Make your functions return something meaningful, typed, and safe!
Features
- Brings functional programming to Python land
- Provides a bunch of primitives to write declarative business logic
- Enforces better architecture
- Fully typed with annotations and checked with
mypy, PEP561 compatible - Adds emulated Higher Kinded Types support
- Provides type-safe interfaces to create your own data-types with enforced laws
- Has a bunch of helpers for better composition
- Pythonic and pleasant to write and to read
π - Support functions and coroutines, framework agnostic
- Easy to start: has lots of docs, tests, and tutorials
Quickstart right now!
Installation
pip install returns
You are also required to configure mypy correctly and install our plugin to fix this existing issue:
# In setup.cfg or mypy.ini:
[mypy]
plugins =
returns.contrib.mypy.returns_plugin
We also recommend to use the same mypy settings we use.
Make sure you know how to get started, check out our docs! Try our demo.
Contents
- Maybe container that allows you to write
None-free code - RequiresContext container that allows you to use typed functional dependency injection
- Result container that let's you to get rid of exceptions
- IO container and IOResult that marks all impure operations and structures them
- Future container and FutureResult to work with
asynccode - Write your own container! You would still have all the features for your own types (including full existing code reuse and type-safety)
Maybe container
None is called the worst mistake in the history of Computer Science.
So, what can we do to check for None in our programs? You can use builtin Optional type and write a lot of if some is not None: conditions. But, having null checks here and there makes your code unreadable.
user: Optional[User]
discount_program: Optional['DiscountProgram'] = None
if user is not None:
balance = user.get_balance()
if balance is not None:
credit = balance.credit_amount()
if credit is not None and credit > 0:
discount_program = choose_discount(credit)
Or you can use Maybe container! It consists of Some and Nothing types, representing existing state and empty (instead of None) state respectively.
from typing import Optional
from returns.maybe import Maybe, maybe
@maybe # decorator to convert existing Optional[int] to Maybe[int]
def bad_function() -> Optional[int]:
...
maybe_number: Maybe[float] = bad_function().bind_optional(
lambda number: number / 2,
)
# => Maybe will return Some[float] only if there's a non-None value
# Otherwise, will return Nothing
You can be sure that .bind_optional() method won't be called for Nothing. Forget about None-related errors forever!
We can also bind a Optional-returning function over a container. To achieve this, we are going to use .bind_optional method.
And that's how your initial refactored code will look like:
user: Optional[User]
# Type hint here is optional, it only helps the reader here:
discount_program: Maybe['DiscountProgram'] = Maybe.from_optional(
user,
).bind_optional( # This won't be called if `user is None`
lambda real_user: real_user.get_balance(),
).bind_optional( # This won't be called if `real_user.get_balance()` is None
lambda balance: balance.credit_amount(),
).bind_optional( # And so on!
lambda credit: choose_discount(credit) if credit > 0 else None,
)
Much better, isn't it?
RequiresContext container
Many developers do use some kind of dependency injection in Python. And usually it is based on the idea that there's some kind of a container and assembly process.
Functional approach is much simpler!
Imagine that you have a django based game, where you award users with points for each guessed letter in a word (unguessed letters are marked as '.'):
from django.http import HttpRequest, HttpResponse
from words_app.logic import calculate_points
def view(request: HttpRequest) -> HttpResponse:
user_word: str = request.POST['word'] # just an example
points = calculate_points(user_word)
... # later you show the result to user somehow
# Somewhere in your `words_app/logic.py`:
def calculate_points(word: str) -> int:
guessed_letters_count = len([letter for letter in word if letter != '.'])
return _award_points_for_letters(guessed_letters_count)
def _award_points_for_letters(guessed: int) -> int:
return 0 if guessed < 5 else guessed # minimum 6 points possible!
Awesome! It works, users are happy, your logic is pure and awesome. But, later you decide to make the game more fun: let's make the minimal accountable letters threshold configurable for an extra challenge.
You can just do it directly:
def _award_points_for_letters(guessed: int, threshold: int) -> int:
return 0 if guessed < threshold else guessed
The problem is that _award_points_for_letters is deeply nested. And then you have to pass threshold through the whole callstack, including calculate_points and all other functions that might be on the way. All of them will have to accept threshold as a parameter! This is not useful at all! Large code bases will struggle a lot from this change.
