They propose just using an async iterator of UInt8Array. I almost like this idea, but it's not quite all the way there.
They propose this:
type Stream<T> = {
next(): Promise<{ done, value: UInt8Array<T> }>
}
type Stream<T> = {
next(): { done, value: T } | Promise<{ done, value: T }>
}
- I can easily make mine from theirs
- In theirs the conceptual "stream" is defined by an iterator of iterators, meaning you need a for loop of for loops to step through it. In mine it's just one iterator and it can be consumed with one for loop.
- I'm not limited to having only streams of integers, they are
- My way, if I define a sync transform over a sync input, the whole iteration can be sync making it possible to get and use the result in sync functions. This is huge as otherwise you have to write all the code twice: once with sync iterator and for loops and once with async iterators and for await loops.
- The problem with thrashing Promises when splitting input up into words goes away. With async iterators, creating two words means creating two promises. With stream iterators if you have the data available there's no need for promises at all, you just yield it.
- Stream iterators can help you manage concurrency, which is a huge thing that async iterators cannot do. Async iterators can't do this because if they see a promise they will always wait for it. That's the same as saying "if there is any concurrency, it will always be eliminated."
> - I can easily make mine from theirs
That... doesn't make it superior? On the contrary, theirs can't be easily made out of yours, except by either returning trivial 1-byte chunks, or by arbitrary buffering. So their proposal is a superior primitive.
On the whole, I/O-oriented iterators probably should return chunks of T, otherwise you get buffer bloat for free. The readv/writev were introduced for a reason, you know.
This lines up with my thinking. The proposal should give us a building block in the form of the primitive. I would expect the grandparent comment’s API to be provided in a library built on top of a language level primitive.
Come on, it's how (mature libraries of) parser combinators work. The only slightly tricky part here is detecting leftover data in the pipeline.
> If you want to stream arbitrary JavaScript values, use async iterables directly
OK, so we have to do this because code points are numbers larger than 8 bits, so they're arbitrary JS values and we have to use async iterables directly. This is where the amount of per-item overhead in an async iterable starts to strangle you because most of the actual work being done at that point is tearing down the call stack between each step of each iterator and then rebuilding it again so that the debugger has some kind of stack traces (if you're using for await of loops to consume the iterables that is).
Plus theirs involves the very concrete definition of an array, which might have 100 prototype methods in JS, each part of their API surface. I have one function in my API surface.
If you go back a few versions, that number goes up to around 105x. I don’t recall now if I tested back to 14. There was an optimization to async handling in 16 that I recall breaking a few tests that depended on nextTick() behavior that stopped happening, such that the setup and execution steps started firing in the wrong order, due to a mock returning a number instead of a Promise.
I wonder if I still have that code somewhere…
I dabble in JS and… what?! Any idea why?
If you’re doing real work, 90 instructions ain’t much but it’s not free either. If you’ve got an async accumulator (eg, otel, Prometheus) that could be a cost you care about.
I was bumping into PRs trying to eliminate awaits in long loops and thinking surely the overhead can’t be so high to warrant doing this, especially after node ~16. I was wrong.
From what it looks like, they want their streams to be compatible with AsyncIterator so it'd fit into existing ecosystem of iterators.
And I believe the Uint8Array is there for matching OS streams as they tend to move batches of bytes without having knowledge about the data inside. It's probably not intended as an entirely new concept of a stream, but something that C/C++ or other language that can provide functionality for JS, can do underneath.
For example my personal pet project of a graph database written in C has observers/observables that are similar to the AsyncIterator streams (except one observable can be listened to by more than one observer) moving about batches of Uint8Array (or rather uint8_t* buffer with capacity/count), because it's one of the fastest and easiest thing to do in C.
It'd be a lot more work to use anything other than uint8_t* batches for streaming data. What I mean by that, is that any other protocol that is aware of the type information would be built on top of the streams, rather than being part of the stream protocol itself for this reason.
And yes, because it's a new abstraction the compat story is interesting. We can easily wrap any source so we'll have loads of working sources. The fight will be getting official data sinks that support a new kind of stream
Adding types on top of that isn't a protocol concern but an application-level one.
