I don't get the author's conclusion section, especially "structured concurrency does not suffer the colored function problem".
Structured concurrency has nothing to do with function colouring. It is how you organise and spawn tasks, so that you can have certain guarantees e.g. the group responds to cancellation as you'd expect, etc.
I don't see how async/await plays a part in this. You could have structured concurrency with a os threading model, as opposed to a userland threading model.
You can argue that the ergonomics of Zig's async story needs better documentation, especially if you're playing with function pointers. But please don't mislead readers that structured concurrency solves this.
> But please don't mislead readers that structured concurrency solves this.
It absolutely does. The reason is that, in structured concurrency, when a function spawns a thread, you can expect that that thread is joined before the function returns.
For callers, this means that they don't have to worry about whether the function they are calling spawns threads or not; they just call the function, and the function does its thing.
It is the same principle as standard control flow constructs: they use `goto` underneath, but you don't need to worry about the use of `goto` because they guarantee that execution will always end up at the same place afterward (barring early returns and things like that).
So when I say that structured concurrency does not suffer from the colored function problem, what I am saying is that callers, and programmers, do not need to care what functions do underneath when they are called.
Also, structured concurrency solves the function coloring problem because you can use any function, whether it spawns threads or not, as first-class functions, or function pointers, and it all works without needing any special code, or work, or compiler trickery.
> So when I say that structured concurrency does not suffer from the colored function problem, what I am saying is that callers, and programmers, do not need to care what functions do underneath when they are called
Thanks for clarifying what you meant with your original comment in the article. I completely agree that this is a very big advantage, as opposed to not using structured concurrency.
However, it seems that you are conflating two things with this assertion. For example you say:
> The reason is that, in structured concurrency, when a function spawns a thread, you can expect that that thread is joined before the function returns
While structured concurrency gives you this guarantee, this is orthogonal to whether a function the thread runs is "red" or "blue".
"Red" functions exist in a language because you want blocking to happen on userland so it's cheap to unschedule the task that is blocking.
Structured concurrency doesn't remove "red" functions from a language. They do make the ergonomics of using such functions much easier on the caller e.g. python's trio library [0].
To actually "remove Red functions from the language" you can make the user think they're just using plain "blue" function techniques for blocking.
Go achieves this by only allowing use userland threading.
Java's project Loom [1] achieves this by making common suspension/yielding points (e.g. IO) compatible with both userland and OS threading.
To me, this is what solves the colouring problem i.e. you only have one colour and keep using and writing your code as you always have been.
Structured concurrency, much like you said, is the added bonus that helps reasoning about the lifecycle of tasks a function may or may not spawn.
I was going to say that this isn't quite right, that you could implement structured concurrency at the language level and have a compiler know how to thread the needle between the different "colors".
But then I realized this is exactly what Go does (with the scheduler intercepting system calls), just minus the structured concurrency point, strengthening your point that these are orthogonal.
Go solves the coloring issue without structured concurrency, that's true. But that statement says nothing about whether structured concurrency can or cannot solve the coloring issue, and if it could, then they're not orthogonal.
I think OP saying that structured concurrency solves the coloring problem is a very similar argument to the authors of Zig saying Zig is colorblind; it might be true for a portion of the userbase, mainly those who are callers, and mostly not for those who are library implementers.
I also think that the structured concurrency feature of "you can expect that the {thread, task, fiber, goroutine} is joned before the function returns" is weaker for the purpose of colorblindness than "you can expect that the {thread, task, fiber, goroutine} is joined before the next instruction is executed" which is what I think you'd need as a single tool for colorblindness
> While structured concurrency gives you this guarantee, this is orthogonal to whether a function the thread runs is "red" or "blue".
I disagree with this.
Making sure that threads join before return is a big reason why I can pass functions around easily as function pointers and not worry about it.
An example is in my up-and-coming build system. [1]
In that code, I have a function that opens a threadset (my term for trio's nurseries) and executes a build. Once my build system is out of alpha, the direct call to that function will be replaced by a function pointer to the build type of the user's choice. (For example, you might want a "quick build" to just rebuild a file saved by your editor, or a "full build" regardless of what is already built, or just a normal build.)
Despite using function pointers, I don't need to worry about whether threads are spawned or not. In Zig, you do need to worry about if a function is async or not when using function pointers.
This includes if I make a build type whose function will not spawn threads at all, by the way. (For example, if building on a platform where you don't have the memory to spare for extra threads.) I can use that build type's function as a function pointer the same way I would use any function that does use a threadset.
In other words, while Zig is colorblind at compile time, structured concurrency is colorless at both compile time and runtime.
Isn't structured conc. about using scopes (, syntax) to provide concurrency "control flow"?
If so, then it does solve the color problem. The alternative to structural (ie., scopy-syntaxy) flow primatives, is to reuse existing control flow primtives (ie., function calls) and just layer on coloring.
By declining to reuse functions this way, you are dropping the need for color.
I may be totally wrong with this assumption, but the way I understoog Zig's color-less async support is that the compiler either creates a "red" or "blue" function body from the same source code based on how the function is called (so on the language level, function coloring doesn't matter, but it does in compiler output).
The compiler still needs to stamp out colored function bodies because the generated code for a function with async support needs to look different - the compiler needs to turn the code into a state machine instead of a simple sequence).
It's a bit unfortunate that red and blue functions appear to have a different "ABI signature", but I guess that's needed to pass an additional context pointer into a function with async support (which would otherwise be the implicit stack pointer).
That's exactly it. It just enables code reuse. You still have to think about how your application will behave, but you won't have to use an async-flavored reimplementaion of another library. Case in point: zig-okredis works in both sync and async applicatons, and I don't have to maintain two codebases.
I thought using "colorblind" in the title of my original blogpost would be a clear enough hint to the reader that the colors still exist, but I guess you can never be too explicit.
This is all correct. I don't think it needs to be this way though. It could plausibly work in a way that doesn't make async functions special other than that the compiler must calculate the size of the deepest stack they can use, and then TFA's code would work rather than not working. (Well, other than the thing where the author wanted to make a function that suspends only once, then call it, then resume it, the await it, which will never work by design)
Pretty sure stack growth knowledge like this is on the Zig roadmap. Stack memory is a free-for-all in most languages, but in many domains, it's as crucial to manage as the heap.
A similar type of information is already recorded by the compiler for async functions to figure out how large their frames must be. So the proposed change would change the size of some frames but otherwise do nothing to most code.
