderr_default.__le__

                if hasattr( _stderr_default, "__lt__" ):
                    __lt__ = _stderr_default.__lt__

                if hasattr( _stderr_default, "__module__" ):
                    __module__ = _stderr_default.__module__

                if hasattr( _stderr_default, "__mro__" ):
                    __mro__ = _stderr_default.__mro__

                if hasattr( _stderr_default, "__name__" ):
                    __name__ = _stderr_default.__name__

                if hasattr( _stderr_default, "__ne__" ):
                    __ne__ = _stderr_default.__ne__

                if hasattr( _stderr_default, "__new__" ):
                    __new__ = _stderr_default.__new__

                if hasattr( _stderr_default, "__prepare__" ):
                    __prepare__ = _stderr_default.__prepare__

                if hasattr( _stderr_default, "__qualname__" ):
                    __qualname__ = _stderr_default.__qualname__

                if hasattr( _stderr_default, "__reduce__" ):
                    __reduce__ = _stderr_default.__reduce__

                if hasattr( _stderr_default, "__reduce_ex__" ):
                    __reduce_ex__ = _stderr_default.__reduce_ex__

                if hasattr( _stderr_default, "__repr__" ):
                    __repr__ = _stderr_default.__repr__

                if hasattr( _stderr_default, "__setattr__" ):
                    __setattr__ = _stderr_default.__setattr__

                if hasattr( _stderr_default, "__sizeof__" ):
                    __sizeof__ = _stderr_default.__sizeof__

                if hasattr( _stderr_default, "__str__" ):
                    __str__ = _stderr_default.__str__

                if hasattr( _stderr_default, "__subclasscheck__" ):
                    __subclasscheck__ = _stderr_default.__subclasscheck__

                if hasattr( _stderr_default, "__subclasses__" ):
                    __subclasses__ = _stderr_default.__subclasses__

                if hasattr( _stderr_default, "__subclasshook__" ):
                    __subclasshook__ = _stderr_default.__subclasshook__

                if hasattr( _stderr_default, "__text_signature__" ):
                    __text_signature__ = _stderr_default.__text_signature__

                if hasattr( _stderr_default, "__weakrefoffset__" ):
                    __weakrefoffset__ = _stderr_default.__weakrefoffset__

                if hasattr( _stderr_default, "mro" ):
                    mro = _stderr_default.mro

                def __init__(self):
                    """
                        Override any super class `type( _stderr_default )` constructor, so we can 
                        instantiate any kind of `sys.stderr` replacement object, in case it was already 
                        replaced by something else like on Sublime Text with `_LogWriter()`.

                        Assures all attributes were statically replaced just above. This should happen in case
                        some new attribute is added to the python language.

                        This also ignores the only two methods which are not equal, `__init__()` and `__getattribute__()`.
                    """
                    different_methods = ("__init__", "__getattribute__")
                    attributes_to_check = set( dir( object ) + dir( type ) )

                    for attribute in attributes_to_check:

                        if attribute not in different_methods \
                                and hasattr( _stderr_default, attribute ):

                            base_class_attribute = super( _stderr_default_class_type, self ).__getattribute__( attribute )
                            target_class_attribute = _stderr_default.__getattribute__( attribute )

                            if base_class_attribute != target_class_attribute:
                                sys.stderr.write( "    The base class attribute `%s` is different from the target class:\n%s\n%s\n\n" % (
                                        attribute, base_class_attribute, target_class_attribute ) )

                def __getattribute__(self, item):

                    if item == 'write':
                        return _sys_stderr_write

                    try:
                        return _stderr_default.__getattribute__( item )

                    except AttributeError:
                        return super( _stderr_default_class_type, _stderr_default ).__getattribute__( item )

            _stderr_singleton = StdErrReplamentHidden()
            sys.stderr = _stderr_singleton

        return cls

    @classmethod
    def unlock(cls):
        """
            Detach this `stderr` writer from `sys.stderr` and allow the next call to `lock()` create
            a new writer for the stderr.
        """

        if cls.is_active:
            global _sys_stderr_write_hidden

            cls.is_active = False
            _sys_stderr_write_hidden = _stderr_default.write



class StdOutReplament(object):
    """
        How to redirect stdout and stderr to logger in Python
        https://mcmap.net/q/37814/-how-to-redirect-stdout-and-stderr-to-logger-in-python

