a Python library for shelling out.
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.


duct.py Build Status Build status Coverage Status PyPI version

A Python library for shelling out. One of the goals of duct is to be easily portable to other languages, and there's a Rust version happening in parallel.

PyPI package

Easy things should be easy.

But always be explicit about what happens to output.

from duct import cmd, sh

# Read the name of the current git branch.
current_branch = sh('git symbolic-ref --short HEAD').read()

# Log the current branch, with git taking over the terminal as usual.
cmd('git', 'log', current_branch).run()

That's exactly the same as the following in standard Python 3.5:

from subprocess import run, PIPE

result = run('git symbolic-ref --short HEAD', shell=True, stdout=PIPE,
             universal_newlines=True, check=True)
current_branch = result.stdout.rstrip('\n')

run(['git', 'log', current_branch], check=True)

Crazy things should be possible.

Sometimes you have to write ridiculous pipelines in bash:

(echo error >&2 && echo output) 2>&1 | grep stuff

The duct version is longer, but duct expressions are composable objects, so we can build the whole command piece-by-piece:

from duct import cmd, sh

echoes = cmd('echo', 'error').stdout_to_stderr().then(cmd('echo', 'output'))
pipeline = echoes.stderr_to_stdout().pipe(sh('grep stuff'))
pipeline.run()  # This will raise an exception! See below.

Errors should never pass silently.

Because grep in the example above doesn't match any lines, it's going to return an error code, and duct will raise an exception. To ignore the error, you can use unchecked:

result = cmd('false').unchecked().run()
print(result.status)  # 0

If you need to know the value of a non-zero exit status, you can catch the exception it raises and inspect it like this.

from duct import cmd, StatusError

except StatusError as e:
    print(e.result.status)  # 1

Note that duct treats errors in a pipe like bash's pipefail option: they count even when they happen on the left. This can be surprising in cases where we usually ignore errors. In the following example, cat returns an error because its stdout is closed:

# Raises an exception, because cat returns an error.

Work with pathlib.

If you have a Path object, you can use it anywhere you would use a string.

from duct import cmd
from pathlib import Path

myscript = Path('foo')
mydir = Path('bar')


Expression starting functions


Create a command expression from a program name and optional arguments. This doesn't require escaping any special characters or whitespace. If your arguments are anything other than constant strings, this is definitely what you want to use.

x = "hi"
cmd("echo", x).run()


Create a command expression from a string of shell code, executed with the shell=True flag in the subprocess module. This can spare you from typing a lot of quote characters, or even whole pipelines, but please don't use it with anything other than a constant string, because shell escaping is tricky.

sh("echo hi").run()

Execution methods


Execute the expression and return a Result object, which has fields stdout, stderr, and status. By default, the child process shares the stdin/stdout/stderr pipes of the parent, and no output is captured. If the expression has a non-zero exit status, run will raise an exception.

result = sh("echo foo").stdout_capture().run()
assert result.status == 0
assert result.stdout == b"foo\n"
assert result.stderr == b""


Execute the expression and capture its output, similar to backticks or $() in bash. This is a convenience wrapper around run which sets stdout_capture, decodes stdout as UTF-8, trims trailing newlines, and returns it directly instead of returning a Result. Note that in Python 2 the return value is a unicode string.

output = sh("echo 日本語").read()
assert output == u"日本語"


Start the expression running in the background and immediately return a WaitHandle. Calling wait on the handle waits for the expression to finish running and then returns a Result, so start followed by wait is equivalent to run.

handle = sh("echo foo").stdout_capture().start()
result = handle.wait()
assert result.status == 0
assert result.stdout == b"foo\n"
assert result.stderr == b""

Pipe building methods


Create a pipe expression, similar to | in bash. The argument becomes the right side of the pipe, and it can be any duct expression. The status of a pipe expression is equal to the right side's status if it's nonzero, otherwise the left side's.

output = sh("echo dog").pipe(sh("sed s/o/a/")).read()
assert output == "dag"


Create a sequence expression, similar to && in bash, and used like pipe above. The left side runs, and then if its status is zero, the right side runs. If you want to ignore errors on the left side, similar to ; in bash, use unchecked around the left expression.

cmd("false").then(sh("echo we never get here")).run()  # StatusError

Redirections etc.


Redirects an expression's stdin to read from a string or bytes object. Duct will spawn a writer thread at runtime.

output = cmd("cat").input("stuff").read()
assert output == "stuff"

stdin, stdin_file, stdin_null

Redirects an expression's stdin to read from a file. The file can be a string/bytes/pathlib path to open at runtime, or with stdin_file an already open file or descriptor. stdin_null redirects stdin to /dev/null on Unix and nul on Windows.


stdout, stdout_file, stdout_null, stdout_capture, stdout_to_stderr

Redirects an expression's stdout to write to a file. The file can by a string/bytes/pathlib path to open at runtime, or with stdout_file an already open file or descriptor. stdout_null redirects to /dev/null on Unix or nul on Windows. stdout_capture redirects to a pipe whose output bytes end up as Result.stdout. stdout_to_stderr replaces stdout with a copy of the stderr pipe.

from duct import sh
from pathlib import Path

temp_dir = sh("mktemp -d").read()
temp_file = Path(temp_dir) / "file.txt"
sh("echo some stuff").stdout(temp_file).run()

result = sh("echo more stuff").stdout_capture().run()
assert result.stdout == b"more stuff\n"

stderr, stderr_file, stderr_null, stderr_capture, stderr_to_stdout

Analogous to the stdout methods. stderr_capture redirects to a pipe whose output bytes end up as Result.stderr.

from duct import sh

sh("echo output && echo junk >&2").stderr_null().run()

result = sh("echo error stuff >&2").stderr_capture().run()
assert result.stderr == b"error stuff\n"


Sets the working directory an expression will execute with. The default is the working directory of the parent.

output = cmd("pwd").dir("/").read()
assert output == "/"


Sets an environment variable for an expression, given a name and a value.

output = sh("echo $FOO").env("FOO", "bar").read()
assert output == "bar"


Unsets an environment variable for an expression, whether it's from the parent environment, or from an exterior (but not interior) call to env.

os.environ["FOO"] = "bar"
output = sh("echo $FOO").env_remove("FOO").read()
assert output == ""


Sets the entire environment for an expression, so that nothing is inherited. This includes both the parent processes's environment, and any calls to env in parent expressions.

# BAR and BAZ are guaranteed to be undefined when this runs.
prog = sh("echo $FOO$BAR$BAZ").full_env({"FOO": "1"})

# This env var would normally get inherited by the child, but full_env
# above will prevent it.
os.environ["BAR"] = "2"

# This env call also gets suppressed.
output = prog.env("BAZ", "3").read()
assert output == "1"


Prevents a non-zero exit status from short-circuiting then expressions or from causing run and friends to return an error. The unchecked exit code will still be there on the Result returned by run; its value doesn't change.

"Uncheckedness" sticks to an exit code as it bubbles up through complicated expressions, but it doesn't "infect" other exit codes. So for example, if only one sub-expression in a pipe has unchecked, then errors returned by the other side will still be checked. That said, most commonly you'll just call unchecked right before run, and it'll apply to an entire expression. This sub-expression stuff doesn't usually come up unless you have a big pipeline built out of lots of different pieces.

# Raises a StatusError!

# Does not raise an error.
result = cmd("false").unchecked().run()
assert result.status == 0