[0] https://github.com/microsoft/TypeScript/blob/924810c077dd410...
I agree with this.
I have had to handle raw byte streams at lower levels for a lot of use-cases (usually optimization, or when developing libs for special purposes).
It is quite helpful to have the choice of how I handle the raw chunks of data that get queued up and out of the network layer to my application.
Maybe this is because I do everything from C++ to Javascript, but I feel like the abstractions of cleanly getting a stream of byte arrays is already so many steps away from actual network packet retrieval, serializing, and parsing that I am a bit baffled folks want to abstract this concern away even more than we already do.
I get it, we all have our focuses (and they're ever growing in Software these days), but maybe it's okay to still see some of the bits and bytes in our systems?
But what if you just want to do a simple decoding transform to get a stream of Unicode code points from a steam of bytes? If your definition of a stream is that it has UInt8 values, that simply isn't possible. And there's still gonna be waaay too many code points to fall back to an async iterator of code points.
Also, I wasn't talking about building network layers, I was explicitly referring to things that use a network layer... That is, an application receiving streams of enumerable network data.
I also agree with what you're saying, we don't want UInt8, we want bits and bytes.
I'm really confused as to why the parent comment was edited so heavily. Oh well, that's social media for you.
Ironically, naively, I'd expect something more like a callback where you would specify how your input gets written to a buffer, but again im definitely losing a lot of nuance from not doing JS in a long while.
They are those languages versions of goroutines, and JavaScript doesn’t have one. Generators sort of, but people don’t use them much, and they don’t compose them with each other.
So if we are going to fix Streams, an implementation that is tuned only for IO-bound workflows at the expense of transform workflows would be a lost opportunity.
type Stream<T> = {
next(): { done, value: T } | Promise<{ done, value: T }>
}
type Stream<T> = {
next(): { done, value: T }
}
type Stream<T> = {
next(): Promise<{ done, value: T }>
}
> My way, if I define a sync transform over a sync input, the whole iteration can be sync making it possible to get and use the result in sync functions. This is huge as otherwise you have to write all the code twice: once with sync iterator and for loops and once with async iterators and for await loops.
Writing all the code twice is cleaner in every implementation scenario I can envisage. It's very rare I want generalised flexibility on an API call - that leads to a lot of confusion & ambiguity when reading/reviewing code, & also when adding to/editing code. Any repetitiveness in handling both use-cases (separately) can easily be handled with well thought-out composition.
But the bigger problem here is that sync and async aren't enough. You almost need to write everything three times: sync, async, and async-batched. And that async-batched code is gonna be gnarly and different from the other two copies and writing it in the first place and keeping it in sync is gonna give you headaches.
To see how it played out for me take a look at the difference between:
https://github.com/iter-tools/regex/blob/a35a0259bf288ccece2... https://github.com/iter-tools/regex/blob/a35a0259bf288ccece2... https://github.com/iter-tools/regex/blob/a35a0259bf288ccece2...
While I understand the logic, that's a terrible idea.
* The overhead is massive. Now every 1KiB turns into 1024 objects. And terrible locality.
* Raw byte APIs...network, fs, etc fundamentally operate on byte arrays anyway.
In the most respectful way possible...this idea would only be appealing to someone who's not used to optimizing systems for efficiency.
Small, short-lived objects with known key ordering (monomorphism) are not a major cost in JS because the GC design is generational. The smallest, youngest generation of objects can be quickly collected with an incremental GC because the perf assumption is that most of the items in the youngest generation will be garbage. This allows collection to be optimized by first finding the live objects in the gen0 pool, copying them out, then throwing away the old gen0 pool memory and replacing it with a new chunk.
Are there any concerns that the extra array overhead will make the application even more vulnerable to out of memory errors while it holds off on GC to process the big stream (or multiple streams)?
I am mostly curious, maybe this is not a problem for JS engines, but I have sometimes seen GC get paused on high throughput systems in GoLang, C#, and Java, which causes a lot of headaches.
