A library which provides a slightly more convinient way to launch processes, compared to Python's subprocess module.
While Python’s subprocess library is great, it may not be the easiest library to use. This is the reason I created nicecall: it allows to do simple tasks with processes very easily.
Note that nicecall is not a substitute to subprocess, because much of subprocess functionality doesn’t exist. For instance, one can’t use stdin or pipes with nicecall. The goal is not to replace subprocess, but only to provide an easy way to do the most common tasks.
Note: make sure you also check the tests which give a few examples of how to use the library. Most pieces of code below are in tests/smoke/test_docs.py.
The library uses method chaining, which allows to add logic on the fly before actually launching the process. Methods such as on_stdout, ignore, etc. create a copy of the object, modify this copy, and return it to the caller. This makes it possible to reuse base objects in multiple locations in your code, reducing code duplication.
Let’s start by executing a task:
result = nicecall.Process().execute(["touch", "/tmp/hello"])
The result contains the exit code, which makes it possible to determine whether the process terminated successfully. Below, the value of result is expected to be zero. You may also make it fail:
result = nicecall.Process().execute(["touch", "/tmp/a/b/c/d"])
The result should now be 1, assuming you don’t have /tmp/a/b/c directory.
Sometimes, one may prefer raising an exception if the process fails, instead of checking for the exit code manually. The raise_if_error() method can be used for that. When called, exit code different from zero will lead to subprocess.CalledProcessError being raised.
One can also perform a bunch of actions on stdout and stderr. Let’s display stdout in terminal:
The output should be:
a b c
If you just want to read stdout output, then instead of execute, you can use read_stdout. The following piece of code produces the same output:
lines = nicecall.Process().read_stdout(["echo", "a\nb\nc"]) print("\n".join(lines))
Note that when using read_stdout, a non-zero exit code will lead to subprocess.CalledProcessError being raised, independently of the presence of raise_if_error() in the chain. The reason for this choice is that with read_stdout, there is no way to check for the actual exit code, and plainly ignoring it would lead to difficult to debug situations.
Also note that read_stdout stores stdout in a list and is not lazy. This means that:
- The method is appropriate only for commands which output a small quantity of lines.
- Looping through the lines and stopping in the middle wouldn’t affect neither the processing of the exit code, nor the other actions such as logging which may have been specified earlier in the chain through on_stdout.
- The lines from stderr are not returned. The on_stderr method can still be used, exactly in the same way it is with execute.
- All the actions defined through on_stdout will be performed before the list is returned.
- Filters added through keep and ignore apply to the lines in the result as well.
A common thing, at least in my case, is to log stdout or stderr to syslog. With nicecall, it’s easy:
# Initialize logging. log_handler = logging.handlers.SysLogHandler(address="/dev/log") formatter = logging.Formatter("demo: [%(levelname)s] %(message)s") log_handler.setFormatter(formatter) log_handler.setLevel(logging.DEBUG) demo_logger = logging.getLogger("demo") demo_logger.setLevel(logging.DEBUG) demo_logger.addHandler(log_handler) ... # Log stdout. logger = nicecall.StdoutLogger("test") nicecall.Process().on_stdout(logger.log).execute(["echo", "a\nb"])
Note that nicecall.StdoutLogger can be initialized with either the name of the logger, or the instance of the logger itself.
The library itself logs calls (INFO level) and call failures (WARNING level) through the logger named nicecall.process. For instance, executing touch /tmp/a/b/c/d will produce two log messages:
INFO:nicecall.process:Called “touch /tmp/a/b/c/d”. WARNING:nicecall.process:“touch /tmp/a/b/c/d” failed with exit code 1.
Sometimes, you don’t want to process specific content such as empty lines or whitespace. This is what filters are about:
nicecall \ .Process() \ .ignore(nicecall.filters.whitespace) \ .on_stdout(print) \ .execute(["echo", "a\n\nb"])
Here, a and b will be displayed in terminal; however, the empty line will be ignored. The reverse is called keep. Both keep and ignore accept any function which takes a string as a parameter and returns a boolean. For instance, this will print only stdout content longer than fifteen characters:
nicecall \ .Process() \ .keep(lambda line: len(line) > 15) \ .on_stdout(print) \ .execute(["echo", "Hello World!\nWhat a lovely day!"])
