Skip to content

asyncio library that provides an API for running subprocesses

License

Notifications You must be signed in to change notification settings

byllyfish/shellous

Repository files navigation

Async Processes and Pipelines

PyPI docs CI codecov Downloads

shellous provides a concise API for running subprocesses using asyncio. It is similar to and inspired by sh.

import asyncio
from shellous import sh

async def main():
    result = await sh("echo", "hello")
    print(result)

asyncio.run(main())

Benefits

  • Run programs asynchronously in a single line.
  • Redirect stdin, stdout and stderr to files, memory buffers, async streams or loggers.
  • Iterate asynchronously over subprocess output.
  • Set timeouts and reliably cancel running processes.
  • Run a program with a pseudo-terminal (pty).
  • Use send() and expect() to manually control a subprocess.
  • Construct pipelines and use process substitution directly from Python (no shell required).
  • Runs on Linux, MacOS, FreeBSD and Windows.
  • Monitor processes being started and stopped with audit_callback API.

Requirements

  • Requires Python 3.9 or later.
  • Requires an asyncio event loop.
  • Pseudo-terminals require a Unix system.
  • Process substitution requires a Unix system with /dev/fd support.

Running a Command

The tutorial in this README uses the asyncio REPL built into Python. In these examples, >>> is the REPL prompt.

Start the asyncio REPL by typing python3 -m asyncio, and import sh from the shellous module:

>>> from shellous import sh

Here's a command that runs echo "hello, world".

>>> await sh("echo", "hello, world")
'hello, world\n'

The first argument to sh is the program name. It is followed by zero or more arguments. Each argument will be converted to a string. If an argument is a list or tuple, it is flattened recursively.

>>> await sh("echo", 1, 2, [3, 4, (5, 6)])
'1 2 3 4 5 6\n'

A command does not run until you await it. When you run a command using await, it returns the value of the standard output interpreted as a UTF-8 string. It is safe to await the same command object more than once.

Here, we create our own echo command with "-n" to omit the newline. Note, echo("abc") will run the same command as echo -n "abc".

>>> echo = sh("echo", "-n")
>>> await echo("abc")
'abc'

Commands are immutable objects that represent a program invocation: program name, arguments, environment variables, redirection operators and other settings. When you use a method to modify a Command, you are returning a new Command object. The original object is unchanged.

You can wrap your commands in a function to improve type safety:

>>> from shellous import Command
>>> def exclaim(word: str) -> Command[str]:
...   return sh("echo", "-n", f"{word}!!")
... 
>>> await exclaim("Oh")
'Oh!!'

The type hint Command[str] indicates that the command returns a str.

Arguments

Commands use positional arguments only; keyword arguments are not supported.

In most cases, shellous automatically converts Python objects passed as command arguments to str or bytes. As described above, the list and tuple types are an exception; they are recursively flattened before their elements are converted to strings.

Dicts, sets, and generator types are not supported as arguments. Their string format doesn't make sense as a command line argument.

Results

When a command completes successfully, it returns the standard output (or "" if stdout is redirected). For a more detailed response, you can specify that the command should return a Result object by using the .result modifier:

>>> await echo.result("abc")
Result(exit_code=0, output_bytes=b'abc', error_bytes=b'', cancelled=False, encoding='utf-8')

A Result object contains the command's exit_code in addition to its output. A Result is True if the command's exit_code is zero. You can access the string value of the output using the .output property:

if result := await sh.result("cat", "some-file"):
    output = result.output
else:
    print(f"Command failed with exit_code={result.exit_code})

You can retrieve the string value of the standard error using the .error property. (By default, only the first 1024 bytes of standard error is stored.)

If a command was terminated by a signal, the exit_code will be the negative signal number.

The return value of sh.result("cmd", ...) uses the type hint Command[Result].

