Like, no careful thinking and good 80/20 solution this time. Just "huh, we'd need coroutines to do this `right` so let's just do that"
When they added generics, they really, really thought long and hard, and came up with a compromise that was brilliant and innovative in the balance it struck.
I would have hoped to see something like that here, like "we're adding this one feature to goroutines to have control in specific situations" feels like something that would be better than "we're going full Rust on this problem and just adding it."
https://github.com/golang/go/issues/43557
https://github.com/golang/go/discussions/56413
I'm not sure Russ's personal blog is any kind of official statement "this is what we're doing" yet?
I've found it's generally a very strong indicator.
But to your point, this isn't a new issue, there's been a good amount of discussion around it over the years.
https://github.com/golang/go/discussions/56413#discussioncom...
for {
next := getNext()
...
}
What is the advantage of writing this as: for next := range getNext {
...
} getNext := iterableThing.Iterator()
for {
next, ok := getNext()
if !ok {
break
}
...
}
vs. for next := range iterableThing.Iterator() {
...
}
One advantage is that it's slightly shorter, which matters for very common patterns--people complain about `err != nil` after all. Another advantage is there isn't another variable for everyone to name differently. Another advantage is that everyone can't do the control flow slightly differently either.Not to nitpick this specifically but as a generic reminder not all complaints are worthy of shifting the trajectory of a massively popular programming language.
Balancing "worthy" and "unworthy" changes is really hard both in the community and discussions like this one. I don't envy the teams that have to do it.
getNext := iterableThing.Iterator()
for next, ok := getNext(); ok; next, ok = getNext() {
...
}
Which, yeah, the range cleans it up a bit, but it's not doing quite as much work as you're implying. for i.HasNext() {
current := i.GetNext()
...
}
And just mutate the internal state of the struct. Probably throw it behind some sort of interface like: type Iterator[T any] interface { // feel free to strip out the generic
HasNext() bool // if you really only have one type
GetNext() T // you do this with
}
I think this still loses to ranging over a function, but I still think it should be compared to what you can do with the current language as opposed to what was originally posted.Go has simplicity as a design goal. Part of that simplicity is not adding all the features we can think of to the language. Just because it would provide slightly leaner syntax for a relatively small group of use cases isn't a good reason to add new language features (imho, from my 10+ years using Go and watching it evolve).
If we can implement this using current language features, but it's complex and messy to implement, then it's a great case for a new library, possibly even inclusion in the standard library. If we can't implement this at all using current language features, then maybe it's a case for a new language feature to enable this. If we can implement it relatively cleanly using current language features, then we're good and don't need to do anything. This seems like a "can implement, but not very cleanly" case, which would be a great justification for a library, but not a language feature. Again, imho.
First sentence of OP:
> This post is about why we need a coroutine package for Go
Then in the girst section after the intro:
> Later, I will argue for an optimized implementation provided directly by the runtime, but that implementation should be indistinguishable from the pure Go definition.
The point I was trying to make is that some amount of people see the benefit in removing much smaller boilerplate. And so presumably at least as many people should see the benefit of removing a larger amount of boilerplate.
I'm just talking about the benefits here. There is also a cost to expanding the language. The costs might outweigh the benefits, but it doesn't mean that the benefits don't exist.
Agreed. I think this is where the discussion will focus. I'm glad that historically the Go team have weighted the costs heavily, and hope they continue to do so.
I guess I'm in the minority but boilerplate really doesn't bother me. Going the other way gets too much "magic" and we lose the ability to fine-tune behaviour.
That is not an advantage. That's a slight convenience feature that saves maybe a few seconds every now and then, from the least time-consuming part of programming, which is typing.
The price to pay in the long run for such conveniences is high, as demonstrated in nearly every language which went down that path. Once textbooks start advising against the use of language features because they carry the risk of making the code less legible, a language design has gone down a wrong path.
> people complain about `err != nil` after all.
People complain about many things. People complain about having to wait in line. People complain about taxes. People complain about having to use 2FA. People complain about trees taking up space they would like to have as parking lots for their SUVs.
The fact that people complain about X, doesn't mean X is bad, it doesn't mean X is wrong, it doesn't mean X has to change.
The point of coroutines is that they are little contexts that serve to be called
- many times
- sometimes with different parameters
- change state
- might be interrupted by callers
- might interrupt callers themselves
All of this can be done by other means. Just like any sorting can be done by copy pasting the same code, but generics make it less tedious. That's the same idea here. Some problems can be implemented as interleaving coroutines, and their definition is simple enough that you want to write it all in the body of some CreateCoroutine() function instead of taking out another struct with 5 almost empty functions. It will not solve all problems, but can more clearly separate business logic and orchestration.
The harder case to deal with is a "push iterator", which are often much simpler to implement, but less flexible to use. See https://github.com/golang/go/discussions/56413
The OP is about converting a push iterator into a pull iterator efficiently by using coroutines. This provides the best of both worlds, simple iterator implementation and flexible use by its caller.
...
if getNext() == nil {
break
}
...
This isn't a huge boon, and is mostly a matter of style. But I prefer the latter because it's logic that gets handled with the range builtin, so my code can be more about application logic, rather than muddling in breaking out of loop logic. for {
next := <-channel
...
}From TFA:
"Because scheduling is explicit (without any preemption) and done entirely without the operating system, a coroutine switch takes at most around ten nanoseconds, usually even less. Startup and teardown is also much cheaper than threads."
