Go beyond Goroutines: introducing the Reactive paradigm
samuelberthe.substack.com
samuelberthe.substack.com
Meanwhile the samber/ro example is incomplete (Subscribe(...)?), and the source includes some weird stuff: https://github.com/samber/ro/blob/22b84c8296c01c4085e8913944...
Not to mention heaps of reflection and panics: https://github.com/samber/ro/blob/22b84c8296c01c4085e8913944...
The functionality and expressiveness might be fantastic, but I would evaluate it very carefully before use.
You can see this your self if you edit the markup in your browser’s inspector and add `contenteditable` attribute to the surrounding <pre>, then navigate down a line... it will jump forward just by slightly less then a column per indent level.
Libraries that enable terse seemingly magical things tend to just hide a lot and make code harder to read. You end up having to become an expert in what amounts to a DSL on top of the language.
Edit: I've maintained a full codebase with R2DBC and I can assure you many developers came scratching their heads to me sometimes on tell me why are we doing this again when we can only have finite connections to DB and those connections can be 1-1 mapped to worker threads.
Speaking of the code examples, I am not convinced at all. The supposedly bad example is idiomatic and understandable to me, and I have a much better idea of what's going on than in the supposedly good examples. It contains the kind of constructions I expect to see when working on a Go codebase, and it will be easier to modify correctly for another Go developer coming in later. Please, work with the language instead of forcing it to wear a programming paradigm it doesn't like.
Such as:
> Together, these tools make Go perfect for microservices, real-time systems, and high-throughput backends.
Real-time systems?! I have never heard of anyone using Go for realtime systems because of its GC and preemptive scheduler. Seems like the sort of thing an LLM would slip in because it sounds good and nails that 3 item cadence.
> Built on top of Go channels → broken backpressure.
But then the example is about ordering. Maybe I'm being pedantic or missing the specific nomenclature the ReactiveX community uses, but backpressure and ordering are different concerns to me.
Then the Key Takeaways at the end just seems like an LLMism to me. It's a short article! Do we really need another 3 item list to summarize it?
I'm not anti-LLM, but the sameness of the content it generates grates on me.
I don't know. I've seen "realtime" used quite often in a sort of colloquial sense, where it means something fairly different from "hard realtime system" as an embedded systems person might use the term. I think there's a pretty large base of people who use "realtime" to mean something that others might call "near real-time" or just "continually updating" or something along those lines, where's there's no implication of needing nanosecond level predictable scheduling and what-not.
That's not to say that the article isn't AI generated, or whatever. Just that I wouldn't necessarily see the use of the "realtime" nomenclature as strong support for that possibility.
So many Go developers ignore some tools because they consider them "not idiomatic".
But why not use abstractions when available ??? Did we forget to be productive ?
And the smallest possible team is the programmer and their future self.
Even then the hard thing is to predict what will be better for our future selves. Maybe we will be rusty in our Go skills or maybe we will have embraced idiomatic Go, and the approach that makes sense now will require our future self to puzzle through the code.
Of course maybe we will have kept programming the same way because it still feels like the better way and our future self will be efficient. But again that's only for the smallest possible team. If the team expands, then all bets are off.
This is precisely the premise for their library: I don't have the mental context to fit all the boilerplate in, nor do I have the brainpower to sift through it.
Sure, assembly is readable: every line is one machine instruction. But that's way too many details. On the other hand, C++ templates are not enough details.
The way you approach this in Go (and I would argue in any other language) is by building small abstractions when and if it makes sense to do so. Not by introducing abstractions early, or in order to make a piece of code slightly easier to follow. A simple comment might help to explain tricky logic, but otherwise you shouldn't need explanations about core language features.
Abstractions are not free. They can be poorly written, leaky, abstract too much or too little, inflexible, increase overall complexity and thus cognitive load, impact performance, introduce bugs, etc.
So relying on them only when absolutely necessary is often the sensible choice.
Also, if possible, building your own bespoke abstraction is often preferable to using an external package. You can tailor it to your exact use case, without adding another dependency, which carries its own risks and problems.
This specific package seems designed to be syntax sugar over language features, which is not a good idea for future maintainers of the code that uses it. They would need to understand how this 3rd-party library works, even if the author claims it to be more readable, ergonomic, or what have you.
var result := []string{}
for i = 0; i < items.Length(); i++ {
item := items.Get(i)
if item < threshold {
result = append(result, converToString(item))
}
}I more readable than this code:
return items
.filter { it < threshold }
.map { convertToString(it) }
Of course, once you do understand these very complicated concepts, the code above is far more readable: it clearly states WHAT the program does (filtering all values below the threshold and converting them to string) rather than HOW it does that (which nobody cares about). "readability" here is only counted from the narrow perspective of an imperative programmer who is not familiar with functional declarative data processing.