Multiple keep and ignore methods can be combined. The output will keep the lines which match all predicates from keep methods and none from ignore ones.
Filters apply to both stdout and stderr; there is no way to apply them to only one of the streams.
In order to be able to test your code, the library provides a NullProcess class, a stub and a mock.
This class creates an object which will not launch any process when execute is called. The purpose of this class is to replace the actual Process class during testing.
The stub makes it possible to emulate Process without actually doing the system calls. The difference with NullProcess is that the stub makes it possible to define the exit codes and stdout/stderr output for specific commands.
The stub allows to define associations between the arguments and the expected response. For instance, imagine a situation where the tested code is expected to perform two calls: one to create a directory, another one to create a file in it. We want to test how the code under testing will perform if the second command fails: are the developers handling this edge case? For this purpose, one can use the stub like this:
stub = nicecall.tests.ProcessStub() stub.add_match(["mkdir", "/tmp/a"], 0) stub.add_match( ["touch", "/tmp/a/b"], 1, stderr=["touch: cannot touch '/tmp/a/b': No such file or directory"])
The stub can now be passed to the code under tests instead of nicecall.Process(). The tested code will run, perform a mkdir, and, when executing the touch command, will get back the exit code 1 and a call to the actions, if any, set through on_stderr.
The mock performs in a similar way to a stub, but also records the activity of the code under tests, i.e. the parameters which were passed to different methods of the mock. Usually, the mock is used this way:
with nicecall.tests.ProcessMockContext() as context: # Code under tests goes here. # The mock is `context.mock`. ... # Follows the assertions. In this example, I'm just ensuring that the # code under tests added `print` to the `stdout` actions, i.e. ran the # `...on_stdout(print)...` command. actual = context.on_stdout_actions expected = [print] self.assertCountEqual(expected, actual)
The mock makes it possible to check the following elements:
- executed_args: the args which were used when calling execute() method.
- ignore_predicates: the list of predicates added by the tested code using the ignore method.
- keep_predicates: same as previous, but for keep.
- on_stdout_actions: the list of actions added by the tested code using the on_stdout method.
- on_stderr_actions: same as previous, but for on_stderr.
The class is the entry point of the library. It makes it possible to specify different options before actually starting the process.
execute: actually executes the process and blocks until the process finishes.
args is an array which indicates the process to start, and its parameters. Example: ["touch", "/tmp/hello"].
Returns the exit code.
keep: specifies a filter to apply to determine if the line of stdout or stderr should be processed by the actions specified through on_stdout and on_stderr.
The method can be called multiple times and mixed with ignore to aggregate multiple filters.
predicate is a function which takes a string as a parameter and returns a boolean value: true if the line should be processed, or false otherwise.
Returns a new instance of the Process class with the new filter.
ignore: see keep. Here, the predicate is reverted.
on_stdout: adds an action to perform when a line from stdout is received.
The method can be called multiple times if multiple actions should be performed for every line of stdout.
action: a function which takes a string as a parameter and doesn’t return anything.
Returns a new instance of the Process class with the new action.
on_stderr: see on_stdout. Here, it deals with stderr instead.
The file contains a bunch of filters which can be used in Process.keep and Process.ignore.
This class is used to log output from stdout or stderr.
The library was written for Python 3 under Linux. I haven’t tested it neither with Python 2, nor under Windows.
While I used Test Driven Development when creating this library and naturally have a 100% branch coverage, I don’t know neither Python, nor subprocess well enough to be sure that the library can be used reliably in production. Use at own risk.
If you want to contribute, contact me at firstname.lastname@example.org. You’ll be able to contribute to the project using the official SVN repository. If you find it more convinient to clone the source to GitHub, you can do that too.
The source code of the library and the corresponding documentation are covered by the MIT License.