Ok, you can directly use django.settings (or similar) in your _award_points_for_letters function. And ruin your pure logic with framework specific details. That's ugly!
Or you can use RequiresContext container. Let's see how our code changes:
from django.conf import settings
from django.http import HttpRequest, HttpResponse
from words_app.logic import calculate_points
def view(request: HttpRequest) -> HttpResponse:
user_word: str = request.POST['word'] # just an example
points = calculate_points(user_words)(settings) # passing the dependencies
... # later you show the result to user somehow
# Somewhere in your `words_app/logic.py`:
from typing_extensions import Protocol
from returns.context import RequiresContext
class _Deps(Protocol): # we rely on abstractions, not direct values or types
WORD_THRESHOLD: int
def calculate_points(word: str) -> RequiresContext[int, _Deps]:
guessed_letters_count = len([letter for letter in word if letter != '.'])
return _award_points_for_letters(guessed_letters_count)
def _award_points_for_letters(guessed: int) -> RequiresContext[int, _Deps]:
return RequiresContext(
lambda deps: 0 if guessed < deps.WORD_THRESHOLD else guessed,
)
And now you can pass your dependencies in a really direct and explicit way. And have the type-safety to check what you pass to cover your back. Check out RequiresContext docs for more. There you will learn how to make '.' also configurable.
We also have RequiresContextResult for context-related operations that might fail. And also RequiresContextIOResult and RequiresContextFutureResult.
Result container
Please, make sure that you are also aware of Railway Oriented Programming.
Straight-forward approach
Consider this code that you can find in any python project.
import requests
def fetch_user_profile(user_id: int) -> 'UserProfile':
"""Fetches UserProfile dict from foreign API."""
response = requests.get('/api/users/{0}'.format(user_id))
response.raise_for_status()
return response.json()
Seems legit, does it not? It also seems like a pretty straightforward code to test. All you need is to mock requests.get to return the structure you need.
But, there are hidden problems in this tiny code sample that are almost impossible to spot at the first glance.
Hidden problems
Let's have a look at the exact same code, but with the all hidden problems explained.
import requests
def fetch_user_profile(user_id: int) -> 'UserProfile':
"""Fetches UserProfile dict from foreign API."""
response = requests.get('/api/users/{0}'.format(user_id))
# What if we try to find user that does not exist?
# Or network will go down? Or the server will return 500?
# In this case the next line will fail with an exception.
# We need to handle all possible errors in this function
# and do not return corrupt data to consumers.
response.raise_for_status()
# What if we have received invalid JSON?
# Next line will raise an exception!
return response.json()
Now, all (probably all?) problems are clear. How can we be sure that this function will be safe to use inside our complex business logic?
We really cannot be sure! We will have to create lots of try and except cases just to catch the expected exceptions. Our code will become complex and unreadable with all this mess!
Or we can go with the top level except Exception: case to catch literally everything. And this way we would end up with catching unwanted ones. This approach can hide serious problems from us for a long time.
Pipe example
import requests
from returns.result import Result, safe
from returns.pipeline import flow
from returns.pointfree import bind
def fetch_user_profile(user_id: int) -> Result['UserProfile', Exception]:
"""Fetches `UserProfile` TypedDict from foreign API."""
return flow(
user_id,
_make_request,
bind(_parse_json),
)
@safe
def _make_request(user_id: int) -> requests.Response:
# TODO: we are not yet done with this example, read more about `IO`:
response = requests.get('/api/users/{0}'.format(user_id))
response.raise_for_status()
return response
@safe
def _parse_json(response: requests.Response) -> 'UserProfile':
return response.json()
Now we have a clean and a safe and declarative way to express our business needs:
- We start from making a request, that might fail at any moment,
- Then parsing the response if the request was successful,
- And then return the result.
Now, instead of returning regular values we return values wrapped inside a special container thanks to the @safe decorator. It will return Success[YourType] or Failure[Exception]. And will never throw exception at us!
We also use flow and bind functions for handy and declarative composition.
This way we can be sure that our code won't break in random places due to some implicit exception. Now we control all parts and are prepared for the explicit errors.
