alleycat-reactive 0.2.0

A simple Python library to provide an API to implement the Reactive Object Pattern (ROP).

AlleyCat - Reactive

A part of the AlleyCat project which supports the Reactive Object Pattern.

Introduction

AlleyCat Reactive is a project to explore the possibility of bridging the gap between the two most widely used programming paradigms, namely, the object-oriented programming (OOP), and functional programming (FP).

It aims to achieve its goal by proposing a new design pattern based on the Reactive Extensions (Rx).

Even though it is already available on PyPI repository as alleycat-reactive package, the project is currently at a proof-of-concept stage and highly experimental.

As such, there can be significant changes in the API at any time. Furthermore, the project may be discontinued in future if the idea proves to be an unuseful one.

Reactive Object Pattern

The library provides an API to implement what we term as Reactive Object Pattern or ROP for short. Despite its rather pretentious name, it merely means defining class properties which can also serve as an Observable in Rx.

But why do we need such a thing?

If you are a seasoned programmer of an OOP language (which includes Python, by the way), you may feel like the paradigm already provides everything you need.

On the other hand, if you are already familiar with Rx, you may wonder what OOP has to do with the concept, as it's mostly about composing functions, not classes or properties.

Arguably, the most significant benefit that any functional approach brings could be its ability to define a process or a value in a declarative manner.

We won't delve into this subject too much since you can learn about the concept from many other websites.

In short, it's much better to define data in a declarative manner, or as "data pipelines", especially when it's changing over time. Rx is all about composing and manipulating such pipelines in a potentially asynchronous context.

But what if the data does not come from an asynchronous source, like Tweets or GUI events, but simple properties objects? Of course, Rx can handle synchronous data as well, but the cost of using it may outweigh the benefits in such a scenario.

In a traditional OOP system, properties of an object are mere values which are often mutable, but not observable by default. To observe the change of a property over time, we must use an observer pattern, usually in the form of a separate event.

Some implementation of Rx allows converting such an event into an Observable. Still, it can be a tedious task to define an event for every such property, and it requires a lot of boilerplate code to convert them into Rx pipelines.

More importantly, the resulting pipelines and their handling code don't have any clear relationship with the original object or its properties from which they got derived. They are just a bunch of statements which you can put anywhere in the project, and that's not how you want to design your business logic in an OOP project.

In a well-designed OOP project, classes form a coherent whole by participating in inheritance hierarchies. They may reference, extend, or redefine properties of their parents or associated objects to express the behaviours and traits of the domain concepts they represent.

And there is another problem with using Observables in such a manner. Once you build an Rx pipeline, you can't retrieve the value flowing inside unless you write still more boilerplate code to subscribe to the stream and store its value to an outside variable.

This practice is usually discouraged as an anti-pattern, and so is the use of Subjects. However, the object is not observing some outside data (e.g. Tweets) but owns it, which is one of the few cases where the use of a Subject can be justified. In such a context, it's perfectly reasonable to assume that an object always has access to a snapshot of all the states it owns.

So, what if we can define such state data of an object as Observables which can also behave like ordinary properties? Wouldn't it be nice if we can easily access them as in OOP while still being able to observe and compose them like in Rx?

And that is what this project is trying to achieve.

Usage

To achieve the goal outlined in the previous section, we provide a way to define a property which can also turn into an Observable. And there are two different types of such property classes you can use for that purpose.

Reactive Property

Firstly, there is a type of property that can manage its state, which is implemented by ReactiveProperty[T] class. To define such a property using an initial value, you can use a helper function from_value as follows:

from alleycat.reactive import functions as rv

class Toto:

    # Create an instance of ReactiveProperty[int] with an initial value of '99':
    value = rv.from_value(99) # You know the song, don't you? 

Sometimes, you need to determine the initial value by referencing another value supplied as a constructor argument. In that case, you can lazily initialize the property by using new_property as shown below:

from alleycat.reactive import functions as rv

class MyClass:

    # Declare an empty property first.
    value = rv.new_property()

    def __init__(self, init_value: int):

        # Then assign a value as you would do to an ordinary property.
        self.value = init_value

Whichever way you choose, it can be read and modified like an ordinary property. If you want to make it a read-only property, you can set the read_only argument to True as below:

from alleycat.reactive import functions as rv

class ArcadiaBay:

    writeable = rv.from_value("life is strange")

    read_only = rv.from_value("the past", read_only=True)


place = ArcadiaBay()

print(place.writeable) # "life is strange"
print(place.read_only) # "the past"

place.writeable = "It's awesome"

print(place.writeable) # "It's awesome"

place.read_only = "Let me rewind." # Throws an AttributeError

# Of course, you can't change the past. But the game is hella cool!

But haven't we talked about Rx? Of course, we have! And that's the whole point of the library, after all.

To convert a reactive property into an Observable of the same type, you can use observe method like this:

from alleycat.reactive import functions as rv

class Nena:

    ballons = rv.from_value(98)


nena = Nena()

luftballons = []

rv.observe(nena.ballons).subscribe(luftballons.append) # Returns a Disposable. See Rx API.

print(luftballons) # Returns [98].

nena.ballons = nena.ballons + 1

print(luftballons) # [98, 99] # Ok, I lied before. It's about Nena. not Toto :P

If you are familiar with Rx, you may notice the similarities between ReactiveProperty with BehaviorSubject. In fact, the former is a wrapper around the latter, and observe returns an Observable instance backed by such a subject.

