Skip to content
/ pipepy Public

A Python library for invoking and interacting with shell commands.


Notifications You must be signed in to change notification settings


Repository files navigation

A Python library for invoking and interacting with shell commands.


Table of contents

Why? Comparison with other similar frameworks

  1. Xonsh: Xonsh allows you to combine shell and Python and enables very powerful scripting and interactive sessions. This library does the same to a limited degree. However, Xonsh introduces a new language that is a superset of Python. The main goal of this library that sets it apart is that it is intended to be a pure Python implementation, mainly aimed at scripting.

  2. sh and pieshell: These are much closer to the current library in that they are pure Python implementations. The current library, however, tries to improve on the following aspects:

    • It tries to apply more syntactic sugar to make the invocations feel more like shell invocations.

    • It tries to offer ways to have shell commands interact with python code in powerful and intuitive ways.

Installation and testing

python -m pip install pipepy

Or, if you want to modify the code while trying it out:

git clone
cd pipepy
python -m pip install  -e .

To run the tests, you need to first install the testing requirements:

python -m pip install -r test_requirements.txt

pymake test
# or

There are a few more pymake targets to assist with testing during development:

  • covtest: Produces and opens a coverage report
  • watchtest: Listens for changes in the source code files and reruns the tests automatically
  • debugtest: Runs the tests without capturing their output so that you can insert a debug statement

pymake is a console script that is part of pipepy that aims to be a replacement for GNU make, with the difference that the Makefiles are written in Python. More on this below.

Intro, basic usage

from pipepy import ls, grep

print(ls)  # prints contents of current folder
if ls | grep('info.txt'):
      print('info.txt found')

Most shell commands are importable straight from the pipepy module. Dashes in commands' names are converted to underscore (docker-composedocker_compose). Commands that cannot be found automatically can be created with the PipePy constructor:

from pipepy import PipePy

custom_command = PipePy('./bin/custom')
python_script = PipePy('python', '')

Customizing commands

Calling a command with non empty arguments will return a modified unevaluated copy. So the following are equivalent:

from pipepy import PipePy
ls_l = PipePy('ls', '-l')
# Is equivalent to
ls_l = PipePy('ls')('-l')

There is a number of other ways you can customize a command:

  • Globs: globbing will be applied to all positional arguments:

    from pipepy import echo
    print(echo('*'))  # Will print all files in the current folder

    You can use glob.escape if you want to avoid this functionality:

    import glob
    from pipepy import ls, echo
    # <<< **a *a *aa
    # <<< **a *a *aa
    # <<< *a
  • Keyword arguments:

    from pipepy import ls
    ls(sort="size")     # Equivalent to ls('--sort=size')
    ls(I="files.txt")   # Equivalent to ls('-I', 'files.txt')
    ls(sort_by="size")  # Equivalent to ls('--sort-by=size')
    ls(escape=True)     # Equivalent to ls('--escape')
    ls(escape=False)    # Equivalent to ls('--no-escape')

    Since keyword arguments come after positional arguments, if you want the final command to have a different ordering you can invoke the command multiple times:

    from pipepy import ls
    ls('-l', sort="size")  # Equivalent to ls('-l', '--sort=size')
    ls(sort="size")('-l')  # Equivalent to ls('--sort=size', '-l')
  • Attribute access:

    from pipepy import git
    git.push.origin.bugfixes  # Equivalent to git('push', 'origin', 'bugfixes')
  • Minus sign:

    from pipepy import ls
    ls - 'l'        # Equivalent to ls('-l')
    ls - 'default'  # Equivalent to ls('--default')

    This is to enable making the invocations look more like the shell:

    from pipepy import ls
    l, t = 'l', 't'
    ls -l -t  # Equivalent to ls('-l', '-t')

    You can call pipepy.overload_chars(locals()) in your script to assign all ascii letters to variables of the same name.

    import pipepy
    from pipepy import ls
    ls -l -t  # Equivalent to ls('-l', '-t')


Commands are evaluated lazily. For example, this will not actually do anything:

from pipepy import wget

Invoking a PipePy instance with non-empty arguments will return an unevaluated copy supplied with the extra arguments. A command will be evaluated when its output is used. This can be done with the following ways:

  • Accessing the returncode, stdout and stderr properties:

    from pipepy import echo
    command = echo("hello world")
    # <<< 0
    # <<< 'hello world\n'
    # <<< ''
  • Evaluating the command as a string object

    from pipepy import ls
    result = str(ls)
    # or

    Converting a command to a str returns its stdout.