Cheers. I must admit I've completely ignored the async aspect of Zig for the time being, so I know nothing about how close or far it is from already doing enough of this work. Thanks for the info.
This article was my first exposure to async in Zig, and I find it rather at odds with the "No hidden control flow" selling point of the language.
fn foo() void {
bar();
std.log.info("bar is finished");
baz();
std.log.info("baz is finished");
}
I cannot just look at this function and know what it's doing and how to use it. If bar()'s implementation contains a suspend or await keyword then this won't compile when called from main() unless its call is prefixed with async, and even if I do so then I also have to make sure I resume foo enough times for bar (or baz) to finish too. Am I missing some key detail here? It seems out of place compared to many other language features.
fn foo() void {
bar();
std.log.info("foo() is finished", .{});
}
It looks dead simple, but it's a pain in the butt to use because it's actually async, and you can know neither that nor how to get this function to actually finish without looking inside of bar().
You can't know this function will actually finish without looking inside of bar() in any Turing Complete language, because this is the Halting Problem.
Not in a trivial way either. A good async language will never let std.log.info print until/unless bar resumes.
The difference between a "blue foo" and a "red foo" is also subtler than we pretend it is. Both will finish eventually, unless they don't, and both may be paused for as long as the OS wants in any preëmptive operating system.
Ok, sure, there are a whole bunch of reasons that my function might not actually finish. In this case, though, it's specifically not finishing due to invisible compile-time additions Zig is making to the function I've written.
They could instead require me to explicitly write `await bar();` within foo(), and I could find comfort in the "no hidden control flow" tenet of the language once more. I recognise now that this would break the colour-blindness of their async implementation and that this is a pretty significant trade-off, but my perspective remains that this breaks one of the rules they've laid out on their homepage.
No, you can't, not without knowing what is inside of bar().
bar() could contain an infinite loop. bar() could call rand() and try to reference the null pointer 1 in n times. bar() could issue forth nasal demons to haunt you. bar() could call os.exit. bar() could recurse until the stack is exhausted. bar() could suspend and not resume. The OS could kill bar() from outside before it returns.
Who knows what shadows lurk within a function's stack frame?
Run it and find out.
The invariant which needs to be preserved here is that if bar() returns, the log will print, and equally important, if bar does not return, the log will not print.
That's what a good async system can preserve without knowing what's inside bar.
Within main(), your snippet will give you an error at compile time "function with calling convention 'Inline' cannot be async". In my snippet foo() is invoked in an async context using the async keyword, and when foo() suspends (due to it awaiting bar()) control will be returned to main(). If you change your first line from
foo();
to
_ = async foo();
Then your program will compile and your log is not going to be prevented by Zig's async shenanigans.
That seems to be the same kind of confusion I ran into the last time I tried to do async in Zig. I also had trouble figuring out who was responsible for keeping track of async frames.
The details are not fresh, but it went something like: if a function called an async function like so `the_frame = async foo();`, you couldn't really reliably `resume the_frame;` unless you were responsible for all suspend points hit inside `foo`. If `foo` tried to call some other function `bar` you had no idea had a suspend point, you couldn't resume the frame anymore since you'd need to resume the inner frame instead.
So, yeah, they were colorless as in you could call any function you wanted and they don't advertise suspend points or async usage, but they are actually colorful inside and calling the wrong color breaks things.
On a cursory look at the problematic examples, this seems to be an issue with Zig’s type system. There are async functions and non-async functions, and they have the same syntax, but they have what should be different inferred types. But Zig allows pointers to both to be stored in the same array without an obviously unsafe cast, and the result doesn’t work.
ISTM this could be fixed in the type system, possibly at the cost of breaking some existing working code.
There seems to be a defect in the type inference. In TFA's programs, using (but not calling) red gets you a value of the wrong type. Adding an explicit callconv(.Async) to red causes you to get this error:
./src/main.zig:42:32: error: expected type 'fn(*u64) void', found 'fn(*u64) callconv(.Async) void'
const fns = [2]@TypeOf(blue) { red, blue };
After that, changing the array to be an array of @TypeOf(red) gets you a working program. That is, it's apparently fine to store blue in a value of red's type and use blue with @asyncCall as if it was red's type at runtime, but trying to store red in a variable of blue's type is a compile error.
Hmm, seems like it’s maybe not in type inference per se. It seems more like the compiler actually believes that red (used as a function pointer) is not async! So it has the wrong type in @TypeOf and it also coerces itself with no warning to that wrong type. Ouch.
The docs are light on details, but the compiler should figure out that red is async, and it does if you call it, but not if you take a pointer. I would describe this as a defect! We don't actually disagree about any facts though, just labels.
Sounds about right. Zig couldn’t claim to be colorless anymore, but this seems like an unsound type system to me. Still think Zig‘s approach is a step into the right direction. Is the JVM the only remaining runtime that properly fixes the issue?
> But I have not needed to prove it because (as far as I know) only one programming language that uses async as its concurrency model claims to not have function colors.
The zig "colourless" innovation was informed by my work with cqueues in Lua[1]; which I gave as an example of good colourless-ness.
This appears to redefine the concept of function colors to something different and not particularly problematic.
The issue that the function color concept illustrates is how subtle function incomposibility can seem perfectly workable for a while, but sometimes leads you to a deep dead end that takes a lot of reworking of otherwise good code to resolve.
The key limitation is that “red” functions can only be called from other “red” functions. So that if you’ve developed a bunch of code composed of “blue” functions, possibly with deep, complex call chains, and then find you need to call a “red” function at a low level, you suddenly find yourself in a dead end with no good way forward.
This article seems to drop that key point from function colors. (I say seems because it only speaks of its own definition indirectly. Maybe it’s in the deleted post mentioned at the start?)
I think that makes this whole article pretty much pointless.
The article would be much better if you dropped mentioning function colors at all then, since your definition really has little to do with the original concept.
As-is, bringing it up just muddies your argument.
Critically, function colors are an issue of the language design and here you’re finding limitations in the language implementation.
> The article would be much better if you dropped mentioning function colors at all then, since your definition really has little to do with the original concept.
I heavily disagree, but near the end of the post, I also show that Zig does fit the original definition:
> Third, async is, in fact, viral. The language reference says,
>> Zig infers that a function is async when it observes that the function contains a suspension point. Async functions can be called the same as normal functions. A function call of an async function is a suspend point.
> (Emphasis added.)