If you make all your memory usage patterns possible for the incremental collector to collect, you won't experience noticeable hangups because the incremental collector doesn't stop the world. This was already pretty important for JS since full collections would (do) show up as hiccups in the responsiveness of the UI.
Makes a lot of sense, cool that the garbage collector can run independently of the call stack and function scheduler.
At the moment the consensus seems to be that these language features haven't been worth investing much in optimizing because they aren't widely used in perf-critical pathways. So there's a chicken and egg problem, but one that gives me some hope that these APIs will actually get faster as their usage becomes more common and important, which it should if we adopt one of these proposed solutions to the current DevX problems
My reference point is from a noob experience with Golang - where I was losing a bunch of efficiency to channel overhead from sending millions of small items. Sending batches of ~1000 instead cut that down to a negligible amount. It is a little less ergonomic to work with (adding a nesting level to your loop).
To see the problem let's create a stream with feedback. Lets say we have an assembly line that produces muffins from ingredients, and the recipe says that every third muffin we produce must be mushed up and used as an ingredient for further muffins. This works OK until someone adds a final stage to the assembly line, which puts muffins in boxes of 12. Now the line gets completely stuck! It can't get a muffin to use on the start of the line because it hasn't made a full box of muffins yet, and it can't make a full box of muffins because it's starved for ingredients after 3.
If we're mandated to clump the items together we're implicitly assuming that there's no feedback, yet there's also no reason that feedback shouldn't be a first-class ability of streams.
>> fn double(iter: $iterator<i32>) {
return *for x in iter { $yield( x * 2 )}
}
>> fn add_ten(iter: $iterator<i32>) {
return *for x in iter { $yield( x + 10 )}
}
>> fn print_all(iter: $iterator<i32>) {
for x in iter { $print( x )}
}
>> const source = *for x in [1, 2, 3] { $yield( x )}
>> source |> double |> add_ten |> print_all
12
14
16
If you're dealing with small objects at the production side, like individual tag names, attributes, bindings, etc. during SSR., the natural thing to do is to just write() each string. But then you see that performance is terrible compared to sync iterables, and you face a choice:
1. Buffer to produce larger chunks and less stack switching. This is the exact same thing you need to do with Streams. or
2. Use sync iterables and forgo being able to support async components.
We faced this problem in Lit-SSR and our solution was to move to sync iterables that can contain thunks. If the producer needs to do something async it sends a thunk, and if the consumer receives a thunk it must call and await the thunk before getting the next value. If the consumer doesn't even support async values (like in a sync renderToString() context) then it can throw if it receives one.
This produced a 12-18x speedup in SSR benchmarks over components extracted from a real-world website.
I don't think a Streams API could adopt such a fragile contract (ie, you call next() too soon it will break), but having some kind of way where a consumer can pull as many values as possible in one microtask and then await only if an async value is encountered would be really valuable, IMO. Something like `write()` and `writeAsync()`.
The sad thing here is that generators are really the right shape for a lot of these streaming APIs that work over tree-like data, but generators are far too slow.
Also I'm curious why you say that generators are far too slow. Were you using async generators perhaps? Here's what I cooked up using sync generators: https://github.com/bablr-lang/stream-iterator/blob/trunk/lib...
This is the magic bit:
return step.value.then((value) => {
return this.next(value);
}); type Stream<T> = {
next(): { done, value: T } | Promise<{ done, value: T }>
}
Engineers had a collective freak out panic back in 2013 over Do not unleash Zalgo, a worry about using callbacks with different activation patterns. Theres wisdom there, for callbacks especially; it's confusing if sometime the callback fires right away, sometimes is in fact async. https://blog.izs.me/2013/08/designing-apis-for-asynchrony/
And this sort of narrow specific control has been with us since. It's generally not cool to use MaybeAsync<T> = T | Promise<T>, for similar "it's better to be uniform" reasons. We've been so afraid of Zalgo for so long now.
That fear just seems so overblown and it feels like it hurts us so much that we can't do nice fast things. And go async when we need to.
