pipetools enables function composition similar to using Unix pipes.

It allows forward-composition and piping of arbitrary functions - no need to decorate them or do anything extra.

It also packs a bunch of utils that make common operations more convenient and readable.

Source is on github.

Why?

Piping and function composition are some of the most natural operations there are for plenty of programming tasks. Yet Python doesn’t have a built-in way of performing them. That forces you to either deep nesting of function calls or adding extra glue code.

Example

Say you want to create a list of python files in a given directory, ordered by filename length, as a string, each file on one line and also with line numbers:

>>> print(pyfiles_by_length('../pipetools'))
1. ds_builder.py
2. __init__.py
3. compat.py
4. utils.py
5. main.py

All the ingredients are already there, you just have to glue them together. You might write it like this:

def pyfiles_by_length(directory):
    all_files = os.listdir(directory)
    py_files = [f for f in all_files if f.endswith('.py')]
    sorted_files = sorted(py_files, key=len, reverse=True)
    numbered = enumerate(py_files, 1)
    rows = ("{0}. {1}".format(i, f) for i, f in numbered)
    return '\n'.join(rows)

Or perhaps like this:

def pyfiles_by_length(directory):
    return '\n'.join('{0}. {1}'.format(*x) for x in enumerate(reversed(sorted(
        [f for f in os.listdir(directory) if f.endswith('.py')], key=len)), 1))

Or, if you’re a mad scientist, you would probably do it like this:

pyfiles_by_length = lambda d: (reduce('{0}\n{1}'.format,
    map(lambda x: '%d. %s' % x, enumerate(reversed(sorted(
        filter(lambda f: f.endswith('.py'), os.listdir(d)), key=len))))))

But there should be one – and preferably only one – obvious way to do it.

So which one is it? Well, to redeem the situation, pipetools give you yet another possibility!

pyfiles_by_length = (pipe
    | os.listdir
    | where(X.endswith('.py'))
    | sort_by(len).descending
    | (enumerate, X, 1)
    | foreach("{0}. {1}")
    | '\n'.join)

Why would I do that, you ask? Comparing to the native Python code, it’s

  • Easier to read – minimal extra clutter
  • Easier to understand – one-way data flow from one step to the next, nothing else to keep track of
  • Easier to change – want more processing? just add a step to the pipeline
  • Removes some bug opportunities – did you spot the bug in the first example?

Of course it won’t solve all your problems, but a great deal of code can be expressed as a pipeline, giving you the above benefits. Read on to see how it works!

Installation

$ pip install pipetools

Uh, what’s that?

Usage

The pipe

The pipe object can be used to pipe functions together to form new functions, and it works like this:

from pipetools import pipe

f = pipe | a | b | c

# is the same as:
def f(x):
    return c(b(a(x)))

A real example, sum of odd numbers from 0 to x:

from functools import partial
from pipetools import pipe

odd_sum = pipe | range | partial(filter, lambda x: x % 2) | sum

odd_sum(10)  # -> 25

Note that the chain up to the sum is lazy.

Automatic partial application in the pipe

As partial application is often useful when piping things together, it is done automatically when the pipe encounters a tuple, so this produces the same result as the previous example:

odd_sum = pipe | range | (filter, lambda x: x % 2) | sum

As of 0.1.9, this is even more powerful, see X-partial.

Built-in tools

Pipetools contain a set of pipe-utils that solve some common tasks. For example there is a shortcut for the filter class from our example, called where():

from pipetools import pipe, where

odd_sum = pipe | range | where(lambda x: x % 2) | sum

Well that might be a bit more readable, but not really a huge improvement, but wait!

If a pipe-util is used as first or second item in the pipe (which happens quite often) the pipe at the beginning can be omitted:

odd_sum = range | where(lambda x: x % 2) | sum

See pipe-utils’ documentation.

OK, but what about the ugly lambda?

where(), but also foreach(), sort_by() and other pipe-utils can be quite useful, but require a function as an argument, which can either be a named function – which is OK if it does something complicated – but often it’s something simple, so it’s appropriate to use a lambda. Except Python’s lambdas are quite verbose for simple tasks and the code gets cluttered…

X object to the rescue!

from pipetools import where, X

odd_sum = range | where(X % 2) | sum

How ‘bout that.

Read more about the X object and it’s limitations.

Automatic string formatting

Since it doesn’t make sense to compose functions with strings, when a pipe (or a pipe-util) encounters a string, it attempts to use it for (advanced) formatting:

>>> countdown = pipe | (range, 1) | reversed | foreach('{}...') | ' '.join | '{} boom'
>>> countdown(5)
'4... 3... 2... 1... boom'

Feeding the pipe

Sometimes it’s useful to create a one-off pipe and immediately run some input through it. And since this is somewhat awkward (and not very readable, especially when the pipe spans multiple lines):

result = (pipe | foo | bar | boo)(some_input)

It can also be done using the > operator:

result = some_input > pipe | foo | bar | boo

Note

Note that the above method of input won’t work if the input object defines __gt__ for any object - including the pipe. This can be the case for example with some objects from math libraries such as NumPy. If you experience strange results try falling back to the standard way of passing input into a pipe.