> In other words, if you call an async function, your function gets a suspend point, which means that your function becomes async.
> In fact, if you look at Bob Nystrom’s point 3, “You can only call a red function from within another red function,” this means that Zig fits point 3!
> What Zig does differently than Bob Nystrom’s “hypothetical” language is that if it sees you call a red function from a blue function, it shrugs its shoulders, makes the blue function red, and does not tell you. This gives the appearance of being able to call red functions from blue functions while still only allowing red functions to call red functions.
That’s just changing the definition of function colors. In the original definition the fundamental distinction between functions of different colors is which colors can call which other colors.
In zig async functions can call async and non-async functions, and non-async functions can call async and non-async functions. Thus, no “color” distinction.
It’s just pointless to ascribe the existing terminology of function colors while fundamentally changing the concept to something else. You end up making no point at all.
Then you and I disagree about the definition of function colors.
But Zig's language reference is clear: when needed, blue functions are turned into red functions transparently at compile time. What that does is make it seem as though there are no colors. That's why the dichotomy can be exposed at runtime: the compiler can't intervene and turn blue functions into red functions.
No matter how much you say that changes the definition, it still means that blue functions cannot call red functions. The fact that functions are turned into red functions does not magically make the colors disappear; it just hides their existence until it cannot be hidden anymore.
You omit the most important point about the definition of function colors. Point 3. You can only call red functions from other red functions. That's the only problem with `async` functions.
Now, I don't know much about Zig's async story to comment on this post, but this is exactly my problem with Zig's async.
Zig's top self-promoted feature is language simplicity. "You don't have to debug your knowledge about the language". But with async, that's exactly what I would have to do. I tried to read the articles about Zig's async several times but the idea never clicked in my head.
Besides where a sibling comment says I mentioned it, I also mentioned it near the end:
> Third, async is, in fact, viral. The language reference says,
>> Zig infers that a function is async when it observes that the function contains a suspension point. Async functions can be called the same as normal functions. A function call of an async function is a suspend point.
> (Emphasis added.)
> In other words, if you call an async function, your function gets a suspend point, which means that your function becomes async.
> In fact, if you look at Bob Nystrom’s point 3, “You can only call a red function from within another red function,” this means that Zig fits point 3!
> What Zig does differently than Bob Nystrom’s “hypothetical” language is that if it sees you call a red function from a blue function, it shrugs its shoulders, makes the blue function red, and does not tell you. This gives the appearance of being able to call red functions from blue functions while still only allowing red functions to call red functions.
That's interesting. So this means that the only thing Zig does is auto-insert the 'async' qualifiers in front of functions instead of requiring you to insert them?
As far as I understand it, yes, as well as, insert an `await` right after, turning the enclosing function itself `async`, which means that anything that calls the enclosing function (the caller) would also need to have `async` and `await` inserted, and the process would keep going.
That is what I mean by viral: if a function low in the call chain is turned `async`, every function above it is too.
> My definition, however, does not fit point 3. This is on purpose; from the first time that I read Bob Nystrom’s post, I never believed that point 3 was a prerequisite for function colors.
> This difference of opinion is probably what caused most of the conflict I alluded to earlier, so I thought that I would lay it out, then make my case for it, along with the case that Zig has function colors according to my definition.
You're right, but it's not like they don't address that.
I'm trying to grasp the whole fuss about the colors of the functions still, and yes, I did read all the related posts back then, and skimmed them today as a refresher.
You can call an async function in a sync context. You just can't unwrap the value from the result container, as it's not available. But you do get a result, which is Promise<T>.
Any language that has concurrency, pointers or any mechanism that allows the value to change from under your feet has colored functions, doesn't it?
Also IMHO: Not being able to use "await" or not depending on the context, and wrong usage being a parse error, is much better than black magic happening at the compilation.
Putting aside details, I would like to read something about Zig that addresses both the "no hidden control flow" claim and the "colorblind async/await" claim.
On the face of it those 2 things seem incompatible. Why are they compatible?
If Zig followed the C philosophy, then your async code would basically look like nginx, i.e. state machines and callbacks. Not saying this is desirable, but it does work (for some definition of "work").
> This post is not here to ascribe any intention to Zig proponents, especially the leadership and employees of the Zig Software Foundation. I did that once, and I regret it.
Maybe we should respect his implied wishes to not go there again, since he's clearly trying his best for this blog post to be useful constructive critique?
What point? It's the reason why I asked. You were implying I want to read the other post to judge the author, I am explaining that it didn't even cross my mind and I am miles away from that intention.
I didn't imply anything about you, I just reminded you that it is the author's wish that we don't dig that up. I see no reason to not respect that request
That article mostly calls out some Rust-vs-C comparison points about a specific Python library that were honestly reasonably interesting to read.
The Zig part has a number of updates and clarifications that contribute helpful balance while not disrupting the original text. Quite a few of the points have ended up being superceded by developments and evolution in the language.
Ok, thanks for the clarification! I won't but at the same time I think everyone can make mistakes, despite the unforgiving nature of written digital records.
Didn't really get much out of the blog post. Seems obvious to me that written zig is colorless but compiled zig is not. Except for a few use cases that the zig authors pointed out.
Is that really much different than how python abstracts away types?
This is a worthwhile post, and I hope to have more to say about it soon. For now: It seems like we need some other terminology to refer to this group of issues, not the terminology that we use to refer to another much more annoying group of issues. This group of issues is similar to the group of issues you would face if you wanted to dynamically decide to call one of two functions with different signatures, and we don't generally say that language X has function colors because functions with different signatures must be called differently while language Y doesn't have function colors because it just fills in all the omitted values with `undefined` and drops extra arguments on the floor. In fact, we might criticize language Y for doing that, and here we could reasonably criticize Zig for even letting us store pointers to async and non-async functions in the same array of function pointers, given that they must be called differently.
As an aside, while Zig really does end up with some detectable differences between functions that are async and functions that are not, various other implementations of similar ideas like greenlet or Lua coroutines do not have these issues. The author maybe did not look into those implementations because they don't call themselves `async`, but this is a superficial detail. In Lua for example, `suspend` is written `coroutine.yield`, `resume` is written `coroutine.resume`, and `await` is written by calling `coroutine.resume` in a loop until the coroutine has errored or returned. The author could conduct a similar investigation of Lua coroutines and not run into any of the same issues because there really is only one calling convention in Lua (as far as call sites written in Lua are concerned, anyway). C primitives like ucontext_t or those exposed by libco also expose the same operations without introducing additional calling conventions.