ResultError

If you are not using the .result modifier and a command fails, it raises a ResultError exception:

>>> await sh("cat", "does_not_exist")
Traceback (most recent call last):
  ...
shellous.result.ResultError: Result(exit_code=1, output_bytes=b'', error_bytes=b'cat: does_not_exist: No such file or directory\n', cancelled=False, encoding='utf-8')

The ResultError exception contains a Result object with the exit_code and the first 1024 bytes of standard error.

In some cases, you want to ignore certain exit code values. That is, you want to treat them as if they are normal. To do this, you can set the exit_codes option:

>>> await sh("cat", "does_not_exist").set(exit_codes={0,1})
''

If there is a problem launching a process, shellous can also raise a separate FileNotFoundError or PermissionError exception.

Async For

Using await to run a command collects the entire output of the command in memory before returning it. You can also iterate over the output lines as they arrive using async for.

>>> [line async for line in echo("hi\n", "there")]
['hi\n', ' there']

Use an async for loop when you want to examine the stream of output from a command, line by line. For example, suppose you want to run tail on a log file.

async for line in sh("tail", "-f", "/var/log/syslog"):
    if "ERROR" in line:
        print(line.rstrip())

Async With

You can use a command as an asynchronous context manager. There are two ways to run a program using a context manager: a low-level API and a high-level API.

Byte-by-Byte (Low Level)

Use async with directly when you need byte-by-byte control over the individual streams: stdin, stdout and stderr. To control a standard stream, you must tell shellous to "capture" it (For more on this, see Redirection.)

cmd = sh("cat").stdin(sh.CAPTURE).stdout(sh.CAPTURE)
async with cmd as run:
    run.stdin.write(b"abc")
    run.stdin.close()
    print(await run.stdout.readline())

result = run.result()

The streams run.stdout and run.stderr are asyncio.StreamReader objects. The stream run.stdin is an asyncio.StreamWriter object. If we didn't specify that stdin/stdout are sh.CAPTURE, the streams run.stdin and run.stdout would be None.

The return value of run.result() is a Result object. Depending on the command settings, this function may raise a ResultError on a non-zero exit code.

⚠️ When reading or writing individual streams, you are responsible for managing reads and writes so they don't deadlock. You may use run.create_task to schedule a concurrent task.

You can also use async with to run a server. When you do so, you must tell the server to stop using run.cancel(). Otherwise, the context manager will wait forever for the process to exit.

async with sh("some-server") as run:
    # Send commands to the server here...
    # Manually signal the server to stop.
    run.cancel()

Prompt with Send/Expect (High Level API)

Use the prompt() method to control a process using send and expect. The prompt() method returns an asynchronous context manager (the Prompt class) that facilitates reading and writing strings and matching regular expressions.

cmd = sh("cat").set(pty=True)

async with cmd.prompt() as client:
  await client.send("abc")
  output, _ = await client.expect("\r\n")
  print(output)

The Prompt API automatically captures stdin and stdout.

Here is another example of controlling a bash co-process running in a docker container.

async def list_packages():
    "Run bash in an ubuntu docker container and list packages."
    bash_prompt = re.compile("root@[0-9a-f]+:/[^#]*# ")
    cmd = sh("docker", "run", "-it", "--rm", "-e", "TERM=dumb", "ubuntu")

    async with cmd.set(pty=True).prompt(bash_prompt, timeout=3) as cli:
        # Read up to first prompt.
        await cli.expect()

        # Disable echo. The `command()` method combines send *and* expect methods.
        await cli.command("stty -echo")

        # Return list of packages.
        result = await cli.command("apt-cache pkgnames")
        return result.strip().split("\r\n")

    # You can check the result object's exit code. You can only
    # access `cli.result` outside the `async with` block.
    assert cli.result.exit_code == 0

The prompt() API does not raise a ResultError when a command exits with an error status. Typically, you'll see an EOFError when you were expecting to read a response. You can check the exit status by retrieving the Prompt's result property outside of the async with block.

Redirection

shellous supports the redirection operators | and >>. They work similar to how they work in the unix shell. Shellous does not support use of < or > for redirection. Instead, replace these with |.

