How to convert Node.js async streaming callback into an async generator?

Question

I have a function that streams data in batches via a callback.

Each batch will await the callback function before fetching another batch and the entire function returns a promise that resolves when all batches are finished.

(I'm using TypeScript annotations to help with readability)

async function callbackStream(fn: (batch: Array<number>) => Promise<void>) {}

How do I to turn this function into an async generator that yields one value at a time?

async function* generatorStream(): AsyncIterableIterator<number> {}

This has proven to be quite a difficult task.

I've toyed around with this problem and I've built something that works, but its very convoluted and I can't justify merging this code and making others on my team deal with it.

---

Here's my current implementation:

I'm using this helper function that created a "deferred" promise which helps with passing promises around callbacks.

interface DeferredPromise<T> {
    resolve: (value: T) => void
    reject: (error: any) => void
    promise: Promise<T>
}

function deferred<T>(): DeferredPromise<T> {
    let resolve
    let reject
    const promise = new Promise<T>((res, rej) => {
        resolve = res
        reject = rej
    })
    return {
        resolve: resolve as (value: T) => void,
        reject: reject as (error: any) => void,
        promise,
    }
}

Next I have this hairball of logic that linearizes the promise callbacks into a chain where each promise resolved a batch with next function that will return another promise fetching the next batch.

type Done = { done: true }
type More = { done: false; value: Array<number>; next: () => Promise<Result> }
type Result = More | Done

async function chainedPromises() {
    let deferred = PromiseUtils.deferred<Result>()

    callbackStream(async batch => {
        const next = PromiseUtils.deferred<null>()
        deferred.resolve({
            done: false,
            value: batch,
            next: () => {
                deferred = PromiseUtils.deferred<Result>()
                next.resolve(null)
                return deferred.promise
            },
        })
        await next.promise
    }).then(() => {
        deferred.resolve({ done: true })
    })

    return deferred.promise
}

From here, creating a generator that yields one item at a time isn't very difficult:

async function* generatorStream(): AsyncIterableIterator<number> {
    let next = chainedPromises
    while (true) {
        const result = await next()
        if (result.done) {
            return
        }
        for (const item of result.value) {
            yield item
        }
        next = result.next
    }
}

---

I think we can all agree that the intermediate chainedPromises function is very confusing and convoluted. Is there any way I can transform callbackStream into generatorStream in a way that is easy to understand and easy to follow? I don't mind using a library if its well established, but I would also appreciate a simple implementation from first-principles.

Answer 1

No, I don't think there's a way to implement this transformation in a way that's easy to understand and easy to follow. However, I would recommend to drop the deferred s (you're never reject ing anyway) and just use the promise constructor. Also I'd rather implement an asynchronous generator right away.

function queue() {
    let resolve = () => {};
    const q = {
        put() {
            resolve();
            q.promise = new Promise(r => { resolve = r; });
        },
        promise: null,
    }
    q.put(); // generate first promise
    return q;
}
function toAsyncIterator(callbackStream) {
    const query = queue();
    const result = queue();
    const end = callbackStream(batch => {
        result.put(batch);
        return query.promise;
    }).then(value => ({value, done: true}));
    end.catch(e => void e); // prevent unhandled promise rejection warnings
    return {
        [Symbol.asyncIterator]() { return this; },
        next(x) {
            query.put(x);
            return Promise.race([
                end,
                result.promise.then(value => ({value, done:false})
            ]);
        }
    }
}
async function* batchToAsyncIterator(batchCallbackStream) {
    for await (const batch of toAsyncIterator(batchCallbackStream)) {
        // for (const val of batch) yield val;
        // or simpler:
        yield* batch;
    }
}

Answer 2

You need a event bucket, here is an example:

function bucket() {
  const stack = [],
                iterate = bucket();
    
  var next;
  
  async function * bucket() {
        while (true) {
            yield new Promise((res) => {
                if (stack.length > 0) {
                    return res(stack.shift());
                }

                next = res;
            });
        }
  }  
  
  iterate.push = (itm) => {
    if (next) {
      next(itm);
      next = false;
      return;
    }

    stack.push(itm);
  }  
  
  return iterate;
}

;(async function() {
  let evts = new bucket();

  setInterval(()=>{
    evts.push(Date.now());
    evts.push(Date.now() + '++');
  }, 1000);

  for await (let evt of evts) {
    console.log(evt);
  }
})();

Answer 3

Would it work if there will be typescript solution?

It should handle condition when callback is called faster then promise is resolved a couple of times. Callback can be a method that has this signature callback(error, result, index) It is set to finish when callback is called with no arguments. Usage:

asAsyncOf(this.storage, this.storage.each);

Solution:

function asAsyncOf<T1, T2, T3, T4, Y>(c, fn: { (a: T1, a1: T2, a2: T3, a3: T4, cb: { (err?, res?: Y, index?: number): boolean }): void }, a: T1, a1: T2, a2: T3, a3: T4): AsyncGenerator<Y>
function asAsyncOf<T1, T2, T3, Y>(c, fn: { (a: T1, a1: T2, a2: T3, cb: { (err?, res?: Y, index?: number): boolean }): void }, a: T1, a1: T2, a3: T3): AsyncGenerator<Y>
function asAsyncOf<T1, T2, Y>(c, fn: { (a: T1, a1: T2, cb: {(err?, res?: Y, index?: number): boolean}): void}, a: T1, a1: T2): AsyncGenerator<Y>
function asAsyncOf<T, Y>(c, fn: { (a: T, cb: { (err?, res?: Y, index?: number): boolean }): void }, a: T): AsyncGenerator<Y>
function asAsyncOf<Y>(c, fn: { (cb: {(err?, res?: Y, index?: number): boolean}): void}): AsyncGenerator<Y>
async function* asAsyncOf(context, fn, ...args) {
    let next = (result?) => { };
    let fail = (err) => { };
    let finish = {};
    const items = [];
    let started = true;
    try {
        fn.apply(context, [...args, function (err, result, index) {
            const nextArgs = [].slice.call(arguments, 0);
            if (nextArgs.length === 0) {
                started = false;
                next(finish);
                return true;
            }
            if (err) {
                fail(err);
                return true;
            }
            items.push(result);
            next(result);
        }]);
    } catch (ex) {
        fail(ex);
    }
    while (true) {
        const promise = started ? new Promise((resolve, error) => {
            next = resolve;
            fail = error;
        }) : Promise.resolve(finish);
        const record = await promise;
        if (record === finish) {
            while (items.length) {
                const item = items.shift();
                yield item;
            }
            return;
        }
        while (items.length) {
            const item = items.shift();
            yield item;
        }
    }
}
export { asAsyncOf };