subprocess-shell 1.1.0

A shell for subprocess

subprocess_shell

is a Python package providing an alternative interface to sub processes. The aim is simplicity comparable to shell scripting and transparency for more complex use cases.

[[TOC]]

Update: the showcase presents an earlier version which didn't provide run() and wait(logs=...)

Features

• Simple
  • e.g. 5 functions (start, write, wait, read, run) and 3 operators (>>, +, -)
• Transparent
  • usability layer for subprocess except streams
• Separates streams
  • no interleaving of stdout and stderr and from different processes of a chain
• Avoids deadlocks due to OS pipe buffer limits by using queues
• Uses Rich if available
• Supports Windows[^r4]

[^r4]: Insofar as tests succeed most of the time. On my system, tests freeze up sometimes for no apparent reason. If you experience the same and can reproduce it consistently, please open an issue!

images/rich_output.png

Examples

	bash -e	subprocess_shell	subprocess	Plumbum[^r1]
initialization		from subprocess_shell import *	import subprocess	from plumbum import local
run command	echo this	["echo", "this"] >> run()	assert subprocess.Popen(["echo", "this"]).wait() == 0	local["echo"]["this"].run_fg()
redirect stream	echo this > /path/to/file	["echo", "this"] >> run(start(stdout="/path/to/file"))	with open("/path/to/file", "wb") as stdout: assert subprocess.Popen(["echo", "this"], stdout=stdout).wait() == 0	(local["echo"]["this"] > "/path/to/file").run_fg()
read stream	a=$(echo this)	a = ["echo", "this"] >> run()	process = subprocess.Popen(["echo", "this"], stdout=subprocess.PIPE) a, _ = process.communicate() assert process.wait() == 0	a = local["echo"]("this")
write stream	cat - <<EOF this EOF	["cat", "-"] >> run(write("this"))	process = subprocess.Popen(["cat", "-"], stdin=subprocess.PIPE) process.communicate(b"this") assert process.wait() == 0	(local["cat"]["-"] << "this").run_fg()
chain commands	echo this | cat -	["echo", "this"] >> start() + ["cat", "-"] >> run()	process = subprocess.Popen(["echo", "this"], stdout=subprocess.PIPE) assert subprocess.Popen(["cat", "-"], stdin=process.stdout).wait() == 0 assert process.wait() == 0	(local["echo"]["this"] | local["cat"]["-"]).run_fg()
branch out	?	import sys _v_ = "import sys; print('stdout'); print('stderr', file=sys.stderr)" arguments = [sys.executable, "-c", _v_] process = arguments >> start(pass_stdout=True, pass_stderr=True) process + ["cat", "-"] >> run() process - ["cat", "-"] >> run()	import sys _v_ = "import sys; print('stdout'); print('stderr', file=sys.stderr)" arguments = [sys.executable, "-c", _v_] process = subprocess.Popen(arguments, stdout=subprocess.PIPE, stderr=subprocess.PIPE) assert subprocess.Popen(["cat", "-"], stdin=process.stdout).wait() == 0 assert subprocess.Popen(["cat", "-"], stdin=process.stderr).wait() == 0 assert process.wait() == 0	not supported[^r2]
errors in chains	?	_v_ = ["echo", "this"] >> start(return_codes=(0, 1)) - ["cat", "-"] _v_ >> run(wait(return_codes=(0, 2)))	first_process = subprocess.Popen(["echo", "this"], stderr=subprocess.PIPE) second_process = subprocess.Popen(["cat", "-"], stdin=first_process.stderr) assert first_process.wait() in (0, 1) and second_process.wait() in (0, 2)	not supported[^r2]
callbacks		["echo", "this"] >> run(start(stdout=print))	process = subprocess.Popen(["echo", "this"], stdout=subprocess.PIPE) for bytes in process.stdout: print(bytes) assert process.wait() == 0 !![^r3]

[^r1]: Mostly adapted versions from https://www.reddit.com/r/Python/comments/16byt8j/comment/jzhh21f/?utm_source=share&utm_medium=web2x&context=3 [^r2]: Has been requested years ago [^r3]: This is very limited and has several issues with potential for deadlocks. An exact equivalent would be too long for this table.