To redirect to or from a file, use a pathlib.Path object. Alternatively, you can redirect input/output to a StringIO object, an open file, a Logger, or use a special redirection constant like sh.DEVNULL.

⚠️ When combining the redirect operators with await, you must use parentheses; await has higher precedence than | and >>.

Redirecting Standard Input

To redirect standard input, use the pipe operator | with the argument on the left-side. Here is an example that passes the string "abc" as standard input.

>>> cmd = "abc" | sh("wc", "-c")
>>> await cmd
'       3\n'

To read input from a file, use a Path object from pathlib.

>>> from pathlib import Path
>>> cmd = Path("LICENSE") | sh("wc", "-l")
>>> await cmd
'     201\n'

Shellous supports different STDIN behavior when using different Python types.

Python Type Behavior as STDIN
str Read input from string object.
bytes, bytearray Read input from bytes object.
Path Read input from file specified by Path.
File, StringIO, ByteIO Read input from open file object.
int Read input from existing file descriptor.
asyncio.StreamReader Read input from StreamReader.
sh.DEVNULL Read input from /dev/null.
sh.INHERIT Read input from existing sys.stdin.
sh.CAPTURE You will write to stdin interactively.

Redirecting Standard Output

To redirect standard output, use the pipe operator | with the argument on the right-side. Here is an example that writes to a temporary file.

>>> output_file = Path("/tmp/output_file")
>>> cmd = sh("echo", "abc") | output_file
>>> await cmd
''
>>> output_file.read_bytes()
b'abc\n'

To redirect standard output with append, use the >> operator.

>>> cmd = sh("echo", "def") >> output_file
>>> await cmd
''
>>> output_file.read_bytes()
b'abc\ndef\n'

Shellous supports different STDOUT behavior when using different Python types.

Python Type Behavior as STDOUT/STDERR
Path Write output to the file path specified by Path.
bytearray Write output to a mutable byte array.
File, StringIO, ByteIO Write output to an open file object.
int Write output to existing file descriptor at its current position. ◆
logging.Logger Log each line of output. ◆
asyncio.StreamWriter Write output to StreamWriter. ◆
sh.CAPTURE Capture output for async with. ◆
sh.DEVNULL Write output to /dev/null. ◆
sh.INHERIT Write output to existing sys.stdout or sys.stderr. ◆

◆ For these types, there is no difference between using | and >>.

Shellous does not support redirecting standard output/error to a plain str or bytes object. If you intend to redirect output to a file, you must use a pathlib.Path object.

Redirecting Standard Error

By default, the first 1024 bytes read from standard error are stored in the Result object. Any further bytes are discarded. You can change the 1024 byte limit using the error_limit option.

To redirect standard error, use the stderr method. Standard error supports the same Python types as standard output. To append, set append=True in the stderr method.

To redirect stderr to the same place as stdout, use the sh.STDOUT constant. If you also redirect stdout to sh.DEVNULL, you will only receive the standard error.

>>> cmd = sh("cat", "does_not_exist").stderr(sh.STDOUT)
>>> await cmd.set(exit_codes={0,1})
'cat: does_not_exist: No such file or directory\n'

To redirect standard error to the hosting program's sys.stderr, use the sh.INHERIT redirect option.

>>> cmd = sh("cat", "does_not_exist").stderr(sh.INHERIT)
>>> await cmd
cat: does_not_exist: No such file or directory
Traceback (most recent call last):
  ...
shellous.result.ResultError: Result(exit_code=1, output_bytes=b'', error_bytes=b'', cancelled=False, encoding='utf-8')

If you redirect stderr, it will no longer be stored in the Result object, and the error_limit option will not apply.

Default Redirections

For regular commands, the default redirections are:

  • Standard input is read from the empty string ("").
  • Standard out is buffered and stored in the Result object (BUFFER).
  • First 1024 bytes of standard error is buffered and stored in the Result object (BUFFER).

