xfp

XFP

Since Python 3.0 map, filter & co (but more accurately even since Python 2 list comprehension), new versions of the language keep appending more and more functional elements, last in date being Generics, Union typing or Pattern Matching.
However although functional programming in its roughest form is possible in Python, it fails in our opinion to keep itself nice and readable.

# Look this awful little chunk of code, how cute it is <3

from functools import reduce

initial_value = ["oh", "look", "an", "array", "to", "process", "!"]
camel_cased = map(lambda chain: str(chain[0].upper()) + chain[1:], initial_value)
only_long_word = filter(lambda x: len(x) > 2, camel_cased)
output = reduce(lambda x, y: x + " " + y, only_long_word)

assert output == "Look Array Process"

This project aims to soften the functional syntax already existing within the language, and go even further by enabling more functional concepts.

Github repository HERE

WHY

While python provides today some tools from functional languages (such as map, filter, ...), it fails as making its syntax functional-friendly. For example, multiple mapping/filtering requires a lot of intermediate values with few addition to code readability or robustess. Moreover some paradigms are missing to fully benefits from monadic behavior (map/flat_map/filter/foreach/...).

The goal is to :

• add functional syntax to make functional python code edible
• add tools to complete the functionalities provided by python
• respect python strengthes : time to market, readability, ...

In order to achieve all of this, we propose a functional API adapted for python, taking into account its strengths and weaknesses to enrich the language without twisting it too much.

DOCUMENTATION

Please see the full documentation for more information.

QUICK START

xfp is plublished on PyPI HERE, easy to install with your package manager.

How to run the demos

Some demos are provided in the demo folder. Each one is in a separate subfolder.

For simplicity :

• main code is in the main.py file
• eventual modules are in the same folder and used in main.py with relative imports

To run the demos (here the xlist one):

• make sure xfp is installed on your python environment (eg cd python-fp && pip install .)
• Run the demo from the root of the repo :python -m demo.xiter.main

How to use in your project

Use with Collections

To use XFP on a collection, starts with creating a new Xlist :

from xfp import Xlist, Xiter
xlist = Xlist([1, 2, 3])
# or xiter for the lazy iterator version. Functionalities differ a bit however
xiter = Xiter([1, 2, 3])

You can then start applying operations on the list, either through anonymous functions or defined ones.
The preconised style is to write one operation by line using the '()' operator :

from xfp import Xlist

def under_eight(x: int) -> bool:
    return x < 8

(
    Xlist([1, 2, 3])
    .map(lambda x: x * x)                      # Xlist([1, 4, 9])
    .filter(under_eight)                       # Xlist([1, 4])
    .map(lambda x: f"this is a number : {x}")  # Xlist(["this is a number : 1", "this is a number : 4"])
    .foreach(print)                            # prints each element of the list, return None
)

Side-effects handling

Functional behaviors requires proper encapsulation of 'not a value' meaning (for example, None or raise Exception).
Those ecapsulations are modelised in xfp through the Xresult class. It basically encapsulates a union type under two pathways, either LEFT or RIGHT, in a container. Think of this container as a 'list with one element'. Its API is homogene with the collection one.

from xfp import Xresult, XRBranch

r1 = Xresult(1, XRBranch.RIGHT)
r2 = Xresult(3, XRBranch.LEFT)

(
    r1
    .map_right(lambda x: x + 3)                        # XRBranch.RIGHT : 4
    .flat_map_right(lambda x: r2.map(lambda y: x + y)) # XRBranch.LEFT : 3
    .filter_left(lambda x: x > 5)                      # XRBranch.RIGHT : XresultError(...)
)

