ZIO-py

ZIO for Python (see https://github.com/zio/zio).

This is a fairly straightforward (unofficial and plucky) port of a subset of the ingenious Scala ZIO library along with some other tools for type-safe functional programming in Python.

About the ZIO monad

This particular implementation of the ZIO data structure is based on the functional effects training provided by John De Goes. It is a vastly simplified version of the amazing official Scala library, but is still quite useful: For state of the art "errors as values" computations, this library is for you!

The ZIO[R, E, A] monad is basically three monads rolled into one:

• An "IO monad" for writing pure functional programs. A value of type ZIO[R, E, A] is a program, which when evaluated given input of type R, either fails with a value of type E or succeeds with a value of type A.

• An either monad that allows you to "code to the happy path." If an error is encountered anywhere in a sequence of ZIO operations, it is returned early.

• A reader monad for providing inputs to your program.

Unlike Scala's ZIO, this implementation does not include functionality for parallel/concurrent programming. Since we are stuck in Python with the Global Interpreter Lock, we can't have nice things...so this functionality won't be missed anyways. However, future work will certainly explore supporting this part of the Scala ZIO API.

Perhaps the most important feature of ZIO-py that sets it apart from all other functional programming libraries is its support for type-safe, ergonomic, and quite natural "monadic do notation."

Notable Features

• State of the art "zivariant" ZIO monad data structure.
• Either monad for those times when you don't need all of the power of ZIO.
• ZIOArrow and EitherArrow combinator data structures for ergonomic and type-safe "pipeline-oriented programming" (function composition). These features are awesome for wrapping existing APIs.
• Monadic do notation, what appears to be a significant advancement in bringing ergonomic functional programming to mainstream languages. It looks like this general idea was explored in 2018 in Exceptionally Monadic Error Handling, albeit from the Haskell side. Interesting, I have not seen the idea applied anywhere in the wild.
• HList and Validation implementations for type safe data validation and transformation.
• COMING SOON (ETA December 2023): A state of the art "nano actor" runtime for doing All The Things concurrently! Huge and special thanks to Kit Langton and Adam Fraser for their amazing Zymposium series and guidance while we were working together at Ziverge.

Benefits

• Faster and safer test writing: No more mocking and other monkeypatching tomfoolery. Ok, maybe there is some hyperbole here. But it should significantly reduce the amount of mocking needed. Instead of mocking things, you simply my_program.provide(x) your program a test environment, x, before calling unsafe_run(my_program). When running code in production, you will .provide an instance of a live (real) environment.

• Clear distinction of side-effecting code based on function type signatures. If a function returns a value of type ZIO[R, E, A], you know exactly what that function takes as input, how it can fail, and what it will return if it succeeds. You also know that the function may cause side effects. Any other function can, with some reasonable discipline, be considered free of side effects.

• Code to the happy path while safely handling errors. Since ZIO behaves like a (right-biased) Either monad, it is super easy to do railway-oriented programming.

• Type safety. Everything is statically-checked by mypy. If you get the types wrong, then there is probably a bug in your program. The mypy type checker will find the bugs before they make it to production.

• Implementations of "Kleisli arrow"-like combinator interfaces for creating function pipelines (EitherArrow and ZIOArrow). These interfaces use PEP 612 features to preserve function parameters, making it possible to easily work with pre-existing Python functions/methods. For example, it is super handy for wrapping existing APIs to make them type-safe and composable.

• It's pure, it's functional. It's pure functional programming. It's Pythonic. It shouldn't be possible. (And someone somewhere is upset that these meddling kids have made it possible.)

Installation

At the command line:

$ pip install zio-py

Alternatively, you can install zio-py from source by cloning this repo and running the provided setup.sh script.

How to Use It

Check out the Scala ZIO documentation for the definitive guide on basic usage of ZIO. In Scala. :)

Here, we will introduce you to the style of programming that uses the generalized monadic do notation that is unique to ZIO-py.

Using the "Monadic Do Notation"

ZIO-py features a kind of type-safe monadic do notation that obviates the need to festoon your pure functional programs with unruly nested .flat_map calls. Unlike other "monadic do notation as a library" implementations, this one is 100% type-safe.

To use it within the body of a function:

1. Decorate your function using the @ziopy.zio.monadic decorator. (Or, correspondingly, decorate your method with @ziopy.zio.monadic_method. Two different decorators are needed because methods take an implicit self argument.)
2. The first parameter to your function must be of type ziopy.zio.ZIOMonad[R, E], where R represents the environment type and E represents the error type. A good name for this parameter is typically do.
3. Add any other parameters to your function after the ZIOMonad parameter.
4. Return a value of type ziopy.zio.ZIO[R, E, A] from your function, where A represents the type returned when your function returns successfully.

The types R and E have to coincide for type safety. The PEP 612 features of the mypy type checker will check this for you.

Then, instead of writing

a.flat_map(lambda b: ...)

you can write

b = do << a
...