Notes

• bash -e because errors can have serious consequences
  • e.g.

a=$(failing command)
sudo chown -R root:root "$a/"

• assert process.wait() == 0 is the shortest (readable) code waiting for a process to stop and asserting the return code
• complexity of code for Plumbum can be misleading because it has a much wider scope (e.g. remote execution and files)

Quickstart

• Prepare virtual environment (optional but recommended)

  • e.g. Pipenv: python -m pip install -U pipenv

• Install subprocess_shell

  • e.g. python -m pipenv run pip install subprocess_shell

• Import and use it

  • e.g. from subprocess_shell import * and python -m pipenv run python ...

• Prepare tests

  • e.g. python -m pipenv run pip install subprocess_shell[test]

• Run tests

  • e.g. python -m pipenv run pytest ./tests

Documentation

from subprocess_shell import *

Start process

process = arguments >> start(
    stdin=subprocess.PIPE,
    stdout=subprocess.PIPE,
    pass_stdout=False,
    stderr=subprocess.PIPE,
    pass_stderr=False,
    queue_size=0,
    logs=None,
    return_codes=(0,),
    force_color=True,
    async_=False,
    **{},
)

arguments	iterable	arguments are converted to string using str(...) and passed to subprocess.Popen(...)
stdin	subprocess.PIPE	provide stdin
	any object	same as subprocess.Popen(..., stdin=object)
stdout	subprocess.PIPE	provide stdout
	string or pathlib.Path	redirect stdout to file
	function(chunk: bytes | str) -> typing.Any	call function for each chunk from stdout
	any object	same as subprocess.Popen(..., stdout=object)
pass_stdout	False	if stdout=subprocess.PIPE: queue chunks from stdout
	True	don't use stdout
stderr	subprocess.PIPE	provide stderr
	string or pathlib.Path	redirect stderr to file
	function(chunk: bytes | str) -> typing.Any	call function for each chunk from stderr
	any object	same as subprocess.Popen(..., stderr=object)
pass_stderr	False	if stderr=subprocess.PIPE: queue chunks from stderr
	True	don't use stderr
queue_size	0	no limit on size of queues
	int > 0	wait for other threads to process queues if full; !! can lead to deadlocks !!
logs	None	if in a chain: analog of wait(logs=None)
	boolean	if in a chain: analog of wait(logs=False) or wait(logs=True)
return_codes	(0,)	if in a chain: analog of wait(return_codes=(0,))
	collection object or None	if in a chain: analog of wait(return_codes=object) or wait(return_codes=None)
force_color	False	don't touch environment variable FORCE_COLOR
	True	if environment variable FORCE_COLOR is not set: set to 1
async_	False	if Windows: use asyncio; else: use selectors
	True	use asyncio
**	{}	passed to subprocess.Popen(...)

Write to stdin

process = process >> write(object, close=False, encoding=None)

object	string or bytes	en/decoded if necessary, written to stdin and flushed
close	False	keep stdin open
	True	close stdin after flush
encoding	None	use the default encoding (UTF-8) for encoding the string
	str	use a different encoding

requires start(stdin=subprocess.PIPE)

Wait for process

return_code = process >> wait(
    stdout=True,
    stderr=True,
    logs=None,
    return_codes=(0,),
    rich=True,
    stdout_style="green",
    log_style="dark_orange3",
    error_style="red",
    ascii=False,
    encoding=None,
)

stdout	True	if stdout is queued: collect stdout, format and print to stdout
	False	don't use stdout
	any object	if stdout is queued: collect stdout, format and print with print(..., file=object)
stderr	True	if stderr is queued: collect stderr, format and print to stderr
	False	don't use stderr
	any object	if stderr is queued: collect stderr, format and print with print(..., file=object)
logs	None	write stdout first and use log_style for stderr if the return code assert succeeds or error_style otherwise
	False	write stdout first and use error_style for stderr
	True	write stderr first and use log_style
return_codes	(0,)	assert that the return code is 0
	collection object	assert that the return code is in object
	None	don't assert the return code
rich	True	use Rich if available
	False	don't use Rich
stdout_style	"green"	use color "green" for stdout frame
	style object or string	use style for stdout frame, see Styles
log_style	"dark_orange3"	use color "dark_orange3" for stderr frame, see argument logs
	style object or string	use style for stderr frame, see argument logs and Styles
error_style	"red"	use color "red" for stderr frame, see argument logs
	style object or string	use style for stderr frame, see argument logs and Styles
ascii	False	use Unicode
	True	use ASCII
encoding	None	use the default encoding (UTF-8) for encoding strings or decoding bytes objects
	str	use a different encoding