We are not yet done with this example, let's continue to improve it in the next chapter.
IO container
Let's look at our example from another angle. All its functions look like regular ones: it is impossible to tell whether they are pure or impure from the first sight.
It leads to a very important consequence: we start to mix pure and impure code together. We should not do that!
When these two concepts are mixed we suffer really bad when testing or reusing it. Almost everything should be pure by default. And we should explicitly mark impure parts of the program.
That's why we have created IO container to mark impure functions that never fail.
These impure functions use random, current datetime, environment, or console:
import random
import datetime as dt
from returns.io import IO
def get_random_number() -> IO[int]: # or use `@impure` decorator
return IO(random.randint(1, 10)) # isn't pure, because random
now: Callable[[], IO[dt.datetime]] = impure(dt.datetime.now)
@impure
def return_and_show_next_number(previous: int) -> int:
next_number = previous + 1
print(next_number) # isn't pure, because does IO
return next_number
Now we can clearly see which functions are pure and which ones are impure. This helps us a lot in building large applications, unit testing you code, and composing business logic together.
Troublesome IO
As it was already said, we use IO when we handle functions that do not fail.
What if our function can fail and is impure? Like requests.get() we had earlier in our example.
Then we have to use a special IOResult type instead of a regular Result. Let's find the difference:
- Our
_parse_jsonfunction always returns the same result (hopefully) for the same input: you can either parse validjsonor fail on invalid one. That's why we return pureResult, there's noIOinside - Our
_make_requestfunction is impure and can fail. Try to send two similar requests with and without internet connection. The result will be different for the same input. That's why we must useIOResulthere: it can fail and hasIO
So, in order to fulfill our requirement and separate pure code from impure one, we have to refactor our example.
Explicit IO
Let's make our IO explicit!
import requests
from returns.io import IOResult, impure_safe
from returns.result import safe
from returns.pipeline import flow
from returns.pointfree import bind_result
def fetch_user_profile(user_id: int) -> IOResult['UserProfile', Exception]:
"""Fetches `UserProfile` TypedDict from foreign API."""
return flow(
user_id,
_make_request,
# before: def (Response) -> UserProfile
# after safe: def (Response) -> ResultE[UserProfile]
# after bind_result: def (IOResultE[Response]) -> IOResultE[UserProfile]
bind_result(_parse_json),
)
@impure_safe
def _make_request(user_id: int) -> requests.Response:
response = requests.get('/api/users/{0}'.format(user_id))
response.raise_for_status()
return response
@safe
def _parse_json(response: requests.Response) -> 'UserProfile':
return response.json()
And later we can use unsafe_perform_io somewhere at the top level of our program to get the pure (or "real") value.
As a result of this refactoring session, we know everything about our code:
- Which parts can fail,
- Which parts are impure,
- How to compose them in a smart, readable, and typesafe manner.
Future container
There are several issues with async code in Python:
- You cannot call
asyncfunction from a sync one - Any unexpectedly thrown exception can ruin your whole event loop
- Ugly composition with lots of
awaitstatements
Future and FutureResult containers solve these issues!
Mixing sync and async code
The main feature of Future is that it allows to run async code while maintaining sync context. Let's see an example.
Let's say we have two functions, the first one returns a number and the second one increments it:
async def first() -> int:
return 1
def second(): # How can we call `first()` from here?
return first() + 1 # Boom! Don't do this. We illustrate a problem here.
If we try to just run first(), we will just create an unawaited coroutine. It won't return the value we want.
But, if we would try to run await first(), then we would need to change second to be async. And sometimes it is not possible for various reasons.
However, with Future we can "pretend" to call async code from sync code:
from returns.future import Future
def second() -> Future[int]:
return Future(first()).map(lambda num: num + 1)
Without touching our first async function or making second async we have achieved our goal. Now, our async value is incremented inside a sync function.
However, Future still requires to be executed inside a proper eventloop:
import anyio # or asyncio, or any other lib
# We can then pass our `Future` to any library: asyncio, trio, curio.
# And use any event loop: regular, uvloop, even a custom one, etc
assert anyio.run(second().awaitable) == 2
As you can see Future allows you to work with async functions from a sync context. And to mix these two realms together. Use raw Future for operations that cannot fail or raise exceptions. Pretty much the same logic we had with our IO container.