That's pretty much it! The type safety guarantees that, if any statement in your monadic code that passed through a do << produces an error, the @monadic function has to be capable of returning that error. The same safety idea is used for accessing stuff out of the environment (the R in ZIO[R, E, A]).

It turns out to be a lot easier to use than Scala's "for comprehension" and Haskell's "do notation" because it's just a regular statement. So you can mix it with loops, conditional logic, etc., which is not possible in those other languages.

How the Monadic Do Notation Works

Each do << program invocation calls the private (and potentially impure) program._run function, which returns a value of type Either[E, A]. More specifically, it returns either an instance of Left[E] or an instance of Right[A]. If left: Left[E] is returned, we wrap left.value in a special exception called RaiseLeft.

Meanwhile, the @monadic function decorator adds an exception handler to the decorated function. It catches raise_left: RaiseLeft exceptions, and returns the wrapped value as a ZIO program ZIO.fail(lambda: raise_left.value).

The end result is a control flow mechanism for early return of Left[E] values from your decorated functions.

Example Programs

from typing import NoReturn, Union

import ziopy.services.console as console
import ziopy.services.system as system
from ziopy.either import EitherArrow
from ziopy.environments import ConsoleSystemEnvironment
from ziopy.services.console import Console, LiveConsole
from ziopy.zio import ZIO, Environment, ZIOMonad, monadic, unsafe_run


@monadic
def program(
    do: ZIOMonad[Console, Union[EOFError, KeyboardInterrupt]]
) -> ZIO[
    Console,
    Union[EOFError, KeyboardInterrupt],
    str
]:
    con = do << Environment()

    do << con.print("Hello, what is your name?")
    name = do << con.input()
    do << con.print(f"Your name is: {name}")
    x = do << ZIO.succeed(1)

    while x < 20:
        x = do << (
            ZIO.succeed(x)
            .map(lambda p: p + 1)
            .flat_map(lambda q: ZIO.succeed(q - 1))
            .flat_map(lambda r: ZIO.succeed(r + 1))
        )

    do << con.print(f"The value of x is: {x}")
    return ZIO.succeed(f"Hello, {name}!")


p = program().provide(LiveConsole())
final_result = unsafe_run(p)
print(f"Final result (1) is: {final_result}")

# You can run the same program (value) over and over again.
final_result_2 = unsafe_run(p)
print(f"Final result (2) is: {final_result_2}")


@monadic
def prog(
    do: ZIOMonad[ConsoleSystemEnvironment, NoReturn]
) -> ZIO[ConsoleSystemEnvironment, NoReturn, int]:
    age = do << console.get_input_from_console(
        prompt="How old are you?\n",
        parse_value=(
            EitherArrow.from_callable(str)
            .map(int)
            .catch(ValueError)
        ),
        default_value=21
    )
    do << console.print(f"You are {age} years old.")
    return ZIO.succeed(age)


unsafe_run(
    prog().provide(
        ConsoleSystemEnvironment(console=LiveConsole(), system=system.LiveSystem())
    )
)

History

ZIO-py grew out of a 2019 Root Insurance Company Hack Days project which experimented with porting ZIO to Python. The barrier to adoption was the fact that Python did not have a good mechanism for handling monadic programming, such as Scala's for comprehension or Haskell's do notation. I implemented the beginnings of an AST transformer that made it possible to use a kind of primitive do notation here, but generalizing it to work with general Python AST transformations was extremely difficult. Without a better syntax for monadic programming, nobody would ever want to use it in Python. Nested .flat_map everywhere is a mess.

After letting the problem simmer in my head for more than a year, I suddenly had an epiphany one morning:

"Solve the inversion of control flow problem, and you'll have a better monadic "do" notation than any programming language currently offers."

So that's what I did. I tried out a few different designs, trying to emulate something analogous to call/cc in a typesafe way in Python. Next, I used a fork/exec strategy to simulate call/cc. Ultimately I was able to construct a design that eschewed call/cc, using only try/catch and an additional thunk in the @monadic decorator to achieve the desired control flow.

One of the limiting reagents was that mypy still has some problems with type inference with code that uses decorators. So, for the short term, I whipped together a simple mypy plugin that properly checks/modifies the type signature of functions that are decorated as @monadic.

Figuring out a way to use the library in a type safe way was tricky. I had to subconsciously think for a few days about how to maintain the type safety. The @monadic decorator, do: ZIOMonad[R, E] argument, and mypy plugin solved that problem pretty well methinks...but YMMV.

What's Next?

Async/concurrent functionality is currently on the radar. The end goal is to make ZIO-py into one of the best solutions for async/concurrent programming, and to help developers take full advantage of the future of Python when the global interpreter lock has been removed.

Statement of Public Good

This project is made possible by:

• The Mathematics and Informatics Institute of Ohio, a nonprofit whose mission is to enrich the citzenry of the State of Ohio via education and public domain contributions to advanced mathematics, computer science, informatics, information theory, data science, and other analytical disciplines.
• Root Insurance Company. This library is an open source version of one of our widely-used internal Python libraries.
• John De Goes and the many Scala ZIO contributors.