        Set a Read-Only Attribute in Python?
        https://mcmap.net/q/37815/-set-a-read-only-attribute-in-python
    """
    is_active = False

    @classmethod
    def lock(cls, logger):
        """
            Attach this singleton logger to the `sys.stdout` permanently.
        """
        global _stdout_singleton
        global _stdout_default
        global _stdout_default_class_type

        # On Sublime Text, `sys.__stdout__` is set to None, because they already replaced `sys.stdout`
        # by some `_LogWriter()` class, then just save the current one over there.
        if not sys.__stdout__:
            sys.__stdout__ = sys.stdout

        try:
            _stdout_default
            _stdout_default_class_type

        except NameError:
            _stdout_default = sys.stdout
            _stdout_default_class_type = type( _stdout_default )

        # Recreate the sys.stdout logger when it was reset by `unlock()`
        if not cls.is_active:
            cls.is_active = True
            _stdout_write = _stdout_default.write

            logger_call = logger.debug
            clean_formatter = logger.clean_formatter

            global _sys_stdout_write
            global _sys_stdout_write_hidden

            if sys.version_info <= (3,2):
                logger.file_handler.terminator = '\n'

            # Always recreate/override the internal write function used by `_sys_stdout_write`
            def _sys_stdout_write_hidden(*args, **kwargs):
                """
                    Suppress newline in Python logging module
                    https://mcmap.net/q/37816/-suppress-newline-in-python-logging-module
                """

                try:
                    _stdout_write( *args, **kwargs )
                    file_handler = logger.file_handler

                    formatter = file_handler.formatter
                    terminator = file_handler.terminator

                    file_handler.formatter = clean_formatter
                    file_handler.terminator = ""

                    kwargs['extra'] = {'_duplicated_from_file': True}
                    logger_call( *args, **kwargs )

                    file_handler.formatter = formatter
                    file_handler.terminator = terminator

                except Exception:
                    logger.exception( "Could not write to the file_handler: %s(%s)", file_handler, logger )
                    cls.unlock()

            # Only create one `_sys_stdout_write` function pointer ever
            try:
                _sys_stdout_write

            except NameError:

                def _sys_stdout_write(*args, **kwargs):
                    """
                        Hides the actual function pointer. This allow the external function pointer to
                        be cached while the internal written can be exchanged between the standard
                        `sys.stdout.write` and our custom wrapper around it.
                    """
                    _sys_stdout_write_hidden( *args, **kwargs )

        try:
            # Only create one singleton instance ever
            _stdout_singleton

        except NameError:

            class StdOutReplamentHidden(_stdout_default_class_type):
                """
                    Which special methods bypasses __getattribute__ in Python?
                    https://mcmap.net/q/37817/-which-special-methods-bypasses-__getattribute__-in-python
                """

                if hasattr( _stdout_default, "__abstractmethods__" ):
                    __abstractmethods__ = _stdout_default.__abstractmethods__

                if hasattr( _stdout_default, "__base__" ):
                    __base__ = _stdout_default.__base__

                if hasattr( _stdout_default, "__bases__" ):
                    __bases__ = _stdout_default.__bases__

                if hasattr( _stdout_default, "__basicsize__" ):
                    __basicsize__ = _stdout_default.__basicsize__

                if hasattr( _stdout_default, "__call__" ):
                    __call__ = _stdout_default.__call__

                if hasattr( _stdout_default, "__class__" ):
                    __class__ = _stdout_default.__class__

                if hasattr( _stdout_default, "__delattr__" ):
                    __delattr__ = _stdout_default.__delattr__

                if hasattr( _stdout_default, "__dict__" ):
                    __dict__ = _stdout_default.__dict__

                if hasattr( _stdout_default, "__dictoffset__" ):
                    __dictoffset__ = _stdout_default.__dictoffset__

                if hasattr( _stdout_default, "__dir__" ):
                    __dir__ = _stdout_default.__dir__

                if hasattr( _stdout_default, "__doc__" ):
                    __doc__ = _stdout_default.__doc__