The current system of calling conventions in Zig is imo somewhat troublesome because "async/not async" and "inlined/not inlined" should be independent concerns. This has caused some issues in my experience where using `suspend` to jump out of a search when a solution is found causes much worse codegen than the alternatives because the compiler ends up emitting many more distinct functions.
Zig has many similarities stackful coroutine runtimes though, and I suspect that our pre-existing notions of stackful and stackless coroutine libraries may not apply well to it. The compiler has to statically know the maximum depth from which a function or its callees can suspend, then this whole maximum-sized stack is saved in the resulting frame type. This is distinct from e.g. Rust, where each stack frame of the async parts of your async call stack has separate storage.
Edit: I guess I mean to say that Zig could in the future remove the async calling convention (but keep the fact that each function is either async or not) and switch to a model like libco without breaking very much code. If this does happen, I don't think it's necessary to rename the keywords at that time.
The original article that introduced the term unfortunately doesn't contain a clear definition of what "colored" functions are, so people endlessly argue about their particular interpretation of the article.
The original article contains an anecdote that describes a limitation specific to JS, and a couple of opinions that aren't universally applicable. Sometimes it's nice to have magic that sweeps away implementation details, sometimes it's good to have full explicit control.
The definition the author uses for colored function is too broad imo.
void green(int p)
void yellow(char k[10)
Under his definition, these are colored functions. The compiler will create colors for them, and you can only pass them to places that expect these colors.
I have no proof of it yet (I want to create the proof), but I believe the structured concurrency equivalent of the Structured Programming Theorem could be something like this:
> For all possible programs in CSP style or in async style, there exists a program that uses structured concurrency that will accomplish the same result.
Maybe, but later in the post, I show that Zig does, in fact, fit Bob Nystrom's definition:
> Third, async is, in fact, viral. The language reference says,
>> Zig infers that a function is async when it observes that the function contains a suspension point. Async functions can be called the same as normal functions. A function call of an async function is a suspend point.
> (Emphasis added.)
> In other words, if you call an async function, your function gets a suspend point, which means that your function becomes async.
> In fact, if you look at Bob Nystrom’s point 3, “You can only call a red function from within another red function,” this means that Zig fits point 3!
> What Zig does differently than Bob Nystrom’s “hypothetical” language is that if it sees you call a red function from a blue function, it shrugs its shoulders, makes the blue function red, and does not tell you. This gives the appearance of being able to call red functions from blue functions while still only allowing red functions to call red functions.
I wish we had more colors for functions. I don't like the whole "colorless functions" idea. I want to, as a function, be able to constrain my caller. "You can only call me in an async function" is a cool thing to communicate. "You can only call me if you can handle errors", "you can only call me if you have read/write access to the file system", etc - I want arbitrary colors for my functions, not colorless functions.
Not only does this mean that I can express arbitrary constraints on callers, it also means that all constraints and stateful expectations are explicit, verifiable, and clear to the reader.
The post is interesting because it highlights that, colorless or not, functions have constraints and you can try to hide them and handle things automatically,
via inference as is demonstrated here, but they're always there. Async is probably one of the less interesting constraints, but I find it so much more confusing when languages have colorless functions... I can't see suspension points, I can't tell wtf is going on. It's extremely frustrating.
> However, before I continue, let me say a little bit of praise for Zig: this is innovation. This is a piece of good work that, most of the time, async functions can be called the same as normal functions.
It's definitely cool and I like the definition of async they have. I think suspension is probably one of the less interesting things that a function can do, vs, say, IO, which is similarly "colorless" in languages. I just wish there were colors for all of these things.
I think you're missing the point. The problem of colour is not a problem of limitations. It is indeed a good idea to limit your functions to the least required amount of power via the type system. One great way of achieving this is using an algebraic effects system which someone already linked to elsewhere in this thread.
The problem of colour is rather that even though two functions should have the same structure, regardless of their effects, the language still forces you to write two functions each with its own syntactical structure. In type theory terms, this is a problem stemming from a lack of effect polymorphism.
Ideally you'd write one function, polymorphic in its effects, and then just be able to use it in both sync and async contexts. In the sync context, the function's type would specialize in a way that it lacks the `Async` effect, so it wouldn't be able to do async stuff there. In the async context it would specialize in a way that it does have the `Async` effect. Therein you would be unable to apply it in contexts requiring the lack of an `Async` effect.
This is my sentiment as well, I understand some people's reservations and the fact that "async pollutes every calling function, as now everything must async as well", but to me this is correct behaviour and beneficial for the programmer. To me, this is akin to the pure/unpure function classification in FP, it makes sense that calling an unpure function would also "taint" the caller. Why should "fetch_user_db_which_may_be_on_the_other_side_of_the_world()" look the same as "Math.abs()", when their behaviour for the caller is so different?
I recently tried Go to speed up some Cloud Functions that use Firestore, it's very difficult to know which of the chained method calls of the Google-provided SDK is the one to actually "blocking" the green thread, perhaps they are all are unnecessarily? Without studying the source code, it's hard to know. I prefer explicitness over absolute simplicity. It also makes it easier to optimise code in my opinion as it's easier to identify sequential async calls that could actually happen concurrently, either by merging SQL queries or by using Promise.all() or the equivilent.
JavaScript does a good job with the await keyword in my opinion, you know exactly when something could take a significant delay, when it's doing something more behind the scenes. Rust does one better by making ".await" a suffix, allowing for nifty call chains. It communicates that this is an explicit yield point and could take a while.
> To me, this is akin to the pure/unpure function classification in FP, it makes sense that calling an unpure function would also "taint" the caller.
Well it also sucks if unpure code has to add a bunch of boilerplate annotations to every single function and call.
> I recently tried Go to speed up some Cloud Functions that use Firestore, it's very difficult to know which of the chained method calls of the Google-provided SDK is the one to actually "blocking" the green thread, perhaps they are all are unnecessarily?
If you have to mark enough of your functions as "possibly blocking" then you can get just as lost about which one is actually blocking in a particular situation.
> To me, this is akin to the pure/unpure function classification in FP, it makes sense that calling an unpure function would also "taint" the caller.
Since this could be read in a very abstract, purist way, I'd like to add that this is also very much an architecture concern for many pieces of code.
At work, I regularly encounter issues with a piece of code that is written in one of the worst imaginable spaghetti styles. And, since it changes things in a database (not even in proper transactions, btw), and is written to pretty much prevent being tested, no one wants to touch it.