To learn about all the exciting things we can do with an Observable, you may want to read the official documentation of Rx. We will introduce a few examples later, but before that, we better learn about the other variant of the reactive value first.

Reactive View

ReactiveView is another derivative of ReactiveValue, from which ReactiveProperty is also derived (hence, the alias of functions module used above, "rv").

The main difference is that while the latter owns a state value itself, a reactive view reflects it from an outside source specified as an Observable. To create a reactive view from an instance of Observable, you can use from_observable function like this:

import rx
from alleycat.reactive import functions as rv

class Joni:

    big = rv.from_observable(rx.of("Yellow", "Taxi"))

If you are familiar with Rx, you may see it as a wrapper around an Observable, while a reactive property can be seen as one around a Subject.

Like its counterpart, you can initialize a reactive view either eagerly or lazily. In order to create a lazy-initializing view, you can use new_view, and later provide an Observable as shown below:

import rx
from alleycat.reactive import functions as rv

class BothSides:

    love = rv.new_view()

    def __init__(self):
        self.love = rx.of("Moons", "Junes", "Ferris wheels")

It also accepts read_only option from its constructor (default to True, in contrast to the case with ReactiveProperty) setting of which will make it 'writeable'.

It may sound unintuitive since a 'view' usually implies immutability. However, what changes when you set a value of a reactive view is the source Observable that the view monitors, not the data itself, as is the case with a reactive property.

Lastly, you can convert a reactive property into a view by calling its as_view method. It's a convenient shortcut to call observe to obtain an Observable of a reactive property so that it can be used to initialize an associated view.

The code below shows how you can derive a view from an existing reactive property:

from alleycat.reactive import functions as rv

class Example:

    value = rv.from_value("Boring!") # I know. But it's not easy to make it interesting, alright?  

    view = value.as_view()

Transformation

As we know how to create reactive properties and values, now it's time to learn how to transform them. Both variants of ReactiveValue provides map method, with which you can map an arbitrary function or lambda expression over each value in the pipeline:

from alleycat.reactive import functions as rv

class Counter:

    word = rv.new_property()  

    count = word.as_view().map(len).map(lambda c: f"The word has {c} letter(s)")

counter = Counter()

counter.word = "Supercalifragilisticexpialidocious!"

print(counter.count) # Prints "The word has 35 letter(s)". Wait, did you actually count that?

But what if you want to use other Rx operators, like scan or merge? In that case, you can use pipe just like you would with an Observable instance:

from rx import operators as ops
from alleycat.reactive import functions as rv

class CrowsCounter:
    animal = rv.new_property()

    crows = animal.as_view().pipe(
        ops.map(str.lower),
        ops.filter(lambda v: v == "crow"),
        ops.map(lambda _: 1),
        ops.scan(lambda v1, v2: v1 + v2, 0))

counting = CrowsCounter()

counting.animal = "cat"
counting.animal = "Crow"
counting.animal = "CROW"
counting.animal = "dog"

print(counting.crows) # Returns 2.

On a side note, the aggregation performed by crows in the above example reports the correct value even when there is no explicit subscription in the code. It is because all properties and views in this library are published as 'hot' Observables by default.

It was a design decision to make aggregation more intuitive and less error-prone. You don't have to understand such an implementation detail unless you come across a problem. However, if you have a use case where it would be either necessary or significantly efficient to make reactive values 'cold', please feel free to open a feature request.

Operators

Sometimes, you may need to combine several properties of an object to derive another of its attribute. For example, you can merge two or more reactive values to define a new one which emits a value whenever one of its sources does, as follows:

from alleycat.reactive import functions as rv

class Fixture:
    cats = rv.new_property()

    dogs = rv.new_property()

    pets = rv.merge(cats, dogs)

pets = []

fixture = Fixture()

rv.observe(fixture, "pets").subscribe(pets.append)

fixture.cats = "Garfield"
fixture.cats = "Grumpy"
fixture.cats = "Cat who argues with a woman over a salad bowl"  # What was his name?

fixture.dogs = "Pompidou"  # Sorry, I'm a cat person so I don't know too many canine celebrities.

print(pets) # The array contains all of the names mentioned above.

Imagine you have a Rectangle class which declares its width and height as reactive properties. It would be nice if you can somehow define its area property based upon the rectangle's dimension in a declarative manner.

However, merely merging width and height properties won't automatically calculate the size of the shape. In this case, you can use combine_latest to calculate the size with the latest values of width and height, whenever either of them is changed:

from alleycat.reactive import functions as rv

class Rectangle:
    width = rv.new_property(100)

    height = rv.new_property(200)

    area = rv.combine_latest(width, height)(lambda v: v[0] * v[1])

rectangle = Rectangle()

print(rectangle.area) # Prints 20,000... you do the math!

rectangle.width = 150

print(rectangle.area) # Prints 30,000.

rectangle.height = 50

print(rectangle.area) # Prints 750.

There's another operator called zip whose semantic matches that of the API with the same name in Rx. In fact, you can use most of the operators Rx provides by applying them over Observables extracted by combine method. For instance, you can rewrite the above example code with combine_latest operator provided by Rx as follows:

import rx
from alleycat.reactive import functions as rv

class Rectangle:
    width = rv.new_property(100)

    height = rv.new_property(200)

    area = rv.combine(width, height)(rx.combine_latest).map(lambda v: v[0] * v[1])

Install

The library can be installed using pip as follows:

pip install alleycat-reactive

License

This project is provided under the terms of MIT License.