I feel the same about the "ro" examples in the OP. I don't particularly like that ro takes (mostly because Go forces its hand, I assume), like having to put everything in an explicit pipe, but I find the example far more readable than the pure Go example which combines loops, channels and WaitGroups. That's far worse than the loop example I gave in this reply, to be honest, and I really don't know why people say this example is readable. I guess you can optimize it a little, but I always found both channel and WaitGroups waitable, unreadable and error prone. They are only "readable" in narrow perverted sense that has somehow become prevalent in the Go community, where "readability" is redefined to mean: no closures, no immutable values, no generics, no type safety and certainly nothing that smells like FP.
My poor memory seems to recall them gaining traction ~10 years ago, but they've fallen hard off my radar.
My fear with adopting a library like this for Go is actually that it might end up being very unfriendly to the profiler once bottlenecks start occurring.
Brian Goetz even went as far as saying loom is going to kill reactive entirely: https://www.youtube.com/watch?v=9si7gK94gLo&t=1165s
So yes, in this particular case, most of the usages of RxJava, Reactive Streams and particularly Spring Reactor is just there because programmers wanted pretty simple composition of asynchronous operations (particularly API calls) and all the other alternatives were worse: threads were too expensive (as Brian Goetz mentions), but Java did have asynchronous I/O since Java 1.4 and proactor-style scheduling (CompletableFuture) since Java 8. You could use CompletableFutures to do everything with asynchronous I/O, but this was just too messy. So a lot of programmers started using Reactive framework, since this was the most ergonomic solution they had (unless they wanted to introduce Kotlin).
That's why you'd mostly see Mono<T> types if you look at a typical reactive Spring WebFlux project. These Mono/Single monads are essentially CompletableFutures with better ergonomics. I don't mean to say that hot and cold multi-value observables were used at all, but in many cases the operator chaining was pretty simple: gathering multiple inputs, mapping, reducing, potentially splitting. Most of this logic is easier to do with normal control flow when you've got proper coroutines.
But that's not all what reactive frameworks can give you. The cases when I'd choose a reactive solution over plain coroutines are few and pretty niche: to be honest, I've only reached to a reactive 3 or 4 times in my career. But they do exist:
1. Some reactive operators are trivially mapped to loops or classic collection operators (map/reduce/filter/flatMap/groupBy/distinct...). But everything that's time-bound, is more complicated to implement in a simple loop, and the result is far less readable. Think about sample or debounce for instance.
2. Complex operation chains do exist and implementing them as a reactive pattern makes the logic far easier to test and reason about. I've had a case where I need to implement a multi-step resource fetching logic, where an index is fetched first, and then resources are fetched based on the index and periodically refreshed with some added jitter to avoid a thundering herd effect, as well as retries with exponential backoff and predictable update interval ranges which is NOT affected by the retries (in other words: no, you can't just put a delay in a loop). My first implementation tried to model that with pure coroutines and it was a disaster. My second implementation was RxJava, which was quite decent, and then Kotlin Flow came out, which was a breeze.
3. I'm not sure if we should call this "Reactive" (since it's not the classic observable), but hot and cached single values (like StateFlow in Kotlin) are extremely useful for many UI paradigms. I found myself reaching to StateFlow extensively when I was doing Android programming (which I didn't do a lot of!).
In short, I strongly disagree with Brian Goetz that Functional Reactive Programming is transitional. I think he's seeing this issue from a very Java-centric perspective, where probably over 90% of the usage we've seen for it was transitional, but that's not all that FRP was all about. FRP will probably lose its status a serious contender for a general tool for expressing asynchronous I/O logic, and that's fine. It was never designed to be that. But let's keep in mind that there are other languages than Java in the world. Most programming languages support some concept of coroutines that are not bound to OS kernel threads nowadays, and FRP is still very much alive.
Yes, a transitional technology for the _Java Language/Platform_ post Loom. He never said anything about Functional Reactive Programming in general.
Maybe this is what Brian Goetz meant to say, but this is not what he said.
I’ve worked in a few complex projects that adopted reactive styles and I don’t think they made things simpler. It was just as easy to couple components with reactive programming as it was without.
(snark aside, that syntax actually looks pretty nice).