                if hasattr( _stdout_default, "__eq__" ):
                    __eq__ = _stdout_default.__eq__

                if hasattr( _stdout_default, "__flags__" ):
                    __flags__ = _stdout_default.__flags__

                if hasattr( _stdout_default, "__format__" ):
                    __format__ = _stdout_default.__format__

                if hasattr( _stdout_default, "__ge__" ):
                    __ge__ = _stdout_default.__ge__

                if hasattr( _stdout_default, "__getattribute__" ):
                    __getattribute__ = _stdout_default.__getattribute__

                if hasattr( _stdout_default, "__gt__" ):
                    __gt__ = _stdout_default.__gt__

                if hasattr( _stdout_default, "__hash__" ):
                    __hash__ = _stdout_default.__hash__

                if hasattr( _stdout_default, "__init__" ):
                    __init__ = _stdout_default.__init__

                if hasattr( _stdout_default, "__init_subclass__" ):
                    __init_subclass__ = _stdout_default.__init_subclass__

                if hasattr( _stdout_default, "__instancecheck__" ):
                    __instancecheck__ = _stdout_default.__instancecheck__

                if hasattr( _stdout_default, "__itemsize__" ):
                    __itemsize__ = _stdout_default.__itemsize__

                if hasattr( _stdout_default, "__le__" ):
                    __le__ = _stdout_default.__le__

                if hasattr( _stdout_default, "__lt__" ):
                    __lt__ = _stdout_default.__lt__

                if hasattr( _stdout_default, "__module__" ):
                    __module__ = _stdout_default.__module__

                if hasattr( _stdout_default, "__mro__" ):
                    __mro__ = _stdout_default.__mro__

                if hasattr( _stdout_default, "__name__" ):
                    __name__ = _stdout_default.__name__

                if hasattr( _stdout_default, "__ne__" ):
                    __ne__ = _stdout_default.__ne__

                if hasattr( _stdout_default, "__new__" ):
                    __new__ = _stdout_default.__new__

                if hasattr( _stdout_default, "__prepare__" ):
                    __prepare__ = _stdout_default.__prepare__

                if hasattr( _stdout_default, "__qualname__" ):
                    __qualname__ = _stdout_default.__qualname__

                if hasattr( _stdout_default, "__reduce__" ):
                    __reduce__ = _stdout_default.__reduce__

                if hasattr( _stdout_default, "__reduce_ex__" ):
                    __reduce_ex__ = _stdout_default.__reduce_ex__

                if hasattr( _stdout_default, "__repr__" ):
                    __repr__ = _stdout_default.__repr__

                if hasattr( _stdout_default, "__setattr__" ):
                    __setattr__ = _stdout_default.__setattr__

                if hasattr( _stdout_default, "__sizeof__" ):
                    __sizeof__ = _stdout_default.__sizeof__

                if hasattr( _stdout_default, "__str__" ):
                    __str__ = _stdout_default.__str__

                if hasattr( _stdout_default, "__subclasscheck__" ):
                    __subclasscheck__ = _stdout_default.__subclasscheck__

                if hasattr( _stdout_default, "__subclasses__" ):
                    __subclasses__ = _stdout_default.__subclasses__

                if hasattr( _stdout_default, "__subclasshook__" ):
                    __subclasshook__ = _stdout_default.__subclasshook__

                if hasattr( _stdout_default, "__text_signature__" ):
                    __text_signature__ = _stdout_default.__text_signature__

                if hasattr( _stdout_default, "__weakrefoffset__" ):
                    __weakrefoffset__ = _stdout_default.__weakrefoffset__

                if hasattr( _stdout_default, "mro" ):
                    mro = _stdout_default.mro

                def __init__(self):
                    """
                        Override any super class `type( _stdout_default )` constructor, so we can 
                        instantiate any kind of `sys.stdout` replacement object, in case it was already 
                        replaced by something else like on Sublime Text with `_LogWriter()`.

                        Assures all attributes were statically replaced just above. This should happen in case
                        some new attribute is added to the python language.