Async code without exceptions
We have already covered how Result works for both pure and impure code. The main idea is: we don't raise exceptions, we return them. It is especially critical in async code, because a single exception can ruin all our coroutines running in a single eventloop.
We have a handy combination of Future and Result containers: FutureResult. Again, this is exactly like IOResult, but for impure async code. Use it when your Future might have problems: like HTTP requests or filesystem operations.
You can easily turn any wild throwing coroutine into a calm FutureResult:
import anyio
from returns.future import future_safe
from returns.io import IOFailure
@future_safe
async def raising():
raise ValueError('Not so fast!')
ioresult = anyio.run(raising.awaitable) # all `Future`s return IO containers
assert ioresult == IOFailure(ValueError('Not so fast!')) # True
Using FutureResult will keep your code safe from exceptions. You can always await or execute inside an eventloop any FutureResult to get sync IOResult instance to work with it in a sync manner.
Better async composition
Previously, you had to do quite a lot of awaiting while writing async code:
async def fetch_user(user_id: int) -> 'User':
...
async def get_user_permissions(user: 'User') -> 'Permissions':
...
async def ensure_allowed(permissions: 'Permissions') -> bool:
...
async def main(user_id: int) -> bool:
# Also, don't forget to handle all possible errors with `try / except`!
user = await fetch_user(user_id) # We will await each time we use a coro!
permissions = await get_user_permissions(user)
return await ensure_allowed(permissions)
Some people are ok with it, but some people don't like this imperative style. The problem is that there was no choice.
But now, you can do the same thing in functional style! With the help of Future and FutureResult containers:
import anyio
from returns.future import FutureResultE, future_safe
from returns.io import IOSuccess, IOFailure
@future_safe
async def fetch_user(user_id: int) -> 'User':
...
@future_safe
async def get_user_permissions(user: 'User') -> 'Permissions':
...
@future_safe
async def ensure_allowed(permissions: 'Permissions') -> bool:
...
def main(user_id: int) -> FutureResultE[bool]:
# We can now turn `main` into a sync function, it does not `await` at all.
# We also don't care about exceptions anymore, they are already handled.
return fetch_user(user_id).bind(get_user_permissions).bind(ensure_allowed)
correct_user_id: int # has required permissions
banned_user_id: int # does not have required permissions
wrong_user_id: int # does not exist
# We can have correct business results:
assert anyio.run(main(correct_user_id).awaitable) == IOSuccess(True)
assert anyio.run(main(banned_user_id).awaitable) == IOSuccess(False)
# Or we can have errors along the way:
assert anyio.run(main(wrong_user_id).awaitable) == IOFailure(
UserDoesNotExistError(...),
)
Or even something really fancy:
from returns.pointfree import bind
from returns.pipeline import flow
def main(user_id: int) -> FutureResultE[bool]:
return flow(
fetch_user(user_id),
bind(get_user_permissions),
bind(ensure_allowed),
)
Later we can also refactor our logical functions to be sync and to return FutureResult.
Lovely, isn't it?
More!
Want more? Go to the docs! Or try our demo. Or read these articles:
- Python exceptions considered an anti-pattern
- Enforcing Single Responsibility Principle in Python
- Typed functional Dependency Injection in Python
- How Async Should Have Been
- Higher Kinded Types in Python
Do you have an article to submit? Feel free to open a pull request!
β
Drylabs maintains dry-python and helps those who want to use it inside their organizations.
Read more at drylabs.io
![Add Future[A] that helps with async code](https://avatars.githubusercontent.com/u/35063695?v=4)
31 Dec 30, 2022
56 Dec 14, 2022
1 Oct 24, 2022
4 Nov 17, 2021
4 Sep 13, 2021
1 Oct 8, 2022
1 Jan 28, 2022
18 Dec 31, 2022
14 May 31, 2022
0 Aug 31, 2022
2 Nov 24, 2021
1 Dec 20, 2021
7 Sep 28, 2021
73 Dec 14, 2022
5 Jul 4, 2022
1 Feb 23, 2022
72 Oct 9, 2022
13 Nov 17, 2022
3 Feb 12, 2022