  • Evaluating the command as a boolean object:

    from pipepy import ls, grep
    command = ls | grep('info.txt')
    # <<< True
    if command:
        print("info.txt found")

    The command will be truthy if its returncode is 0.

  • Invoking the .as_table() method:

    from pipepy import ps
    # <<< [{'PID': '11233', 'TTY': 'pts/4', 'TIME': '00:00:01', 'CMD': 'zsh'},
    # ...  {'PID': '17673', 'TTY': 'pts/4', 'TIME': '00:00:08', 'CMD': 'ptipython'},
    # ...  {'PID': '18281', 'TTY': 'pts/4', 'TIME': '00:00:00', 'CMD': 'ps'}]
  • Iterating over a command object:

    from pipepy import ls
    for filename in ls:

    command.iter_words() iterates over the words of the command's stdout:

    from pipepy import ps
    # <<< ['PID', 'TTY', 'TIME', 'CMD',
    # ...  '11439', 'pts/5', '00:00:00', 'zsh',
    # ...  '15532', 'pts/5', '00:00:10', 'ptipython',
    # ...  '15539', 'pts/5', '00:00:00', 'ps']
  • Redirecting the output to something else (this will be further explained below):

    from pipepy import ls, grep
    ls > 'files.txt'
    ls >> 'files.txt'
    ls | grep('info.txt')  # `ls` will be evaluated, `grep` will not
    ls | lambda output: output.upper()

If you are not interested in the output of a command but want to evaluate it nevertheless, you can call it with empty arguments. So, this will actually invoke the command (and wait for it to finish).

from pipepy import wget

Background commands

Calling .delay() on a PipePy instance will return a copy that, although not evaluated, will have started running in the background (taking inspiration from Celery's .delay() method for the name). Again, if you try to access its output, it will perform the rest of the evaluation process, which is simply to wait for it to finish:

from pipepy import wget
urls = [...]

# All downloads will happen in the background simultaneously
downloads = [wget(url).delay() for url in urls]

# You can do something else here in Python while the downloads are working

# This will call __bool__ on all downloads and thus wait for them
if not all(downloads):
   print("Some downloads failed")

If you are not interested in the output of a background command, you should take care at some point to call .wait() on it. Otherwise its process will not be waited for and if the parent Python process ends, it will kill all the background processes:

from pipepy import wget
download = wget('...').delay()
# Do something else

You can supply the optional timeout argument to wait. If the timeout is set, it expires and the process hasn't finished, a TimeoutExpired exception will be raised. (This is the same TimeoutExpired exception class from the subprocess module but you can import it from the pipepy module too)

from pipepy import sleep
command = sleep(100).delay()
# <<< TimeoutExpired: Command '['sleep', '30']' timed out after 5 seconds

At any point, you can call to get a list of non-waited-for commands. In case you want to do some cleaning up, there is also a pipepy.wait_jobs() function. This should be used with care however as, if any of the background jobs aren't finished or are stuck, wait_jobs() may hang for an unknown amount of time. wait_jobs also accepts the optional timeout argument.

Redirecting output from/to files

The >, >> and < operators work similar to how they work in a shell:

ls               >  'files.txt'  # Will overwrite files.txt
ls               >> 'files.txt'  # Will append to files.txt
grep('info.txt') <  'files.txt'  # Will use files.txt as input

These also work with file-like objects:

import os
from pipepy import ls, grep

buf = io.StringIO()
ls > buf
ls('subfolder') >> buf
grep('filename') < buf

If you want to combine input and output redirections, you have to put the first redirection inside parentheses because of how python likes to deal with comparison chains:

from pipepy import gzip
gzip = gzip(_text=False)
gzip < 'uncompressed.txt' > 'uncompressed.txt.gz'    # Wrong!
(gzip < 'uncompressed.txt') > 'uncompressed.txt.gz'  # Correct!


The | operator is used to customize where a command gets its input from and what it does with its output. Depending on the types of the operands, different behaviors will emerge:

1. Both operands are PipePy instances

If both operands are commands, the result will be as similar as possible to what would have happened in a shell:

from pipepy import git, grep
if git.diff(name_only=True) | grep('readme.txt'):
      print("readme was changed")

If the left operand was previously evaluated, then it's output (stdout) will be passed directly as input to the right operand. Otherwise, both commands will be executed in parallel and left's output will be streamed into right.