                        This also ignores the only two methods which are not equal, `__init__()` and `__getattribute__()`.
                    """
                    different_methods = ("__init__", "__getattribute__")
                    attributes_to_check = set( dir( object ) + dir( type ) )

                    for attribute in attributes_to_check:

                        if attribute not in different_methods \
                                and hasattr( _stdout_default, attribute ):

                            base_class_attribute = super( _stdout_default_class_type, self ).__getattribute__( attribute )
                            target_class_attribute = _stdout_default.__getattribute__( attribute )

                            if base_class_attribute != target_class_attribute:
                                sys.stdout.write( "    The base class attribute `%s` is different from the target class:\n%s\n%s\n\n" % (
                                        attribute, base_class_attribute, target_class_attribute ) )

                def __getattribute__(self, item):

                    if item == 'write':
                        return _sys_stdout_write

                    try:
                        return _stdout_default.__getattribute__( item )

                    except AttributeError:
                        return super( _stdout_default_class_type, _stdout_default ).__getattribute__( item )

            _stdout_singleton = StdOutReplamentHidden()
            sys.stdout = _stdout_singleton

        return cls

    @classmethod
    def unlock(cls):
        """
            Detach this `stdout` writer from `sys.stdout` and allow the next call to `lock()` create
            a new writer for the stdout.
        """

        if cls.is_active:
            global _sys_stdout_write_hidden

            cls.is_active = False
            _sys_stdout_write_hidden = _stdout_default.write

To use this you can just call StdErrReplament::lock(logger) and StdOutReplament::lock(logger) passing the logger you want to use to send the output text. For example:

import os
import sys
import logging

current_folder = os.path.dirname( os.path.realpath( __file__ ) )
log_file_path = os.path.join( current_folder, "my_log_file.txt" )

file_handler = logging.FileHandler( log_file_path, 'a' )
file_handler.formatter = logging.Formatter( "%(asctime)s %(name)s %(levelname)s - %(message)s", "%Y-%m-%d %H:%M:%S" )

log = logging.getLogger( __name__ )
log.setLevel( "DEBUG" )
log.addHandler( file_handler )

log.file_handler = file_handler
log.clean_formatter = logging.Formatter( "", "" )

StdOutReplament.lock( log )
StdErrReplament.lock( log )

log.debug( "I am doing usual logging debug..." )
sys.stderr.write( "Tests 1...\n" )
sys.stdout.write( "Tests 2...\n" )

Running this code, you will see on the screen:

And on the file contents:

If you would like to also see the contents of the log.debug calls on the screen, you will need to add a stream handler to your logger. On this case it would be like this:

import os
import sys
import logging

class ContextFilter(logging.Filter):
    """ This filter avoids duplicated information to be displayed to the StreamHandler log. """
    def filter(self, record):
        return not "_duplicated_from_file" in record.__dict__

current_folder = os.path.dirname( os.path.realpath( __file__ ) )
log_file_path = os.path.join( current_folder, "my_log_file.txt" )

stream_handler = logging.StreamHandler()
file_handler = logging.FileHandler( log_file_path, 'a' )

formatter = logging.Formatter( "%(asctime)s %(name)s %(levelname)s - %(message)s", "%Y-%m-%d %H:%M:%S" )
file_handler.formatter = formatter
stream_handler.formatter = formatter
stream_handler.addFilter( ContextFilter() )

log = logging.getLogger( __name__ )
log.setLevel( "DEBUG" )
log.addHandler( file_handler )
log.addHandler( stream_handler )

log.file_handler = file_handler
log.stream_handler = stream_handler
log.clean_formatter = logging.Formatter( "", "" )

StdOutReplament.lock( log )
StdErrReplament.lock( log )

log.debug( "I am doing usual logging debug..." )
sys.stderr.write( "Tests 1...\n" )
sys.stdout.write( "Tests 2...\n" )

Which would output like this when running:

While it would still saving this to the file my_log_file.txt:

When disabling this with StdErrReplament:unlock(), it will only restore the standard behavior of the stderr stream, as the attached logger cannot be never detached because someone else can have a reference to its older version. This is why it is a global singleton which can never dies. Therefore, in case of reloading this module with imp or something else, it will never recapture the current sys.stderr as it was already injected on it and have it saved internally.

Diep

an amazing level of accidental complexity for duplicating a stream. – Logotype