Regarding the pulling multiple, it really depends doesn't it? It wouldn't be hard to make a utility function that lets you pull as many as you want queueing deferrables, allowing one at a time to flow. But I suspect at least some stream sources would be just fine yielding multiple results without waiting. They can internally wait for the previous promise, use that as a cursor.
I wasn't aware that generators were far too slow. It feels like we are using the main bit of the generator interface here, which is good enough.
I was so sick of being slapped around by LJHarb who claimed to me again and again that TC39 was honoring the Zalgo post (by slapping synthetic deferrals on everything) that I actually got Isaacs to join the forum and set him straight: https://es.discourse.group/t/for-await-of/2452/5
I feel like my deep dives into iterator performance are somewhat wasted because I might have made my project faster, but it's borderline dark magic and doesn't scale to the rest of the ecosystem because the language is broken.
That's an inherent flaw of garbage collected languages. Requiring to explicitly close a resource feels like writing C. Otherwise you have a memory leak or resource exhaustion, because the garbage collector may or may not free the resource. Even C++ is better at this, because it does reference counting instead.
The async iterable approach makes so much more sense because it composes naturally with for-await-of and plays well with the rest of the async/await ecosystem. The current Web Streams API has this weird impedance mismatch where you end up wrapping everything in transform streams just to apply a simple operation.
Node's original stream implementation had problems too, but at least `.pipe()` was intuitive. You could chain operations and reason about backpressure without reading a spec. The Web Streams spec feels like it was written by the kind of person who thinks the solution to a complex problem is always more abstraction.
import { Repeater } from "@repeaterjs/repeater";
const keys = new Repeater(async (push, stop) => {
const listener = (ev) => {
if (ev.key === "Escape") {
stop();
} else {
push(ev.key);
}
};
window.addEventListener("keyup", listener);
await stop;
window.removeEventListener("keyup", listener);
});
const konami = ["ArrowUp", "ArrowUp", "ArrowDown", "ArrowDown", "ArrowLeft", "ArrowRight", "ArrowLeft", "ArrowRight", "b", "a"];
(async function() {
let i = 0;
for await (const key of keys) {
if (key === konami[i]) {
i++;
} else {
i = 0;
}
if (i >= konami.length) {
console.log("KONAMI!!!");
break; // removes the keyup listener
}
}
})();
It’s one of those abstractions that’s feature complete and stable, and looking at NPM it’s apparently getting 6.5mil+ downloads a week for some reason.
Lately I’ve just taken the opposite view of the author, which is that we should just use streams, especially with how embedded they are in the `fetch` proposals and whatever. But the tee critique is devastating, so maybe the author is right. It’s exciting to see people are still thinking about this. I do think async iterables as the default abstraction is the way to go.
edit: I found where stop is created[1]. I can't say I've seen this pattern before, and the traditionalist in me wants to dislike the API for contradicting conventions, but I'm wondering if this was designed carefully for ergonomic benefits that outweigh the cost of violating conventions. Or if this was just toy code to try out new patterns, which is totally legit also
[1]: https://github.com/repeaterjs/repeater/blob/638a53f2729f5197...
let resolveRef;
const promise = new Promise((res) => { resolveRef = res; });
const callback = (data) => {
// Do work...
resolveRef(data); // This "triggers" the await
};
Object.assign(callback, promise);
I'm working on a db driver that uses it by convention as part of connection/pool usage cleanup.
At a native level (C++/rust), a Promise is just a closure added to a list of callbacks for the event loop. Yes, if you did 1 per streamed byte then it would be huge but if you're doing 1 promise per megabyte, (1000 per gig), it really shouldn't add up 1% of perf.
So if you're going to flatten everything into one stream then you can't have a for loop implementation that defensively awaits on every step, or else it'll be slooooooooow. That's my proposal for the change to the language is a syntax like
for await? (value of stream) {
}
the idea is basically just use functions. no classes and very little statefulness
But Observables really do not solve the problems being talked about in this post.