2. Left operand is any kind of iterable (including string)

If the left operand is any kind of iterable, its elements will be fed to the command's stdin one by one:

import random
from pipepy import grep

result = ["John is 18 years old\n", "Mary is 25 years old"] | grep("Mary")
# <<< Mary is 25 years old

def my_stdin():
      for _ in range(500):
            yield f"{random.randint(1, 100)}\n"

result = my_stdin() | grep(17)
# <<< 17
# ... 17
# ... 17
# ... 17
# ... 17

If it's a string, it will be fed all at once

result = "John is 18 years old\nMary is 25 years old" | grep("Mary")

# Equivalent to

result = ["John is 18 years old\nMary is 25 years old"] | grep("Mary")

In both cases, ie in all cases where the right operand is a PipePy object, the return value of the pipe operation will be an unevaluated copy, which will be evaluated when we try to access its output. This means that we can take advantage of our usual background functionality:

from pipepy import find, xargs
command = find('.') | xargs.wc
command = command.delay()

# Do something else in the meantime

for line in command:  # Here we wait for the command to finish
    linecount, wordcount, charcount, filename = line.split()
    # ...

It also means that the left operand, if it's an iterable, will be consumed when the command is evaluated.

from pipepy import grep

iterable = (line for line in ["foo\n", "bar\n"])
command = iterable | grep("foo")
# <<< 'foo\n'
# <<< []

iterable = (line for line in ["foo\n", "bar\n"])
command = iterable | grep("foo")
list(iterable)  # Lets consume the iterable prematurely
# <<< ["foo\n", "bar\n"]
# <<< ''

Also, if you prefer an invocation style that resembles a function call more than a shell pipe operation, ie if you want to pass a command's input as an argument, you can use the _input keyword argument:

from pipepy import grep, ls

grep('setup', _input=ls)
# Is equivalent to
ls | grep('setup')

or use the square-bracket notation:

from pipepy import grep, ls

# Is equivalent to
ls | grep('setup')

(We use parentheses for arguments and square brackets for input because parentheses allow us to take advantage of keyword arguments which are a good fit for command-line options)

This works both for inputs that are iterables and commands.

3. Right operand is a function

The function's arguments need to either be:

  • a subset of returncode, output, errors or
  • a subset of stdout, stderr

The ordering of the arguments is irrelevant since the function's signature will be inspected to assign the proper values.

In the first case, the command will be waited for and its evaluated output will be made available to the function's arguments.

from pipepy import wc

def lines(output):
    for line in output.splitlines():
            lines, words, chars, filename = line.split()
        except ValueError:
        print(f"File {filename} has {lines} lines, {words} words and {chars} "

wc('*') | lines
# <<< File has 6 lines, 15 words and 159 characters
# ... File has 174 lines, 532 words and 4761 characters
# ... File has 10 lines, 28 words and 275 characters
# ... File has 12 lines, 34 words and 293 characters
# ... File total has 202 lines, 609 words and 5488 characters

In the second case, the command and the function will be executed in parallel and the command's stdout and stderr streams will be made available to the function.

import re
from pipepy import ping

def mean_ping(stdout):
    pings = []
    for line in stdout:
        match ='time=([\d\.]+) ms$', line.strip())
        if not match:
        time = float(match.groups()[0])
        if len(pings) % 10 == 0:
            print(f"Mean time is {sum(pings) / len(pings)} ms")

ping('-c', 30, "") | mean_ping
# >>> Mean time is 71.96000000000001 ms
# ... Mean time is 72.285 ms
# ... Mean time is 72.19666666666667 ms

If the command ends before the function, then next(stdout) will raise a StopIteration. If the function ends before the command, the command's stdin will be closed.

The return value of the pipe operation will be the return value of the function. The function can even include the word yield and thus return a generator that can be piped into another command.