Yet, what the code does would really match an "FP" approach:
- Read in all inputs. You might even read during computation if you're simulating FP through transaction isolation
- Compute the outputs and the new desired state
- Pass that to someone who writes it back to the database and commits the transaction
Outputs in our case are messages that get stored in the database, too (in order to allow recovering after transmission issues), so "output" in effect also just means "state". You don't always have to read "FP" in the strictest sense -- there's a spectrum to this.
Now, if that code was written in such a way it would be easily testable and thus refactorable. In fact, most ideas in the clean/hexagonal architectures or "functional core, imperative shell" are to achieve exactly this. But since the author of said code never had any pushback, they just kept making an already grim situation worse with every change. For years.
I see the same issue with the "ice nine" argument for async. If async infects all your business logic, chances are your abstractions suck.
Edit: Also note that goroutines do effectively the same thing as synchronous code in a larger asynchronous function call: You group the code you can sensibly do in one go into a single goroutine. Then you pass the result to someone else to perform the next step.
So, it's very often just another way to express the same process. CSP will invariably be a better abstraction for some problems, and async/await for others.
But with go's green threads, does it really matter if something blocks or not? Given, I'm not too familiar with Go, but in the case of Java's upcoming Loom which will do something similar, you would just create a virtual thread and if you have to do something with the output of a previous call, you just write plain easy "blocking" code. The JVM will switch to another virtual thread during the "blocking" part automagically and you are done. Whether something is async or not doesn't add anything to the idea you want to implement (in managed languages that is -- rust do need the async keyword because it is a low-level language).
There is a fine balance here, I believe. Java plans on introducing something called structured concurrency to tackle this problem, but even without it I don’t believe that this is a unidirectional tradeoff. Async programming even with syntactic sugar can be overwhelming easily as well.
It may be correct for some abstract mathematical concept of correct. But suppose your lang care about side effects and you want to track coloring for side effects. And then, it turns out there is a sporadic case where there is a performance regression and you want to do telemetry and produce spans to track the regression.
Now you will want to blow your brains out because you will have to refactor all of your function calls down the chain.
It could get even worse if at some point, you are passing a function to a library/framework and for whatever reason it refuses to tolerate a function of your color (assuming you can't lambda-ize the color). Now you have to rewrite the library. Or use a library that does tolerate it. Which might lose other features or ergonomics.
When I've encountered colored functions(and here I'm specifically thinking of D's pure annotation) I have noticed that the architecture tends to collapse to the most permissive forms very quickly, because you will discover layers that were meant to encapsulate the coloration, and did "quack like it", but in a strict sense do not.
For example, one's expectations of a pure function would mean that you could do some arbitrary numeric computation on inputs and return a value. Unfortunately, the compiler is not so naive. If you should do some transcendental math and call a standard library function to do so, you have manipulated hardware state. You are now impure, and so are all your callees.
And so you have to decide at that point: is the code you are writing wrong? Or is it just using too coarse an assumption for its invariants?
A huge amount of what determines our code architectures are the way in which we've organized our own thoughts. In this respect we always go from looser to tighter specification over time as the code matures, until tightness has produced brittleness and cracks. I believe coloration forces some tightening decisions to be scheduled earlier. In that respect it should be treated with some suspicion, as it may produce surprising couplings.
Adding colors without a concrete reason to do so will always be a waste of time. Take type checking as an example. If you use type A as a parameter that requires type B, your program will not work. Type checking provides a lot of value. Or in the GP’s example, forcing functions into async contexts ensures that, for example, the main UI thread doesn’t get blocked. The program will appear broken to the user if this rule is broken. Now compare this to pure functions. What happens if you break the pure function contract? Probably nothing. It’s still totally possible for the program to work. So you end up doing a lot of work that ultimately doesn’t change anything, which is what you’re observing with the proliferation of impure code.
If a compiler assumes that a function is pure, while it's actually not pure, it may end up performing optimizations that break the logic, such as changing `x = f(); y = f()` into `x = f(); y = x`. And so might a human when refactoring.
This illustrates my point perfectly. Theoretical refactoring or optimizations are just hypothetical situations that may or may not occur. Needing to proliferate a function color incurs a real cost in terms of development efficiency. You should get something concrete if you pay the price. Until that happens, such language features are unlikely to gain widespread popularity.
I think the x87 status/config registers you’re alluding to are a great example of power to the compiler that function colouring can potentially grant. “De-colouring” the function requires gating around preserving these registers but within that colour/monad there’s no need to preserve them. That’s an upside, not a downside
The problem with colors is, as most things, bad programmers. The seminal example for me is Java Exceptions. Java tried to color functions based on which errors you had to handle in order to call that function. Well, most beginners didn't understand what an "exception" was, so in order to learn they were told to just ignore if for now. A Decade of bad advice like "just extend RuntimeException" and most of the production code at my job now is littered with catch(Exception e) because colored function were too hard for people. I still get confused looks when I say "Well if I change the types of errors this function can produce, i want the compiler to stop me until I've actually handled them".
You see the same thing happen in async circles. All of a sudden every single function has to be labeled async, whether it needs to or not.
Haskell’s use of monads is perhaps the ultimate in colouring (for values not functions, but that doesn’t make that much of a difference); you routinely colour everything with which pieces of state it needs to read and/or write to do its thing. The result is you either have everything painted in “global state”, which just feels like unnecessary boilerplate, or you spend most of your time defining and converting between elaborately carved pieces of said state.
People could and did invent various intricate constructions which would do the conversions and possibly even the carving for you, but overall I think most agreed it was a gigantic pain.
It’s possible the situation is fixable with a correct combination of features (records, polymorphic variants, implicit parameters, possibly even algebraic effects), as a good portion of the pain comes from not having a standard solution every library would use (they all have drawbacks). But then the checked exceptions story in Java, IIRC, is from pre-Java 5 times, and not being able to be polymorphic in the exception signature of a callback you take is really stupid however you look at it, as well as also a case of inadequate language features.
I don't think it's fair to label the failure of checked exceptions as "beginners didn't understand". It's the fact that returning null was deemed an acceptable return value on some types of failure throughout the standard library, rendering the whole idea basically moot since it didn't actually give you much guarantees.
