Rephrased: ocaml is so flexible that async can be implemented as a library with no special support from the language.
This is the beauty of ocaml (and strongly typed functional languages more broadly)
I don't think that's anything specific to strongly typed functional languages. In eg Rust even the standard library relies on third party crates.
Though it is still somewhat amusing to me that loops in Haskell are delivered via a third party library, if you ever actually want them. See https://hackage.haskell.org/package/monad-loops-0.4.3/docs/C...
I do agree that it's good language design, if you can deliver what would be core functionality via a library.
Whether you want to integrate that library into the standard library or not is an independent question of culture and convenience.
(Eg Python does quite well with its batteries-included approach, but if they had better dependency management, using third party libraries wouldn't be so bad. It works well in eg Rust and Haskell.)
Clojure has core.async, which implements "goroutines" without any special support from the language. In fact, the `go` macro[1] is a compiler in disguise: transforming code into SSA form then constructing a state machine to deal with the "yield" points of async code. [2]
core.async runs on both Clojure and ClojureScript (i.e. both JVM and JavaScript). So in some sense, ClojureScript had something like Golang's concurrency well before ES6 was published.
[1] https://github.com/clojure/core.async/blob/master/src/main/c...
[2] https://github.com/clojure/core.async/blob/master/src/main/c...
That's wildly overselling it. Closure core async was completely incapable of doing the one extremely important innovation that made goroutines powerful: blocking.
(let [c (chan)]
;; creates channel that is parked forever
(go
(<! c)))
c := make(chan interface{})
// creates goroutine that is parked forever
go func() {
<-c
}()Call a function that blocks in go, the routine will park. Do that in Clojure and the whole thing stalls.
There is a caveat where Java code using `synchronized` will pin a carrier thread, but this has been addressed in recent versions of Java.[1]
Because JavaScript runs an event loop on a single thread. It's akin to using GOMAXPROCS=1 or -Dclojure.core.async.pool-size=1 . There may be semantic differences depending on the JavaScript engine, but in the case of web browsers, the only function that could possibly stall the event loop is window.alert. As for Node.js, one would have to intentionally use synchronous methods (e.g. readFileSync) instead of the default methods, which use non-blocking I/O.
When using core.async in ClojureScript, one could use the <p! macro to "await" a promise. So there is no use for Node.js synchronous methods when using core.async (or using standard JavaScript with async/await).
I would call this "making use of the platform" rather than "special support from the language". The core.async library does not patch or extend the base language, and the base language does not patch or extend the platform it runs on.
I found F# to be more pleasant to work with async than C# (which is already a breeze). It is true that you still have to define 'task' (or 'async' if you want to use Async CEs) but it is generally there for a reason. I don't think it's too much noise:
let printAfter (s: float<second>) = task {
let time = TimeSpan.FromSeconds (float s)
do! Task.Delay time
printfn $"Hello from F# after {s} seconds"
} let async printAfter (s: float<second>) =
let time = TimeSpan.FromSeconds (float s)
await Task.Delay time
printfn $"Hello from F# after {s} seconds"
'async'-annotated methods in C# enable 'await'ing on task-shaped types. It is bespoke and async-specific. There is nothing wrong with it but it's necessary to acknowledge this limitation.
let!, and!, return!, etc. keywords in F# are generic - you can build your own state machines/coroutines with resumable code, you can author completely custom logic with CEs. I'm not sure what led you to believe the opposite. `await Task.Delay` in C# is `do! Task.Delay` in F#. `let! response = http.SendAsync` is for asynchronous calls that return a value rather than unit.
In the same vein, seq is another CE that is more capable than iterator methods with yield return:
let values = seq {
// yield individual values
for i in 1..10 -> i
// yield a range, merged into the sequence
yield! [11..20] // note the exclamation mark
}
Reference: https://learn.microsoft.com/en-us/dotnet/fsharp/language-ref...
I would disagree. If you need to have a bespoke set of syntax, then something is not integrated where it should be. The language design should not be such that you are writing things differently, depending on the paradigm that you're handling. That's not something that occurs in every language, so it isn't essential that it exists.
We can acknowledge the differences in a way that alerts the programmer, without forcing the programmer to switch syntaxes back and forth when moving between the paradigms. async/await is one method, Promises another, etc. A different syntax is a much, much higher cognitive load.
Support of asynchronous code and of its composition is central to C#, which is why it does it via async/await and Task<T> (and other Task-shaped types). Many other languages considered this important enough to adopt a similar structure to their own rendition of concurrency primitives, inspired by C# either directly or indirectly. Feel free to take issue with the designers of these languages if you have to.
F#, where async originates from, just happens to be more "powerful" as befits an FP language, where resumable code and CEs enable expressing async in a more generalized fashion. I'm not sold on idea that C# needs CEs. It already has sufficient complexity and good balance of expressiveness.
Do-notation-like 'await' is not for calling functions, it is for acting on their return values - to suspend the execution flow until the task completes.
However, the compiler does not, has never, required that it does things via a different syntax. In fact, in the early branches before that was adopted, it didn't! The same behaviour was seen in those branches. This behaviour you expect, was never something that had to be. It was something chosen to simplify the needs of the optimiser, and in fact cut the size of code required in half. It was to reduce the amount of code needed to be maintained by the core team. And so 1087 [1] was accepted.
So perhaps you might need to read more into the process of the why and how async was introduced into C# and F#. It was a maintenance team problem. It was a pragmatic approach for them - not something that was the only way that this became a possibility.
As said, in the original branch for using tasks...
> Having two different but similar ways of creating asynchronous computations would add some cognitive overhead (even now there are times when I am indecisive between using async or mailbox for certain parallelism/concurrency scenarios). [0]
[0] https://github.com/fsharp/fslang-suggestions/issues/581
[1] https://github.com/fsharp/fslang-design/blob/main/FSharp-6.0...
However, this is where our opinions differ. I like task CE (and taskSeq for that matter too). It serves as a good and performant default. It's great to be able to choose the exact behavior of asynchronous code when task CE does not fit.
Simple things like the maybe and either monad are often clearer in this notation. Complex things like alternatives to async (such as CSP derived message passing concurrency), continuations, parser combinators, non-determinism, algebraic effects and dependency tracked incremental computations are naturally modeled with this same machinery, with CE notation being a kind of super helpful DSL builder that makes certain complex computations easier to express in a sequenced manner.
If the custom syntax was only for async you'd have a point, but the general power of the framework make it the more preferable approach by far, in my opinion.
We shoehorn things that feel like, but are structurally different, to DSLs into config files and the like, using JSON/YAML/etc in rough ways, because DSLs introduce a cognitive overhead that doesn't need to be there.
That the shoehorn happens, does mean that DSLs are something natural to reach for. You're right there. But that we have moved away, as an industry, indicates that using any kind of DSL is a smell. That there probably is a better way to do it.
Having a core language feature using a DSL, is a smell. It could be done better.
It's early days in that regard, with some folks doing some really interesting things: Odersky himself / the Ox project.
Gleams a nice one too if you want pure functional, small community though.