[1] https://github.com/WICG/observable [2] https://github.com/WICG/observable/issues/216
Sadly it will never happen. WebAssembly failed to keep some of its promises here.
classic case of not using an await before your promise
In an ideal world you could just ask the host to stream 100MB of stuff into a byte array or slice of the wasm heap. Alas.
for await (const chunk of stream) {
// process the chunk
stream.returnChunk(chunk);
}
(Lately I’ve been thinking that a really nice stream or receive API would return an object with a linear type so that you must consume it and possibly even return it. This would make it impossible to write code where task cancellation causes you to lose received data. Sadly, mainstream languages can’t do this directly.)
> I'm not here to disparage the work that came before — I'm here to start a conversation about what can potentially come next.
Terrible LLM-slop style. Is Mr Snell letting an LLM write the article for him or has he just appropriated the style?
Just ctrl-f'ing through previous public posts, I think there were a total of 7 used across about that many posts. This one for example had 57. I'm not good enough in proper English to know what the normal number is supposed to be, just pointing that out.
It doesn't really matter what tools are used if the result is good
You might say well, it's on the Cloudflare blog so it must have some merit, but after the Matrix incident...
I would instead say that it is written by James Snell, who is one of the central figures in the Node community; and therefore it must have some merit.
I’ve read my fair share of LLM slop. This doesn’t qualify.
Possibly the LLM vendors could bias the models more toward nonprofessional content, but then the quality and utility of the output would suffer. Skip the scientific articles and books, focus on rando internet comments, and you’ll end up with a lot more crap than you already get.
Coalgebras might seem too academic but so were monads at some point and now they are everywhere.
https://github.com/ralusek/streamie
allows you to do things like
infiniteRecords
.map(item => doSomeAsyncThing(item), { concurrency: 5 });
infiniteRecords
.map(item => doSomeAsyncSingularThing(item), { concurrency: 5 })
.map(groupOf10 => doSomeBatchThing(groupsOf10), { batchSize: 10 })
// Can flatten back to single items
.map(item => backToSingleItem(item), { flatten: true });The objection is
> The Web streams spec requires promise creation at numerous points — often in hot paths and often invisible to users. Each read() call doesn't just return a promise; internally, the implementation creates additional promises for queue management, pull() coordination, and backpressure signaling.
But that's 95% manageable by altering buffer sizes.
And as for that last 5%....what are you doing with JS to begin with?
I may be naive in asking this, but what leads someone to building high perf data tools in JS? JS doesn't seem to me like it would be the tool of choice for such things
Performance-wise, I get about half the throughput I had with the same processsing done it rust, which doesn't change anything for my use-case.
However that's not really relevant to the context of the post as I'm using node.js streams which are both saner and fast. I'm guessing that the post is relevant to people using server-side runtimes that only implement web streams.
To your question, I was about to point out Firefox[1], but realized you clarified 'mainstream'[2]...
right now when i need to wrangle bytes, i switch languages to Golang. it’s easy gc language, and all its IO is built around BYOB api:
interface Reader { read(b: Uint8Array): [number, Error?] }
you pass in your own Uint8Array allocation (in go terms, []byte), the reader fills at most the entire thing, and returns (bytes filled, error). it’s a fully pull stream API with one method at its core. now, the api gets to be that simple because it’s always sync, and blocks until the reader can fill data into the buffer or returns an error indicating no data available right now.
go has a TeeReader with no buffering - it too just blocks until it can write to the forked stream.
https://pkg.go.dev/io#TeeReader
we can’t do the same api in JS, because go gets to insert `await` wherever it wants with its coroutine/goroutine runtime. but we can dream of such simplicity combined with zero allocation performance.
This is what UDP is for. Everything actually has to be async all the way down and since it’s not, we’ll just completely reimplement the OS and network on top of itself and hey maybe when we’re done with that we can do it a third time to have the cloud of clouds.
The entire stack we’re using right down to the hardware is not fit for purpose and we’re burning our talent and money building these ever more brittle towering abstractions.
A stream API can layer over UDP as well (reading in order of arrival with packet level framing), but such a stream would a bit weird and incompatible with many stream consumers (e.g. [de]compression). A UDP API is simpler and more naturally event (packet) oriented. The concepts don’t mix well.
Still, it would be nice if they browser supported a UDP API instead of the weird and heavy DTLS and QUIC immitations.