However, the default redirections are adjusted when using a pseudo-terminal (pty):

  • Standard input is captured and ignored (CAPTURE).
  • Standard out is buffered and stored in the Result object (BUFFER).
  • Standard error is redirected to standard output (STDOUT).

When you use the Prompt API, the standard input and standard output are automatically redirected to CAPTURE.

Pipelines

You can create a pipeline by combining commands using the | operator. A pipeline feeds the standard out of one process into the next process as standard input. Here is the shellous equivalent to the bash command: ls | grep README

>>> pipe = sh("ls") | sh("grep", "README")
>>> await pipe
'README.md\n'

A pipeline returns a Result if the last command in the pipeline has the .result modifier. To set other options like encoding for a Pipeline, set them on the last command.

>>> pipe = sh("ls") | sh("grep", "README").result
>>> await pipe
Result(exit_code=0, output_bytes=b'README.md\n', error_bytes=b'', cancelled=False, encoding='utf-8')

Error reporting for a pipeline is implemented similar to using the -o pipefail shell option.

Pipelines support the same await/async for/async with operations that work on a single command, including the Prompt API.

>>> [line.strip() async for line in pipe]
['README.md']

Process Substitution (Unix Only)

You can pass a shell command as an argument to another. Here is the shellous equivalent to the bash command: grep README <(ls).

>>> cmd = sh("grep", "README", sh("ls"))
>>> await cmd
'README.md\n'

Use .writable to write to a command instead.

>>> buf = bytearray()
>>> cmd = sh("ls") | sh("tee", sh("grep", "README").writable | buf) | sh.DEVNULL
>>> await cmd
''
>>> buf
bytearray(b'README.md\n')

The above example is equivalent to ls | tee >(grep README > buf) > /dev/null.

Timeouts

You can specify a timeout using the timeout option. If the timeout expires, shellous will raise a TimeoutError.

>>> await sh("sleep", 60).set(timeout=0.1)
Traceback (most recent call last):
  ...
TimeoutError

Timeouts are just a special case of cancellation. When a command is cancelled, shellous terminates the running process and raises a CancelledError.

>>> t = asyncio.create_task(sh("sleep", 60).coro())
>>> t.cancel()
True
>>> await t
Traceback (most recent call last):
  ...
CancelledError

By default, shellous will send a SIGTERM signal to the process to tell it to exit. If the process does not exit within 3 seconds, shellous will send a SIGKILL signal. You can change these defaults with the cancel_signal and cancel_timeout settings. A command is not considered fully cancelled until the process exits.

Pseudo-Terminal Support (Unix Only)

To run a command through a pseudo-terminal, set the pty option to True.

>>> await sh("echo", "in a pty").set(pty=True)
'in a pty\r\n'

Alternatively, you can pass a pty function to configure the tty mode and size.

>>> ls = sh("ls").set(pty=shellous.cooked(cols=40, rows=10, echo=False))
>>> await ls("README.md", "CHANGELOG.md")
'CHANGELOG.md\tREADME.md\r\n'

Shellous provides three built-in helper functions: shellous.cooked(), shellous.raw() and shellous.cbreak().

Context Objects

You can store shared command settings in an immutable context object (CmdContext). To create a new context object, specify your changes to the default context sh:

>>> auditor = lambda phase, info: print(phase, info["runner"].name)
>>> sh_audit = sh.set(audit_callback=auditor)

Now all commands created with sh_audit will log their progress using the audit callback.

>>> await sh_audit("echo", "goodbye")
start echo
stop echo
'goodbye\n'

You can also create a context object that specifies all return values are Result objects.

>>> rsh = sh.result
>>> await rsh("echo", "whatever")
Result(exit_code=0, output_bytes=b'whatever\n', error_bytes=b'', cancelled=False, encoding='utf-8')

Options

Both Command and CmdContext support options to control their runtime behavior. Some of these options (timeout, pty, audit_callback, and exit_codes) have been described above. See the shellous.Options class for more information.