"For this taxonomy, Go's goroutines are cheap threads: a goroutine switch is closer to a few hundred nanoseconds"
Also check out https://news.ycombinator.com/item?id=29510751
and https://ewencp.org/blog/golang-iterators/index.html:
"Finally, the most natural channel based implementation is… slow. By a factor of nearly 50x. Buffering does help though, reducing that to a factor of 25x."
for {
next := <-channel
...
}
You have to set up the channel and the goroutine that feeds it, you need to safely close the channel when the iteration is done (but not before, unless you like panics), you need to deal with panics inside the goroutine and possibly support cancellation if the iteration breaks early (unless you love memory leaks).If you try to implement this pattern by hand, you are all too likely to make fatal mistake, and this is doubly true in the hands of an inexperienced programmer.
I appreciate the fact that Russ wrote this long post, gradually implementing `coro.New()` and improving its functionality and safety — and only in the very end, we get a short paragraph about performance. Good performance is important to make this feature attractive to use, but if the feature is clunky and error-prone, it wouldn't be worth much, even with great performance.
Google has internal kernel patches that implement cooperating multithreading this way (internally they're called fibers), and the patches exist for precisely this reason: better latency and more predictable scheduling behavior. Paul Turner gave a presentation about this at LPC ~10 years ago that explains more about the motivation for the patches and why they improve efficiency: https://www.youtube.com/watch?v=KXuZi9aeGTw
If programmers try to manually implement iterators, generators and interleaved state machines with their own goroutines and channels, it's not just performance that suffers - there is too much room for error.
[1] I'm using the qualifier "true" here, since many modern languages (such as Python, Kotlin) use the term "coroutines" for something that is more like Go's Goroutines than Lua's coroutines. Unlike Go, they are not preemptible, but they are (at least by default) implicitly resumed when necessary by some scheduler, and they may execute on different kernel threads and switch contexts.
Python has both true (ish) coroutines (or at least coroutines which are entirely user controllable), which it mostly uses for iteration, and the concurrency specialised “async”.
Initially the goal was to reuse “yield” for concurrency (a big reason why “yield from” was added), but the ergonomics of mixing multiple coroutines uses was found to be awful, at least for the langage python is, and trying to provide relevant sugar difficult.
I understand that the historical reasons for that ("generators" originally only supported "yield" and a separate async/await design with implicit scheduling seemed more ergonomic). But that doesn't remove the confusion.
But Python's generators are still fully "true" coroutines. Russ makes the distinction between coroutines and generators based on whether as "Generators provide less power than coroutines, because only the top-most frame in the coroutine is allowed to yield. ". I think this is a bad definition, as generators can clearly "pass" the yielding power to other generators with "yield from", and Russ talks about this feature in the same post. Generators require more ceremony, but I don't think they are "less powerful" in any meaningful way .
Roberto Ierusalimschy, the creator of Lua, and Ana Lucia de Moura proposed the distinction between stackful and stackless coroutines[1]. Their paper revived coroutines as a research subject and had a lasting impact on coroutine-related terminology. Ierusalimschy and Moura made the distinction between "Stackful" coroutines and "non-stackful coroutines" (the term "stackless coroutines" stuck later on, but they did not use it), where Stackful coroutines like Lua's coroutines can "suspend their execution from within nested functions". They also made the same allusion to power that Russ does: "powerful enough to constitute a general control abstraction; in particular, they can- not be used as a concurrent construct".
I think both Ierusalimschy and de Moura are onto something when they call generators "limited", but they are just incorrect when they say that "but are not powerful enough to constitute a general control abstraction; in particular, they cannot be used as a concurrent construct". Early Node.js libraries and frameworks such as "co" and "koa"[2] should count as a proof that you can very much use generators as a concurrency construct. Generators have all the power that is necessary for dealing with normal concurrency cases, although they don't have the best ergonomics.
But there is another distinction where "generators" are not the same as Lua coroutines, and clearly lack some expressive power (even though this power is NOT a necessary condition for implementing full-fledged concurrency). Lua allows coroutines to communicate bidrectionally: yield() can pass a value back the coroutine caller, which is returned from resume() - but when the coroutine caller calls resume() again they can pass an argument which will be returned from yield(). Generators are unidirectional: they only allow you to pass a value to yield(), but not to resume().
[1] http://www.inf.puc-rio.br/~roberto/docs/MCC15-04.pdf [2] https://news.ycombinator.com/item?id=6933358
I might be missing something as I’m not familiar with Lua’s coroutines but that sounds like send (https://docs.python.org/3/reference/expressions.html#generat...). The parameter to `send` (from outside the coroutine) is the return value of `yield` (inside the coroutine).
Why though?
I can get that using a simple channel. The generator sends messages to it instead of "yielding" them. When it's done, it closes the channel. The consumer ranges over the channel.
There, done, function iterator. No library required. No additions to the language required.
I still don’t see the need for coroutines.
The use case motivating all the complexity is function iterators, where `range` can be used on functions of type `func() (T, bool)`. That has been discussed in the Go community for a long time, and the semantics would be intuitive/obvious to most Go programmers.
This post addresses the next thing: Assuming function iterators are added to the language, how do I write one of these iterators that I can use in a for loop?
It starts by noticing that it is often very easy to write push iterators, and builds up to a push-to-pull adapter. It also includes a general purpose mechanism for coroutines, which the adapter is built on.
If all of this goes in, I think it will be bad practice to use coroutines for things other than iteration, just like it's bad practice to use channels/goroutines in places where a mutex would do.