And later on, checked exceptions were incompatible with new stuff like lambdas and streams
One reason checked exceptions are a not great is that library authors would like you to implement their interfaces or extend their classes, but they've decided that you can't throw anything :)
Zig actually does pretty well on this front. you can write a library that accepts functions that return errors, and your functions will be specialized to return those errors in addition to their own if you don't handle them. Callers who want to handle errors exhaustively with a switch will get compile errors if the error set changes and they fail to handle new errors.
The biggest problem with Java checked exceptions was that you couldn't generalize over them. Simply put, you can't write a method in Java that takes some object X, invokes X.Y(), and throws the same checked exceptions that X.Y() throws. This despite the fact that OO design is heavy on such equivalents of higher-order functions (think comparators, observers, visitors etc).
In a similar vein, as another commenter above said, really the only problem with function coloring is that we're not treating it polymorphically.
It is valid Java, but this code lets you specify only a single exception type, while exception specifications can (and usually do) have multiple types. And Java doesn't have the equivalent of C++ template parameter packs.
The other problem with this approach is that it requires the caller to explicitly specify all the exceptions thrown by the function/object being wrapped. So every time you do map() or fold(), you have to spell out the full exception spec for the closure you're passing in. And god help you if there's nested maps...
BTW, the original structurally-typed proposal for closures in Java included generic exception specifications (among many other things). Unfortunately, that was killed off in favor of a much simpler take.
I know this is a very basic question, but what is it about algebraic effects that makes them compose when monad transformers don’t? I’ve read some of Bauer’s early Eff work, which IIUC was explicitly motivated by this, but for all the neat stuff about resumable exceptions and delimited control and whatnot I don’t feel I ever saw (or recognized) the definitive answer to this.
Until a large company either creates and heavily invests in a new language, or integrates an effect system into a very popular language like C# or Java.
Aside from the demand for explicitness, we may not have to choose one or the other. The promise of Zig's semantics w.r.t. async is that you can write a library that accepts allocators, writers, and readers from the end user, and your library's functions will be specialized at compile time to be async or not-async based on what kind of writers and readers were passed in. If we had a broader effects/capabilities/whatever system, it would still be nice if library authors could address the needs of many different users without caring whether those users want to do IO.
If we really need to be explicit, we begin to need different libraries for each combination of effects/capabilities/whatever, sort of like how Rust users need at least 4 separate sync/async/nostd-sync/nostd-async libraries just to speak websockets. This seems like a waste to me.
Zig's error system in which errors must be handled and functions can opt to have their error sets inferred by the compiler (again, this includes specialization to different error sets for instances of the function that are parameterized by functions with different error sets) sort of gives us the same thing for errors that we have for async-ness, but I can sympathize with a perspective that feels that special casing these two things and not allowing any others is inelegant or whatever.
> Async is probably one of the less interesting constraints, but I find it so much more confusing when languages have colorless functions... I can't see suspension points, I can't tell wtf is going on. It's extremely frustrating.
Zig's view on this, which I endorse, is that suspending to your event loop is not substantially different from suspending to the kernel. Zig sort of assumes that we don't want to manually mark each function with which syscalls it and its callees are allowed to make, but maybe it's worthwhile to make a language that does do that.
> The promise of Zig's semantics w.r.t. async is that you can write a library that accepts allocators, writers, and readers from the end user, and your library's functions will be specialized at compile time to be async or not-async based on what kind of writers and readers were passed in. If we had a broader effects/capabilities/whatever system, it would still be nice if library authors could address the needs of many different users without caring whether those users want to do IO.
The problem is that the library author must then code for the worst case and the compiler won’t help them; injecting preemption into synchronous code can expose things about it that used to be unobservable, but now can make the difference between success and deadlock. (Similarly to how injecting mutation into pure code can expose—and bar the optimizer from changing—things that were previously unobservable, like “in which order does map() visit elements?”.)
There still will be async-compatible and async-incompatible libraries, except now you can’t tell which one is which except by looking at the documentation—look at the lengths R5RS goes to in order to define things like returning multiple times from inside a letrec while R6RS just flat out prohibits that.
Yes, this seems to be a pro-colouring comment while elsewhere in this discussion I make what seem to be anti-colouring comments. I really don’t know what the answer is.
I don't have much to add to this. These are real issues, and we have encountered and needed to fix them, and we probably will many more times. I would like to one day use tools that don't have these problems but also don't compromise on composability.
>and your library's functions will be specialized at compile time to be async or not-async based on what kind of writers and readers were passed in
Based on what I've seen in Rust's async, you don't actually want this in a low level language. Those calling scenarios are fundamentally different. If you magically make something async, other code can now suddenly mutate state that your function is referencing on its own stack and really mess things up.
In Rust this isn't an issue because you're not allowed to pass references into an async function that outlive the future. If you take a synchronous function that holds references and add the async keyword, you probably will get compile errors about this. I don't think Zig has anything like that.
>Zig's view on this, which I endorse, is that suspending to your event loop is not substantially different from suspending to the kernel.
No, that doesn't make sense, not even in the context of Zig. Suspending to the event loop allows other coroutines to run. Blocking in a syscall stops the whole program and all the coroutines.
I have bad news for you, there are probably thousands of HTTP implementations out there, thousands of TCP implementations, thousands of SMTP implementations, etc. That's what happens when a protocol becomes an international standard.
But to address your comment directly, no it's not, not when you're actually building a different state machine with different constraints each time. I actually wish we could put this protocol code in one C library in a way that only handles the protocol logic and not the I/O part, and then we'd be done with it forever. But you know that wouldn't be good enough for some, those libraries exist and people still don't use them and instead they go and write their own in the new fancy language with their cool new concept for a state machine.
> I actually wish we could put this protocol code in one C library in a way that only handles the protocol logic and not the I/O part, and then we'd be done with it forever.
You can have one Python library at least[1], h11 and h2 in particular make a pretty convincing case as the ultimate in Python HTTP. (You will always have wrappers, though, because not all levels of detail are appropriate for all uses—similar to how imagemagick/libgd/etc. wrap libtiff which wraps libjpeg and wrap libpng which wraps zlib and so on.)
The problem, as far as I can see, is that to avoid calling out to I/O you need inversion of control, and either you convert your code into CPS and state machines manually, which almost impossible, almost unreadable, and makes layering libraries a distant dream (not that people don’t do it anyway), or you use some sort of generic inversion-of-control mechanism like async/await or call/cc, which really wants dynamic memory allocation.