Putting all of this together, we can do things like:

from pipepy import cat, grep

def my_input():
    yield "line one\n"
    yield "line two\n"
    yield "line two\n"
    yield "something else\n"
    yield "line three\n"

def my_output(stdout):
    for line in stdout:
        yield line.upper()

print(my_input() | cat | grep('line') | my_output | grep("TWO"))
# <<< LINE TWO
# ... LINE TWO

4. Right operand is a generator

This is one of the more exotic forms of piping. Here we take advantage of Python's passing values into a generator functionality. The original generator must send and receive data with the a = (yield b) syntax. The result of the pipe operation will be another generator that will yield whatever the original generator yields while, in the original generator, the return value of each yield command will be the next non-empty line of the PipePy instance:

from pipepy import echo

def upperize():
    line = yield
    while True:
        line = (yield line.upper())

# Remember, `upperize` is a function, `upperize()` is a generator
list(echo("aaa\nbbb") | upperize())
# <<< ["AAA\n", "BBB\n"]

And, since the return value of the pipe operation is a generator, it can be piped into another command:

print(echo("aaa\nbbb") | upperize() | grep("AAA"))
# <<< AAA

Interacting with background processes

There are 3 ways to interact with a background process: read-only, write-only and read/write. We have already covered read-only and write-only:

1. Incrementally sending data to a command

This is done by piping from an iterable to a command. The command actually runs in in parallel with the iterable and the iterable's data is fed to the command as it becomes available. We will slightly modify the previous example to better demonstrate this:

import random
import time
from pipepy import grep

def my_stdin():
    start = time.time()
    for _ in range(500):
        yield f"{time.time() - start} {random.randint(1, 100)}\n"

command = my_stdin() | grep('-E', r'\b17$', _stream_stdout=True)
# <<< 0.3154888153076172 17
# ... 1.5810892581939697 17
# ... 1.7773401737213135 17
# ... 2.8303775787353516 17
# ... 3.4419643878936768 17
# ... 4.511774301528931  17

Here, grep is actually run in in parallel with the generator and matches are printed as they are found since the command's output is being streamed to the console, courtesy of the _stream_stdout argument (more on this below).

2. Incrementally reading data from a command

This can be done either by piping the output of a command to a function with a subset of stdin, stdout and stderr as its arguments, or a generator, as we demonstrated before, or by iterating over a command's output:

import time
from pipepy import ping

start = time.time()
for line in ping('-c', 3, ''):
    print(time.time() - start, line.strip().upper())
# <<< 0.15728354454040527 PING GOOGLE.COM ( 56(84) BYTES OF DATA.
# ... 0.1574106216430664  64 BYTES FROM SOF02S32-IN-F14.1E100.NET ( ICMP_SEQ=1 TTL=103 TIME=71.8 MS
# ... 1.1319730281829834  64 BYTES FROM ( ICMP_SEQ=2 TTL=103 TIME=75.3 MS
# ... 2.1297826766967773  64 BYTES FROM ( ICMP_SEQ=3 TTL=103 TIME=73.4 MS
# ... 2.129857063293457
# ... 2.129875659942627   --- GOOGLE.COM PING STATISTICS ---
# ... 2.1298911571502686  3 PACKETS TRANSMITTED, 3 RECEIVED, 0% PACKET LOSS, TIME 2004MS
# ... 2.129910707473755   RTT MIN/AVG/MAX/MDEV = 71.827/73.507/75.253/1.399 MS

Again, the ping command is actually run in parallel with the body of the for-loop and each line is given to the body of the for-loop as it becomes available.

3. Reading data from and writing data to a command

Lets assume we have a command that makes the user take a math quiz. A normal interaction with this command would look like this:

→ math_quiz
3 + 4 ?
→ 7
8 + 2 ?
→ 12
→ Ctrl-d

Using python to interact with this command in a read/write fashion can be done with a with statement:

from pipepy import math_quiz

result = []
with math_quiz as (stdin, stdout, stderr):
    stdout = (line.strip() for line in stdout if line.strip())
        for _ in range(3)
            question = next(stdout)
            a, _, b, _ = question.split()
            answer = str(int(a) + int(b))
            stdin.write(answer + "\n")
            verdict = next(stdout)
            result.append((question, answer, verdict))
    except StopIteration:

# <<< [('10 + 7 ?', '17', 'Correct!'),
# ...  ('5 + 5 ?', '10', 'Correct!'),
# ...  ('5 + 5 ?', '10', 'Correct!')]

stdin, stdout and stderr are the open file streams of the background process. When the body of the with block finishes, an EOF is sent to the process and it is waited for.