You can retrieve an option from cmd with cmd.options.<option>. For example, use cmd.options.encoding to obtain the encoding:

>>> cmd = sh("echo").set(encoding="latin1")
>>> cmd.options.encoding
'latin1'

Command and CmdContext use the .set() method to specify most options:

Option Description
path Search path to use instead of the PATH environment variable. (Default=None)
env Additional environment variables to pass to the command. (Default={})
inherit_env True if command should inherit the environment variables from the current process. (Default=True)
encoding Text encoding of input/output streams. You can specify an error handling scheme by including it after a space, e.g. "ascii backslashreplace". (Default="utf-8 strict")
exit_codes Set of exit codes that do not raise a ResultError. (Default={0})
timeout Timeout in seconds to wait before cancelling the process. (Default=None)
cancel_timeout Timeout in seconds to wait for a cancelled process to exit before forcefully terminating it. (Default=3s)
cancel_signal The signal sent to a process when it is cancelled. (Default=SIGTERM)
alt_name Alternate name for the process used for debug logging. (Default=None)
pass_fds Additional file descriptors to pass to the process. (Default={})
pass_fds_close True if descriptors in pass_fds should be closed after the child process is launched. (Default=False)
pty Used to allocate a pseudo-terminal (PTY). (Default=False)
close_fds True if process should close all file descriptors when it starts. This setting defaults to False to align with posix_spawn requirements. (Default=False)
audit_callback Provide function to audit stages of process execution. (Default=None)
coerce_arg Provide function to coerce Command arguments to strings when str() is not sufficient. For example, you can provide your own function that converts a dictionary argument to a sequence of strings. (Default=None)
error_limit Maximum number of initial bytes of STDERR to store in Result object. (Default=1024)

env

Use the env() method to add to the list of environment variables. The env() method supports keyword parameters. You can call env() more than once and the effect is additive.

>>> cmd = sh("echo").env(ENV1="a", ENV2="b").env(ENV2=3)
>>> cmd.options.env
{'ENV1': 'a', 'ENV2': '3'}

Use the env option with set() when you want to replace all the environment variables.

input, output, error

When you apply a redirection operator to a Command or CmdContext, the redirection targets are also stored in the Options object. To change these, use the .stdin(), .stdout(), or .stderr() methods or the redirection operator |.

Option Description
input The redirection target for standard input.
input_close True if standard input should be closed after the process is launched.
output The redirection target for standard output.
output_append True if standard output should be open for append.
output_close True if standard output should be closed after the process is launched.
error The redirection target for standard error.
error_append True if standard error should be open for append.
error_close True if standard error should be closed after the process is launched.

Type Checking

Shellous fully supports PEP 484 type hints.

Commands

Commands are generic on the return type, either str or Result. You will specify the type of a command object as Command[str] or Command[Result].

Use the result modifier to obtain a Command[Result] from a Command[str].

from shellous import sh, Command, Result

cmd1: Command[str] = sh("echo", "abc")
# When you `await cmd1`, the result is a `str` object.

cmd2: Command[Result] = sh.result("echo", "abc")
# When you `await cmd2`, the result is a `Result` object.

CmdContext

The CmdContext class is also generic on either str or Result.

from shellous import sh, CmdContext, Result

sh1: CmdContext[str] = sh.set(path="/bin:/usr/bin")
# When you use `sh1` to create commands, it produces `Command[str]` object with the given path.

sh2: CmdContext[Result] = sh.result.set(path="/bin:/usr/bin")
# When you use `sh2` to create commands, it produces `Command[Result]` objects with the given path.

Logging

For verbose logging, shellous supports a SHELLOUS_TRACE environment variable. Set the value of SHELLOUS_TRACE to a comma-delimited list of options:

  • detail: Enables detailed logging used to trace the steps of running a command.

  • prompt: Enables logging in the Prompt class when controlling a program using send/expect.

  • all: Enables all logging options.

Shellous uses the built-in Python logging module. After enabling these options, the shellous logger will display log messages at the INFO level.

Without these options enabled, Shellous generates almost no log messages.