It seems like being able to accurately track stack usage (per module, not per program) could solve this, but that means (we must abandon arbitrary recursion and) the compiler must be able to track arithmetic equations, which is liable to make Gödel and Turing rise from their graves just to shout NOPE! at you. There’s also ABI: what do you do when a new version of a protocol library requires a dozen more bytes on “the” stack, and you were putting these on your stack or in static storage?
(As far as I can tell, Rust’s no_std answer to ABI of this kind is “no.”)
>or you use some sort of generic inversion-of-control mechanism like async/await or call/cc, which really wants dynamic memory allocation.
To that end, Zig does have another kind of function coloring that's useful there: functions which take an allocator parameter. I like that about Zig, I wish C++ and Rust were better about doing it. Sadly you can still hide this by referencing a static allocator in Zig so it's not totally there yet on the language level.
> I actually wish we could put this protocol code in one C library in a way that only handles the protocol logic and not the I/O part, and then we'd be done with it forever.
Yes, this is a great approach. Unfortunately people in e.g. the Rust world tend to write crates that assume they can call malloc, recv, and send, rather than one crate just implementing the protocol logic and another that does all that other stuff.
Zig's way of doing things causes the library people naively write and use to call malloc, recv, and send to also be a library that just implements the protocol logic, reads from and writes to static buffers, and doesn't call malloc.
Edit to reply: Yes, Rust has no_std, but most libraries that implement a protocol are not divided into a no_std library that implements just the protocol logic and a separate std-using library that wraps it, nor are most libraries that implement a protocol asynchronously wrappers around a no_std library that is not async and only implements the protocol and makes 0 syscalls. A consequence of the fundamental non-composability of naively written Rust libraries is that users who want the wrong combination of asyncness and malloc-calling very frequently need to reimplement existing Rust libraries. A consequence of the fundamental composability of naively written Zig libraries is that they never need to do that.
Nim toyed with this at some point as an "effect system". I wrote some pragmas (decorators) that ensured a certain function was only run on the network or GUI thread. In Python, I had the same, but as runtime assertions.
Another thing that you could do is in the command pattern for undo/redo, to make sure changes to your data model only happen from command objects. In Python, I had a decorator that said "when this property is set, assert that Action.do is on the call stack". It would be really powerful to be able to do that at compile time, too.
I think the main problem here is that you would end up with multiple copies of the same API, e.g. node.js has three "stamped out" versions of the filesystem API: (1) synchronous, (2) asynchronous with callbacks, and (3) asynchronous with promises. This duplication sucks both for the library maintainer as well as the user (IMHO).
This seems like more of an issue with porting an existing language than creating a new one. With the “can write to the file system” example, you would only need one version of a “writeJSON” function, for example, because writing without write access is nonsensical. Then you’d specify exactly what you need through these constraints, and if they weren’t met, you couldn’t call the function. This already exists and works well when it comes to const-correctness in a language like C++. You can’t call a non-const method from a const one, for example, and this doesn’t lead to multiple versions of each function, because if a function mutates state, you know that it will never be able to be called from a const method.
You probably need at least one WriteJSON that's synchronous and one that's asynchronous, with different colors. If you want a version that uses memoization internally, it needs the "can allocate memory" color, which is independent of the async/sync color.
Could it be possible to infer optional colors from the environment?
You denote that the color is optional in the function's type signature, and inside the body of the function, you have a construct that lets you execute different code depending on the state of the function's colors. The call site could either automatically infer the colors used or the colors used could be manually specified.
Basically a type parameter with compile-time if statements.
If js had async/await from the beginning there would have been no need for the callback version. Also the sync apis would exist in any case because of the event based architecture of node and the fact that at times you don't want to release control back to the event loop.
That's not a problem, those are all individually useful when doing systems programming. Sometimes you're implementing a thread worker and you want (1). Sometimes you want to plug in existing functions and you want (2). Sometimes you want to use I/O completions so you want (3).
It makes sense to avoid this in a javascript runtime because you actually almost always want (3) there. It's too late for node to change it unfortunately. But I'm also pretty confused as to why zig, a (relatively) low level language, considers it important to remove this distinction.
So in someways, the best generalization like this I already commonly see is a capabilities system, which can be done essentially via function parameters. I've used this as an OOP pattern once or twice and I think it's the easiest way to explain it:
Imagine you have 2 singleton objects in your 'void main()', redkey and bluekey of types Redkey and Bluekey.
Somewhere else, you declare your function: 'int foo(Redkey redkey, int x, int y)' that needs a Redkey object of which only one exists: in your main.
This by itself forces every call on the path between 'main()' and a call to 'foo()' to also include a parameter Redkey.
In the extreme cases where a pattern like this is useful (tracing IO calls, you named it) it can really help cut through a codebase after an hour of refactoring. But it can be limiting. Async and Checked Exceptions are probably the most colored functions and they both need escape hatches because of that.
Let's say that you have a pure function f that does a bunch if calculations and return the result. The caller then prints it. Now consider a the same function f' but instead of returning the result does a tail call to a function r, passed as an additional parameter.
The two scenarios are the same, yet in a colored functions language f' will need to be annotated if r does, for example io.
The way I see it, a language should either allow 'closing' over the color (or effect or capability, or whatever) hiding it, so that f' doesn't beed to care about the details of r; or it should provide enough of polymorphic behaviour so that f' the signature of f' can be inferred from that of r.
Most languages with async do not provide this capability.
> The way I see it, a language should either allow 'closing' over the color (or effect or capability, or whatever) hiding it, so that f' doesn't beed to care about the details of r; or it should provide enough of polymorphic behaviour so that f' the signature of f' can be inferred from that of r.
> Most languages with async do not provide this capability.
Those where async is syntax sugar for something else (e.g., JavaScript, Python, I forget if C# qualifies this way) allow hiding; you don't need annotation to call an async function, you only need it if you choose to use async syntax in the calling function.
The proposed hiding is basically making the caller generic over (and inherit) the effectfulness of the callee. There are probably some FP languages that do this in the general case. Zig can do this for async and for error sets, but you can explicitly opt out of both kinds of hiding if you want to by writing an event loop that uses `resume` but does not use `await` (or uses `nosuspend` when it uses `await`) or specifying your exact error set + handling errors. Lua does this for suspending (and you can opt out with e.g. coroutine.wrap) and errors (and you can opt out using pcall). JavaScript does this for exceptions (and you can opt out by catching them) but not for async.
> you don't need annotation to call an async function, you only need it if you choose to use async syntax in the calling function.