Read from stdout/stderr

string = process >> read(
    stdout=True,
    stderr=False,
    bytes=False,
    encoding=None,
    logs=_DEFAULT,
    return_codes=_DEFAULT,
    wait=None,
)
# optionally one of
.cast_str     # shortcut for `typing.cast(str, ...)`
.cast_bytes   #          for `typing.cast(bytes, ...)`
.cast_strs    #          for `typing.cast(tuple[str, str], ...)`
.cast_bytess  #          for `typing.cast(tuple[bytes, bytes], ...)`

stdout	True	execute process >> wait(..., stdout=False), collect stdout, join and return; requires queued stdout
	False	execute process >> wait(..., stdout=True)
	any object	execute process >> wait(..., stdout=object)
stderr	False	execute process >> wait(..., stderr=True)
	True	execute process >> wait(..., stderr=False), collect stderr, join and return; requires queued stderr
	any object	execute process >> wait(..., stderr=object)
bytes	False	return a string or tuple of strings
	True	return bytes or tuple of bytes
encoding	None	use the default encoding (UTF-8) for encoding strings or decoding bytes objects
	str	use a different encoding
logs	_DEFAULT	use logs from argument wait
	bool or None	replace logs from argument wait
return_codes	_DEFAULT	use return_codes from argument wait
	any object	replace return_codes from argument wait
wait	None	use wait() for waiting
	wait(...)	use a different wait object

process.get_stdout_lines(bytes=False, encoding=None)  # generator[str | bytes]
process.get_stderr_lines(bytes=False, encoding=None)  # generator[str | bytes]
process.join_stdout_strings(encoding=None)  # str
process.join_stderr_strings(encoding=None)  # str
process.get_stdout_strings(encoding=None)  # generator[str]
process.get_stderr_strings(encoding=None)  # generator[str]
process.join_stdout_bytes(encoding=None)  # bytes
process.join_stderr_bytes(encoding=None)  # bytes
process.get_stdout_bytes(encoding=None)  # generator[bytes]
process.get_stderr_bytes(encoding=None)  # generator[bytes]
process.get_stdout_objects()  # generator[str | bytes]
process.get_stderr_objects()  # generator[str | bytes]

bytes	False	return iterable of strings
	True	return iterable of bytes
encoding	None	use the default encoding (UTF-8) for encoding strings or decoding bytes objects
	str	use a different encoding

requires queued stdout/stderr

Chain processes / pass streams

process = source_arguments >> start(...) + arguments >> start(...)
# or
source_process = source_arguments >> start(..., pass_stdout=True)
process = source_process + arguments >> start(...)

process = source_arguments >> start(...) - arguments >> start(...)
# or
source_process = source_arguments >> start(..., pass_stderr=True)
process = source_process - arguments >> start(...)

source_process = process.get_source_process()

• process >> wait(...) waits for the processes from left/source to right/target

Shortcut

object = object >> run(*args)
# optionally one of
.cast_int     # shortcut for `typing.cast(int, ...)`
.cast_str     #          for `typing.cast(str, ...)`
.cast_bytes   #          for `typing.cast(bytes, ...)`
.cast_strs    #          for `typing.cast(tuple[str, str], ...)`
.cast_bytess  #          for `typing.cast(tuple[bytes, bytes], ...)`

*args	none	short for >> start() >> wait()
	sequence of objects returned by start(...), write(...), wait(...) and read(...)	short for >> start(...) {* >> write(...) *} >> {wait(...) | read(...)}

Other

LineStream

If you want to use wait and process the streams line by line at the same time, you can use LineStream.

Example:

import subprocess_shell
import sys

def function(line_string):
    pass

process >> wait(stdout=subprocess_shell.LineStream(function, sys.stdout))

Motivation

Shell scripting is great for simple tasks. When tasks become more complex, e.g. hard to chain or require non-trivial processing, I always switch to Python. The interface provided by subprocess is rather verbose and parts that would look trivial in a shell script end up a repetitive mess. After refactoring up the mess once too often, it was time for a change.

See also

• Plumbum
• sh

Why the name subprocess_shell

Simply because I like the picture of subprocess with a sturdy layer that is easy and safe to handle. Also, while writing import subprocess it is easy to remember to add _shell.

Before subprocess_shell I chose to call it shell. This was a bad name for several reasons, but most notably because the term shell is commonly used for applications providing an interface to the operating system.

---