You need to remember to end lines fed to stdin with a newline character if the command expects it. Also, don't forget to call stdin.flush() every now and then.

You can call with on a pipe expression that involves PipePy objects. In that case, each PipePy object's stdout will be connected to the next one's stdin, the stdin offered to the body of the with block will be the stdin of the leftmost command and the stdout and stderr offered to the body of the with block will be the stdout and stderr of the rightmost command:

from pipepy import cat, grep

command = cat | grep("foo") | cat | cat | cat  # We might as well keep going
with command as (stdin, stdout, stderr):
    assert next(stdout).strip() == "foo1"
    assert next(stdout).strip() == "foo3"

Altering the behavior of commands

Binary mode

All commands are executed in text mode, which means that they deal with str objects. This can cause problems. For example:

from pipepy import gzip
result = "hello world" | gzip
# <<< Traceback (most recent call last):
# ... ...
# ... UnicodeDecodeError: 'utf-8' codec can't decode byte 0x8b in position 1: invalid start byte

gzip cannot work in text mode because its output is binary data that cannot be utf-8-decoded. When text mode is not desirable, a command can be converted to binary mode setting its _text parameter to False:

from pipepy import gzip
gzip = gzip(_text=False)
result = "hello world" | gzip
# <<< b'\x1f\x8b\x08\x00\x00\x00\x00\x00\x00\x03\xcbH\xcd\xc9\xc9W(\xcf/\xcaI\xe1\x02\x00-;\x08\xaf\x0c\x00\x00\x00'

Input and output will be converted from/to binary by using the 'UTF-8' encoding. In the previous example, our input's type was str and was utf-8-encoded before being fed into gzip. You can change the encoding with the _encoding keyword argument:

from pipepy import gzip
gzip = gzip(_text=False)
result = "καλημέρα" | gzip
# <<< b'\x1f\x8b\x08\x00\x00\x00\x00\x00\x00\x03\x01\x10\x00\xef\xff\xce\xba\xce\xb1\xce\xbb\xce\xb7\xce\xbc\xce\xad\xcf\x81\xce\xb1"\x15g\xab\x10\x00\x00\x00'
result = "καλημέρα" | gzip(_encoding="iso-8859-7")
# <<< b'\x1f\x8b\x08\x00\x00\x00\x00\x00\x00\x03{\xf5\xf0\xf5\xf37w?>\x04\x00\x1c\xe1\xc0\xf7\x08\x00\x00\x00'

Streaming to console

During invocation, you can set the _stream_stdout and _stream_stderr keyword arguments to True. This means that the respective stream will not be captured by the result, but streamed to the console. This allows the user to interact with interactive commands. Consider the following 2 examples:

  1. fzf works like this:

    1. It gathers a list of choices from its stdin
    2. It displays the choices on stderr, constantly refreshing it depending on what the user inputs
    3. It starts directly capturing keystrokes on the keyboard, bypassing stdin, to allow the user to make their choice.
    4. When the user presses Enter, it prints the choice to its stdout

    Taking all this into account, we can do the following:

    from pipepy import fzf
    fzf = fzf(_stream_stderr=True)
    # This will open an fzf session to let us choose between "John" and "Mary"
    print("John\nMary" | fzf)
    # <<< Mary
  2. dialog works similar to fzf, but swaps stdout with stderr:

    1. It gathers a list of choices from its arguments
    2. It displays the choices on stdout, constantly refreshing it depending on what the user inputs
    3. It starts directly capturing keystrokes on the keyboard, bypassing stdin, to allow the user to make their choice.
    4. When the user presses Enter, it prints the choice to its stderr

    Taking all this into account, we can do the following:

    from pipepy import dialog
    dialog = dialog(_stream_stdout=True)
    # This will open a dialog session to let us choose between "John" and "Mary"
    result = dialog(checklist=True)('Choose name', 30, 110, 0,
                                    "John", '', "on",
                                    "Mary", '', "off")
    # <<< John

Also, during a script, you may not be interested in capturing the output of a command but may want to stream it to the console to show the command's output to the user. You can force a command sto stream its whole output by setting the _stream parameter:

from pipepy import wget

wget('https://...', _stream=True)()

While stdout and stderr will not be captured, returncode will and thus you can still use the command in boolean expressions:

from pipepy import wget

if wget('https://...', _stream=True):
     print("Download succeeded")
     print("Download failed")