The proposed thing is "I want to get a function, and call it, and use the result, and if it is effectful in a way then I am effectful in that way, and if it is not then I am not" where "the result" is defined (among other things) as the value of the expression that appears in a return statement in the body of the function. Redefining "the result" to be some object representing a continuation or whatever is missing the point.
The lack of async-hiding in JavaScript is why the original "What Color is Your Function?" blog post exists.
I'm not necessarily disagreeing with you, but is there an example for a function that doesn't make sense to call from both an async and non-async function, other than an implementation detail of the programming language?
But continuing on your idea, one colouring I would really like to see everywhere is pureness. C++'s `const` comes close, and perhaps Haskell does it right by having everything be (mostly) pure by default and using the type system to encode other properties. And yeah - different monads (like IO) are likely a good way to handle this whole topic in a user-definable way, but not many people like them (at least when they know they are using them, because unknowingly many programmer do use it, eg. most async syntactic sugar is basically a do notation for a single hard-coded Monad)
in javascript's programming model, literally everything that's running in response to user action (clicks, etc.) absolutely has to complete immediately (from the browser user's point of view), or it will hang the UI. async event handlers really help with that. a non-async handler blocking until a long standing operation is complete (i.e awaiting an async function) is unacceptable
the async/non async functions in javascript are good and well designed. the whole colored functions thing is nonsense and it's weird that it gained so much traction on here
Starting up a new virtual thread and writing “blocking” code in it, while the VM schedules it on top of a single thread is entirely possible and will make everything happen quasi-instantly, the same way as async is.
I think that people that dislike monads mostly fall in one of two categories: those that know about a fancier alternative such as algebraic effects and those that are scared off by an alien-looking unfamiliar concept.
Those two groups are minorities in the set of all people that are even aware of monads.
It was deprecated in C++11, and a removed in C++17, yet now they are discussing adding it back in a flavour similar to Swift, so apparently there is some value into it.
As for being useless, well that is the usual argument from no exception, no RTTI folks, that would be better off using another language instead of messing up with C++'s design.
I'm very much in favour of statically checked exceptions (and noexcept is sorta-kinda like that). It is the old dynamically typed style which was little more than an assertion that wasn't very useful.
Haskell's system (with the io monad + do not all monads or in general) is extremely similar to async/await and most languages with co-routines can simulate do notation in Haskell pretty well syntactically.
The advantage of something like JavaScript's async/await over Haskell's "do" is the fact because async/await is more limited than `do` (or yield and coroutines in JS) it's a lot easier to create tooling for.
I would say it is a huge disadvantage -- do is just syntactic sugar for a monad expression of binds, which is a native part of the language, you can write a simple function to manipulate it however you like. While you would need to effectively parse AST in case of JS.
`yield` is just syntactic sugar for the two-way generator communication protocol where the consumer calls `.next(value)` which returns a value in turn. This is as expressive and you can build `do` semantics on top of stuff like lists/either/state/io on top of it. No AST parsing needed.
And do syntax sugar is just nested calls of `bind` which is a function of the Monad trait. Lists are also a Monad, so do notation just works right now with lists, either, state everything in Haskell.
Basically by limiting the scope significantly and limiting it to language-level Promises you can create things like zero-cost async stack traces, async-call aware flame graphs and more for analysis and better lint rules and more specific types for development.
Despite not liking the idea of function colors, I do think I like your idea of having functions express what capabilities you have to have permission for to call them. A long time ago, I saw someone talk about fine-grained permissions within a programming language, and I think they are the same idea.
I would love if that could be done without function colors, though.
Yeah, I should have elaborated. Like, fundamentally you want to express "caller has capability x", and coloring does that. But coloring isn't ergonomic, so some languages try to infer their way out of it or hide it. But I'd rather they instead just make these things more explicit and lean into it a bit more, like in a capabilities system.
In the "real code" I linked to in the post, I have a concept I call "context stacks". The idea comes from Jai, and I use to do implement things such as error handling and allocators in such a way that functions don't need to take an allocator argument. To make callees use your allocator, you push your allocator onto the allocator context stack and then call the functions you want to use that allocator.
I made it useful for many more things, and users could add their own.
In particular, I realized that it could be extended to where functions could get permissions (such as popping up an Admin dialog box on Windows or asking for sudo on Linux), push a capability [1] onto the capability context stack, and then call the functions that actually do the work. These functions could then grab that capability and do what they need to do.
One good thing about this is that permissions don't leak up to callers that should not have them. In other words, this system keeps permissions from infecting code that really shouldn't have them, but might need them because they might call functions that do.
I don't know if all of this made sense, sorry if not. I wrote it really fast.
People mean one of two different things when they talk about “function colors”.
Let’s take the example of JS. Does JS have function colors?
In one sense, yes, there are async functions and sync functions and you can only call an async function from an async function. So under a certain definition, JS functions are colorful.
But wait... a JS async function is just a normal function that returns a Promise object. So under another definition, JS functions are colorless. Perhaps someone might not like how Promises work in JS, but this has nothing to do with what color the functions are.
There are some languages that do break this and implement async functions as a completely different entity to normal functions. I believe this kind of function colors is lazy and bad language design and these async implementations tend to feel like they don’t belong with the rest of the language and were duct taped on.
The async keyword strongly changes the internals of a function to let it use "await". Even if you do it manually, it's a very particular and restricted style that infects your caller. That's a color.
Wouldn't this be something similiar to Monads? I remember someone gave a talk and described Monads as "worlds". (I forgot which video it was). So a function goes into a "world" does all it's stuff, then it exits and comes down.
Traits is imo kind of this, in rust and to a lesser extent c++. At the very least, a templated function can insist on being injected with what it needs to engage with the outside world as far as an allocator, IO, error handling, and so on.
Isn't this hard to make ergonomic in languages without higher-kinded types, because you need to specialize implementations of functions that otherwise could be generic over colors? like the filter() example in the original What Color is your Function post.
"Wanna know one that doesn’t? Java. I know right? How often do you get to say, “Yeah, Java is the one that really does this right.”? But there you go."
Structured concurrency has nothing to do with function colouring. It is how you organise and spawn tasks, so that you can have certain guarantees e.g. the group responds to cancellation as you'd expect, etc.
I don't see how async/await plays a part in this. You could have structured concurrency with a os threading model, as opposed to a userland threading model.
You can argue that the ergonomics of Zig's async story needs better documentation, especially if you're playing with function pointers. But please don't mislead readers that structured concurrency solves this.