pipe-operator 1.0.1

Elixir's pipe operator in Python

✨ Pipe Operator ✨

• ✨ Pipe Operator ✨
  • ⚡ Quick start
  • 🐍 Pythonic implementation
    • Available classes
    • Limitations
  • 🍹 Elixir-like implementation
    • Overview
    • Operations and shortcuts
    • How it works
    • Linters and type-checkers issues
  • 🔗 Useful links

pipe_operator allows you to use an elixir pipe-like syntax in python. This module provides 2 vastly different implementations, each with its own pros and cons.

⚡ Quick start

As simple as pip install pipe_operator. Then either import the 🐍 pythonic or the 🍹 elixir implementations

from pipe_operator.elixir_flow import elixir_pipe, tap, then
from pipe_operator.python_flow import Pipe, PipeArgs, PipeEnd, PipeStart, Tap, Then

You can use the 🐍 pythonic implementation, which is entirely compatible with linters and type-checkers, but a bit more verbose than the original pipe operator:

from pipe_operator.python_flow import Pipe, PipeArgs, PipeEnd, PipeStart, Tap, Then

result = (
    PipeStart("3")                        # starts the pipe
    >> Pipe(int)                          # function with 1-arg
    >> Pipe(my_func, 2000, z=10)          # function with multiple args
    >> Tap(print)                         # side effect
    >> Then(lambda x: x + 1)              # lambda
    >> Pipe(MyClass)                      # class
    >> Pipe(MyClass.my_classmethod)       # classmethod
    >> Tap(MyClass.my_method)             # side effect that can update the original object
    >> Pipe(MyClass.my_other_method)      # method
    >> Then[int, int](lambda x: x * 2)    # explicitly-typed lambda
    >> PipeArgs(my_other_func, 4, 5, 6)   # special case for functions with no positional/keyword parameters
    >> PipeEnd()                          # extract the value
)

Or the 🍹 elixir-like implementation, whose syntax greatly resembles the original pipe operator, but has major issues with linters and type-checkers.

from pipe_operator.elixir_flow import elixir_pipe, tap, then

@elixir_pipe
def workflow(value):
    results = (
        value                           # raw value
        >> BasicClass                   # class call
        >> _.value                      # property (shortcut)
        >> BasicClass()                 # class call
        >> _.get_value_plus_arg(10)     # method call
        >> 10 + _ - 5                   # binary operation (shortcut)
        >> {_, 1, 2, 3}                 # object creation (shortcut)
        >> [x for x in _ if x > 4]      # comprehension (shortcut)
        >> (lambda x: x[0])             # lambda (shortcut)
        >> my_func(_)                   # function call
        >> tap(my_func)                 # side effect
        >> my_other_func(2, 3)          # function call with extra args
        >> then(lambda a: a + 1)        # then
        >> f"value is {_}"              # formatted string (shortcut)
    )
    return results

workflow(3)

🐍 Pythonic implementation

Available classes

In the python_flow submodule, we expose the following classes:

Class	Description	Examples
PipeStart	The start of the pipe	PipeStart("3")
Pipe	Used to call almost any functions or classes, or methods	Pipe(int), Pipe(my_func, 2000, z=10)
PipeArgs	Same as Pipe but for function with no positional/keyword parameters	PipeArgs(func, 1, 2)
Then	Same as Pipe, but for 1-arg lambda functions	Then(lambda x: x.attribute)
Tap	Used to trigger a side effect (meaning it returns the original value)	Tap(print), Tap(lambda x: x.method())
PipeEnd	The end of the pipe, to extract the raw final result	PipeEnd()

Limitations

property: Properties cannot be called directly. You must resort to the use of Then(lambda x: x.my_property). This will work just fine and ensure type-safety throughout the pipe.

functions without positional/keyword parameters: While they are technically supported by the Pipe class, your type-checker will not handle them properly, because the Pipe class expect the function to have at least 1 positional/keyword parameter (ie the first one, passed down the pipe). To bypass this limitation, you should use PipeArgs instead.

pyright: pyright seems to have trouble dealing with our >> in some specific cases. As such, we advise you set reportOperatorIssue = "none" in your pyright config.

🍹 Elixir-like implementation

Overview

In the elixir_flow submodule, we expose 3 items:

• elixir_pipe: a decorator that enables the use of "pipe" in our function
• tap: a function to trigger a side-effect and return the original value
• then: (optional) the proper way to pass lambdas into the pipe

The elixir_pipe decorator can take arguments allowing you to customize

# Those are the default args
@elixir_pipe(placeholder="_", lambda_var="_pipe_x", operator=">>", debug=False)
def my_function()
    ...

• placeholder: The expected variable used in shortcut like _.property
• lambda_var: The variable named used internally when we generate lambda function. You'll likely never change this
• operator: The operator used in the pipe
• debug: If true, will print the output after each pipe operation

Operations and shortcuts

Initially, all operations can be supported through the base operations, with lambdas allowing you to perform any other operations. To make lambda usage cleaner, you can write them into then calls as well.

Operation	Input	Output
function calls	a >> b(...)	b(a, ...)
class calls	a >> B(...)	B(a, ...)
calls without parenthesis	a >> b	b(a)
lambda calls	a >> (lambda x: x + 4)	(lambda x: x + 4)(a)

However, we've also added shortcuts, based on the placeholder argument, allowing you to skip the lambda declaration and directly perform the following operations:

Operation	Input	Output
method calls	a >> _.method(...)	a.method(...)
property calls	a >> _.property	a.property
binary operators	a >> _ + 3	(lambda Z: Z + 3)(a)
f-strings	a >> f"{_}"	(lambda Z: f"{Z}")(a)
list/set/... creations	a >> [_, 1, 2]	(lambda Z: [Z, 1, 2])(a)
list/set/... comprehensions	a >> [x + _ for x in range(_)]	(lambda Z: [x + Z for x in range(Z)])(a)

How it works

Here's quick rundown of how it works. Feel free to inspect the source code or the tests. Once you've decorated your function and run the code:

• We pull the AST from the original function
• We remove our own decorator, to avoid recursion and impacting other functions
• We then rewrite the AST, following a specific set of rules (as shown in the table below)
• And finally we execute the re-written AST

Eventually, a >> b(...) >> c(...) becomes c(b(a, ...), ...).

Linters and type-checkers issues

Sadly, this implementation comes short when dealing with linters (like ruff or flake8) and type-checkers (like mypy or pyright). Because these are static code analyzers, they inspect the original code, and not your AST-modified version. To bypass the errors, you'll need to disable the followings:

• mypy: Either ignore operator,call-arg,call-overload,name-defined, or ignore just name-defined and use the @no_type_check decorator
• pyright: Set reportOperatorIssue, reportCallIssue, reportUndefinedVariable to none
• ruff: Disable the F821 error
• flake8: Disable the F821 error

🔗 Useful links

• Want to contribute?
• See what's new!
• Originally forked from robinhilliard/pipes