You can call pipepy.set_always_stream(True) to make streaming to the console the default behavior. This may be desirable in some situations, like Makefiles (see below).

import pipepy
from pipepy import ls
ls()  # Alsost equivalent to `ls(_stream=True)()`

Similarly to how setting _stream=True forces a command to stream its output to the console, setting _stream=False forces it to capture its output even if set_always_stream has been called:

import pipepy
from pipepy import ls

ls()                 # Will stream its output
ls(_stream=False)()  # Will capture its output


You can call .raise_for_returncode() on an evaluated result to raise an exception if its returncode is not 0 (think of requests's .raise_for_status()):

from pipepy import ping, PipePyError
result = ping("asdf")()  # Remember, we have to evaluate it first

# <<< PipePyError: (2, '', 'ping: asdf: Name or service not known\n')

except PipePyError as exc:
    # <<< 2
    # <<< ""
    # <<< ping: asdf: Name or service not known

You can call pipepy.set_always_raise(True) to have all commands raise an exception if their returncode is not zero.

import pipepy
from pipepy import ping
# <<< PipePyError: (2, '', 'ping: asdf: Name or service not known\n')

If "always raise" is set, you can still force a command to suppress its exception by setting _raise=False:

import pipepy
from pipepy import ping
    ping("asdf")()  # Will raise an exception
except Exception as exc:
# <<< PipePyError: (2, '', 'ping: asdf: Name or service not known\n')

    ping("asdf", _raise=False)()  # Will not raise an exception
except Exception as exc:

"Interactive" mode

When "interactive" mode is set, the __repr__ method will simply return self.stdout + self.stderr. This enables some very basic functionality for the interactive python shell. To set interactive mode, run pipepy.set_interactive(True):

import pipepy
from pipepy import ls, overload_chars
# <<<
# ...
# ...
# ...

ls -l
# <<< total 20
# ... -rw-r--r-- 1 kbairak kbairak  159 Feb  7 22:05
# ... -rw-r--r-- 1 kbairak kbairak  275 Feb  7 22:04
# ... -rw-r--r-- 1 kbairak kbairak  293 Feb  7 22:04
# ... -rw-r--r-- 1 kbairak kbairak 4761 Feb  8 20:42

Making alterations "permanent"

Since PipePy objects treat their list of arguments as list of strings simply passed onto the subprocess.Popen function, and since there is no special significance to the first argument even though it is technically the command being executed, you can crete PipePy instances with the alterations we discussed and use them as templates for commands that will inherit these alterations:

stream_sh = PipePy(_stream=True)
# <<< PipePy()
# <<< True
# <<< PipePy('ls')
# <<< True

r =
# <<<     setup.cfg  tags
# ... htmlcov       pyproject.toml   test_requirements.txt
# ... LICENSE       src

# <<< None

# <<< 0
raise_sh = PipePy(_raise=True)
# <<< PipePy()
# <<< PipePy('false')
# <<< Traceback (most recent call last):
# ... ...
# ... pipepy.exceptions.PipePyError: (1, '', '')

This can work as a more contained alternative to set_always_stream and set_always_raise.


.terminate(), .kill() and .send_signal() simply forward the method call to the underlying Popen object.

Here are some utilities implemented within pipepy that don't make use of shell subprocesses, but we believe are useful for scripting.


In its simplest form, is an alias to os.chdir:

from pipepy import cd, pwd

# <<< /foo

# <<< /foo/bar

# <<< /foo

But it can also be used as a context processor for temporary directory changes:

# <<< /foo

with cd("bar"):
# <<< /foo/bar

# <<< /foo


In its simplest form, pipepy.export is an alias to os.environ.update:

import os
from pipepy import export

# <<< /home/foo

# <<< /home/foo/bar

But it can also be used as a context processor for temporary environment changes:

# <<< /home/foo

with export(PATH="/home/foo/bar"):
# <<< /home/foo/bar

# <<< /home/foo

If an environment variable is further modified within the body of the with block, it is not reverted upon exit:

with export(PATH="/home/foo/bar"):

# <<< /home/foo/BAR


The source function runs a bash script, extracts the resulting environment variables that have been set in the script and saves them on the current environment. Similarly to export, it can be used as a context processor (in fact, it uses export internally):