Results chaining

You will often have to deal with multiple effects at once. To avoid the vanilla triangle of doom that would cause such dealing, xfp provides a convenient way to handle them altogether.
Let's illustrate it with a mock use case. A table computing and writing from three different sources :

from xfp import Xresult

def load_table(table_name: str) -> Xresult[Exception, DataFrame]:
    pass


def write_table(table_name: str, table: DataFrame) -> Xresult[Exception, None]:
    pass

def process(t1: DataFrame, t2: DataFrame, t3: DataFrame) -> DataFrame:
    pass

# 'Vanilla' xfp processing

load_table('db1.tb1').flat_map(
    lambda t1: load_table('db2.tb2').flat_map(
        lambda t2:load_table('db3.tb3').flat_map(
            lambda t3: write_table('db1.tb4', process(t1, t2, t3))
        )
    )
)

# Xfp result chaining

Xresult.fors(lambda:
    [
        write_table('db1.tb4', process(t1, t2, t3))
        for t1, t2, t3
        in zip(
            load_table('db1.tb1'),
            load_table('db2.tb2'),
            load_table('db3.tb3')
        )
    ])

Quality of life

Util functions

In functional programming, the operation consisting in transforming the function f in g (see below) is called curryfiction :

fromp xfp import Xlist
def f(i: int, j: str) -> Xlist[str]:
    pass

def g(i: int) -> Callable[[str], Xlist[str]]:
    def inner(j: str) -> Xlist[str]:
        pass
    return inner

While the g syntax is often useful (for example to prepare functions to use in a map operation), the writing of such function may be tedious. XFP comes with a convenient decorator curry to infer the g function from the f one:

from xfp import Xlist, curry

# the effective signature of f becomes def f(i: int) -> Callable[[str], Xlist[str]]
@curry
def f(i: int, j: str) -> Xlist[str]:
    return i * j

# notice the usage of only one parameter in f
(
    Xlist(["a", "b", "c"])
    .flat_map(f(3))
    .foreach(print)
)

Xeither, Xtry, Xopt

You can add more semantic to your results by making use of the proxy types Xeither, Xtry, Xopt, respectively indicating "a formal union type", "something that can crash", "the presence or absence of an element".
Those types resolves as an Xresult, but can be used by themselves in pattern matching, and provide tooling revolving around their semantics. Example of Xtry :

from xfp import Xtry, Xresult

def should_raise(x):
    if x > 10:
        raise Exception("too much")
    else:
        return x

r1 = Xtry.from_unsafe(lambda: should_raise(15)) # Xtry.Failure(Exception("too much"))
r2 = Xtry.from_unsafe(lambda: should_raise(8))  # Xtry.Success(8)

# a decorator is provided to automatically convert your functions
@Xtry.safed
def safed_function(x):
    return should_raise(x)

r3: Xresult[Exception, int] = safed_function(15) # Xtry.Failure(Exception("too much"))
r4: Xresult[Exception, int] = safed_function(8)  # Xtry.Success(8)

# Constructors are also available
r5: Xresult[Exception, int] = Xtry.Success(3)

# You can pattern match an expression depending on its pathway
match r3:
    case Xtry.Success(value):
        print(value)
    case Xtry.Failure(exception):
        print(f"Something went wrong : {exception}")

HOW TO CONTRIBUTE

Setup

• clone the repo
• install poetry
• install compatible python version (from 3.12), eg. pyenv install 3.12.4
• install the project : poetry install
• set up the git hook scripts (linter / formatter): pre-commit install

-> poetry installs xfp package in editable mode, so that xfp is available as a package from anywhere and editable.

Linter / formatter = ruff

Ruff is hooked on pre-commit as linter and formatter.
More here : https://github.com/astral-sh/ruff

Pre-commit

More info : https://pre-commit.com/

CI/CD (Github)

• Main branch : no pushes, only merges from branches that pass tests
• github action pytest-action is setup to run pytest on each push
• TODO : auto deploy to PyPI

HOW TO PUBLISH TO PYPI

By now, pypi credentials (token) have to be configured locally with :
poetry config pypi-token.pypi <token>

To publish locally (from given branch) : poetry build && poetry publish