Learning from React series:
- Part 1 - why examining React is useful even if you won't end up using it
- Part 2 - what Facebook wanted to do with React and how to get a grasp on it
- Part 3 (this one) - what is Reactive Programming all about?
- Part 4 - is React functional programming?
- Part 5 - Typescript, for better and for worse
- Part 6 - Single Page Applications are not where they wanted to be
The name React is already declaring that it is used in reactive programming, but what is that? Wikipedia is defining it as "a declarative programming paradigm concerned with data streams and the propagation of change". It expands on that to say that it declares the relationship between elements and updates them when either change. You can easily imagine a graph of elements magically updating as any of them changes. However, the implementation details of that magic matter.
One might argue that Verilog, a hardware description language, is also reactive, as it is based on actions being performed on certain events and it even uses non-blocking assignments, which are like declarations of state change which happen at the same time. Reminds me of the way React is implementing state management.
Of course, reactive programming is also modern UI and when I say modern, I mean everything in the last twenty years. Code gets executed when elements in the user interface change state: on click, on change, on mouse move, on key press etc. That is why, the developers at Facebook argue, browser based UI programming should be reactive at the core. This is not new, it's something you might even be already very familiar with in other contexts. Code that is triggered by events is also called event-driven programming.
But at the same time, others also claim their software is reactive. Microservices are now very fashionable. The concept revolves around organizing your product into completely independent modules that only have one external responsibility, which then one wires together via some sort of orchestrator. The biggest advantage of this is obviously separation of concerns, a classic divide and conquer strategy generalized over all software, but also the fact that you can independently test and deploy each microservice. You don't even have to shut down running ones or you can start multiple instances, perhaps with multiple versions and in different locations. This is also distributed programming. The way the communication between microservices is done is usually via some sort of message queue, like Rabbit MQ, but I am working on a really old software, written like 15 years ago, which uses IBM MQ to communicate between different portions of the software - let's call them macroservices :) Well, this is supposed to be reactive programming, too, because the microservices are reacting to the messages arriving on the queue and/or sent from others.
The observer pattern is old, it's one of the patterns in the original design patterns book Design Patterns: Elements of Reusable Object-Oriented Software, which started the software design pattern craze which rages on even now. Anybody who ever used it extensively in their software can (and many do) claim that they did reactive programming. Then there is something called the actor model (which will probably confuse your Google if you search for it), which is actually a mathematical concept and originated in 1973! Implementations of actors are eerily similar to the microservices concept from above.
And speaking of events, there is another pattern that is focusing on declaring the flow of changes from a given state, given an event. It's called a state machine. It also boasts separation of concerns because you only care about what happens in any state in case of an event. You can visualize all the possible flows in a state machine, too, as names arrows from any state to another, given that such a transition is defined. The implementation of the state machine engine is irrelevant as long as it enables these state transitions as defined by the developer.
And that's the rub. The concept of reactivity is subjective and generally irrelevant. The only thing that changes and matters is the implementation of the event transport mechanism and the handling of state.
In a traditional imperative program we take for granted that the execution of methods will be at the moment of the call and that all methods on that thread will be executed one after the other and that setting a value in memory is atomic and can be read immediately after by any other piece of code and you can even lock that value so it is only read by one entity at a time. Now imagine that you are writing the same program, only we can't make the assumptions above. Calling methods can result in their code getting executed at an arbitrary time or maybe not at all. Whatever you change in a method is only available to that method and there is no way for another method to read the values from another. The result? Your code will take a lot of care to maintain state locally and will start to look more like a state machine, modelling transitions rather than synchronous flows. The order of operations will also be ensured by consuming and emitting the right sort of events. Permanent and/or shared storage will become the responsibility of some of the modules and the idea of "setting data" will become awkward. Keeping these modules in sync will become the greatest hurdle.
That's all it is! By eliminating assumptions about how your code is executed, the result is something more robust, more generic, more compartmentalized. Is it the golden hammer that will solve all problems? Of course it isn't. We've seen how the concepts at the core of reactive programming have been there since forever. If that was the best way, everybody would already be working like that. The biggest problems of this kind of thinking are resource duplication, as everybody has to keep all the data they use locally, and synchronization, as one cannot assume there exists any source of absolute truth that can be accessed by all at the same time. Debugging the system also becomes a bit complicated.
In 2014 a bunch of people created something called "The Reactive Manifesto" and anyone can sign it. In it, they define a reactive system as:
- Responsive - responds quickly
- Resilient - remains responsive in case of failure
- Elastic - remains responsive regardless of workload
- Message Driven - async message passing as the boundary between components
As you can see, reactive mostly means responsive for them and the concept is more one of organization management than a specific set of programming languages and tools. Note that, in fact, you can create a system that communicates via async message passing between components on the client, on the server, between client and server and so on. There can be multiple types of technology and different stacks. As long as they can react to asynchronous messages, they can be integrated into such a system.
This post has reached already a big size and I haven't even touched on functional programming and how it tries to solve... well, everything. I will do that in the next chapter. I have to say that I find the concept of a programming language that eliminates global variable scope and public fields and introduces a delay and a random order of execution of methods or properties from other classes fascinating. Imagine testing and debugging that, then moving the working code to production, where the delay is removed. You will also see that a lot of the ideas above influence how React development is done and perhaps you will understand purists telling everybody how things are not correct until you implement this or that in a certain way. Till next time!