# env
export AAA=aaa
import os
from pipepy import source

with source('env'):
# <<< aaa
'AAA' in os.environ
# <<< False

# <<< aaa

The following keyword-only arguments are available to source:

  • recursive (boolean, defaults to False): If set, all files with the same name in the current directory and all its parents will be sourced, in reverse order. This allows nesting of environment variables:

    - /
      + - home/
          - kbairak/
            + - env:
            |     export COMPOSE_PROJECT_NAME="pipepy"
            + - project/
                + - env:
                      export COMPOSE_FILE="docker-compose.yml:docker-compose-dev.yml"
    from pipepy import cd, source, docker_compose
    source('env', recursive=True)
    # Now I have both `COMPOSE_PROJECT_NAME` and `COMPOSE_FILE`

    The files /home/kbairak/env and /home/kbairak/project/env were sourced, in that order.

  • quiet (boolean, defaults to True): If the sourced file fails, source will usually skip its sourcing without complaint and move on to the next one (if recursive is set). With quiet=False, an exception will be raised and the environment will not be updated.

  • shell (string, defaults to 'bash'): The shell command used to perform the sourcing.


Bundled with this library there is a command called pymake which aims to replicate the syntax and behavior of GNU make as much as possible, but in Python. A file looks like this (this is actually part of the Makefile of the current library):

import pipepy
from pipepy import python, rm


def clean():
    rm('-rf', "build", "dist")()

def build(clean):
    python('-m', "build")()

def publish(build):
    python('-m', "twine").upload("dist/*")()

You can now run pymake publish to run the publish make target, along with its dependencies. The names of the functions' arguments are used to define the dependencies, so clean is a dependency of build and build is a dependency of publish.

(You don't have to use pipepy commands inside, but admittedly it's a very good fit)

The arguments hold any return values of the dependency targets:

def a():
    return 1

def b():
    return 2

def c(a, b):
    print(a + b)
→ pymake c
# ← 3

Each dependency will be executed at most once, even if it's used as a dependency more than once:

def a():
    print("pymake target a")

def b(a):
    print("pymake target b")

def c(a, b):
    print("pymake target c")
→ pymake c
# ← pymake target a
# ← pymake target b
# ← pymake target c

You can set the DEFAULT_PYMAKE_TARGET global variable to define the default target.

from pipepy import pytest


def test():

pymake variables

Apart from dependencies, you can use function arguments to define variables that can be overridden by the invocation of pymake. This can be done in 2 ways:

  1. Using the function's keyword arguments:

    def greeting(msg="world"):
        print(f"hello {msg}")
    → pymake greeting
    # ← hello world
    → pymake greeting msg=Bill
    # ← hello Bill
  2. Using global variables defined in

    msg = "world"
    def greeting():
        print(f"hello {msg}")
    → pymake greeting
    # ← hello world
    → pymake greeting msg=Bill
    # ← hello Bill

Shell completion for pymake

pymake supports shell completion for bash and zsh.

In bash, run:

eval $(pymake --setup-bash-completion)

Then you will be able to see things like (example taken from pipepy's Makefile):

[kbairak@kbairakdelllaptop pipepy]$ pymake <TAB><TAB>
build      clean      debugtest  publish    watchtest
checks     covtest    html       test

In zsh, run:

eval $(pymake --setup-zsh-completion)

Then you will be able to see things like (example taken from pipepy's Makefile):

(pipepy) ➜  pipepy git:(master) ✗ pymake <TAB>
build      -- Build package
checks     -- Run static checks on the code (flake8, isort)
clean      -- Clean up build directories
covtest    -- Run tests and produce coverge report
debugtest  -- Run tests without capturing their output. This makes using an interactive debugger possible
html       -- Run tests and open coverage report in browser
publish    -- Publish package to PyPI
test       -- Run tests
watchtest  -- Automatically run tests when a source file changes

The descriptions are taken from the pymake targets' docstrings.

You can put the eval statements in your .bashrc/.zshrc.


  • Timeout for wait

  • Redirect input/output from/to file-like objects

  • Stream and capture at the same time (wrapper class for file-like object?)

  • with blocks where PipePy invocations forward to the context's stdin, eg:

    from pipepy import ssh
    with ssh("some-host") as host:
        r =  # Will actually send 'ls\n' to ssh's stdin


A Python library for invoking and interacting with shell commands.