events: support once with promise

[benjamingr]

This commit introduces a once overload which returns a promise if a listener isn't passed to once.

This is a pretty common feature request and was constantly brought up when I asked users how we can improve our experience with event emitters for people using async/await.

I have decided to not add a new method but instead overload .once because it was safe to do so given the current behavior. Alternatives for this include adding a oncePromise method.

Ref: #20893

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• documentation is changed or added
• commit message follows commit guidelines

[bnoordhuis]

See #18646. Not 100% the same thing but close enough.

Apropos the PR: overloading the method like this introduces potential for bugs if the caller thinks it's passing a function value when it's not.

[benjamingr]

@bnoordhuis thanks, that PR seems to be about an unrelated feature (method to defer emits) - I'm actually -1 on adding that and +1 on adding this.

Apropos the PR: overloading the method like this introduces potential for bugs if the caller thinks it's passing a function value when it's not.

What do you think a more obvious-for-users behavior would be?

What about still throwing if a second argument is passed (arguments.length > 1) but it is not a function?

[targos]

Previous attempt for something similar: #15204

[benjamingr]

@targos oh yeah - I guess the main difference in my opinion is that this PR uses clearer (IMO) naming which was a primary source of concern in the old PR.

In addition, I have seen a year's worth more of users asking for this - we wanted to see if this is something people were doing and I became convinced we should do this.

Also cc @jasnell

[kirly-af]

Hi @benjamingr, first thank you for your interest in this feature and for opening this PR.

I personally think it would make more sense to add a separate method to avoid confusion with the existing once() method. Btw oncePromise was a terrible name picked in 30 seconds, I'm sure we can do better.

I also believe this should reject the returned Promise in case the "error" event is fired.
The whole point of Promises is to have a nice error handling mechanism for asynchronous code. Paired with async/await that would allow us to do:

constmyEmitter=newEventEmitter()functionfoo(){setTimeout(()=>myEmitter.emit('error',newError('foo')),500)}functionbar(){console.log('bar')}(async()=>{try{foo()awaitmyEmitter.oncePromise('ready')bar()// never gets called}catch(err){console.log(err)// "foo"}})()

As a reminder, on that point @benjamingr opinion was:

I don't think we should do that since event emitters are very general and also because when there is no error event handler Node.js terminates by default - so adding an event handler for error implicitly would be very surprising behavior in Node.

I agree event emitters are very generic, though the "error" event is definitely a special one. However when working with Promises you'd expect to be able to handle errors with the Promise API (.catch(), or try/catch when using await). In case no handler is attached, that would result in an unhandled rejection which is what I would expect when working with any Promise based API. Whether or not terminates the process is directly linked to unhandled rejections behavior.

If error events are handled separately, I don't even see the point of using Promises at all.

constmyEmitter=newEventEmitter()functionfoo(){setTimeout(()=>myEmitter.emit('error',newError('foo')),500)}functionbar(){console.log('bar')}(async()=>{try{foo()// where should I handle the emitted error?awaitmyEmitter.oncePromise('ready')// this will await for everbar()}catch(err){// we are losing this}})()

Of course the event handlers should be removed once the Promise is fulfilled or rejected:

constEventEmitter=require('events')EventEmitter.prototype.oncePromise=function(completeEvent,errorEvent='error'){returnnewPromise((resolve,reject)=>{constonComplete=(value)=>{this.removeListener(errorEvent,onError)resolve(value)}constonError=(err)=>{this.removeListener(completeEvent,onComplete)reject(err)}this.once(completeEvent,onComplete).once(errorEvent,onError)})}

Note that I made the error event overridable here but I'm not even sure this is necessary.

[benjamingr]

I also believe this should reject the returned Promise in case the "error" event is fired.

I am regretfully -1 on any PR with that sort of semantic. Namely because I've seen code like the below "in the wild" in production applications:

varrequests=newEventEmitter();waitForAnalytic().catch(()=>{/* explicitly ignore if this errors */});asyncfunctionwaitForAnalytic(){awaitemitter.once('user-initialized');// using a user module here since we don't have this yetconsole.log('user initialized, reporting analytic');// nothing business critical here }

If an "important" event is thrown, the code would swallow the error and in my opinion the behavior would be confusing to users. You can of course .once('error') and handle it explicitly if you would like.

In general, I am -1 on adding event handlers for any event implicitly, I can see this working for streams (with Symbol.asyncIterator handling this) but not for a promisified once.

---

Regarding the naming - the problem with the old name (when) was that it did not communicate clearly that the event is only happening once which was a major concern since a lot of people (especially new users) aren't really too familiar with the more complex promise semantics and realize that promises are a one time value thing - .once communicates it clearly IMO.

[kirly-af]

@benjamingr I don't care much about the name itself, I'm just saying it should be two different methods as the behavior I'm describing is indeed a bit different from the regular .once().

Concerning your example, how is that different from someone purposely swallowing error events?

var emitter = new EventEmitter();
emitter.on('erro', () => { /* explicitly ignore if this errors */ });
async function waitForAnalytic() {
await emitter.oncePromise('user-initialized'); // using a user module here since we don't have this yet
console.log('user initialized, reporting analytic');
// nothing business critical here }

Error swallowing is a bad practice, whatever it'd be vanilla exceptions, rejections or error events.

Having a different method name would make the behavior difference easy to notice, and as long as that behavior is documented IMHO there's no problem.

Now let's say we go with your approach. That means to get the try/catch construct work well I have to write something like that: (that wouldn't even work actually has the "error" promise might not resolve).

constmyEmitter=newEventEmitter()functionfoo(){setTimeout(()=>myEmitter.emit('error',newError('foo')),500)}functionbar(){console.log('bar')}(async()=>{try{foo()// where should I handle the emitted error?awaitPromise.all([myEmitter.oncePromise('ready'),myEmitter.oncePromise('error').then(err=>Promise.reject(err))])bar()}catch(err){console.error(err)// foo}})()

I think such code would be un-intuitive and cumbersome.

[benjamingr]

Concerning your example, how is that different from someone purposely swallowing error events?

Good question. Mostly the expectation and explicitness.

If someone would like to ignore errors I want it to be as explicit as possible. I don't want anything that doesn't make it obvious that it adds an error handler to add an error handler.

Now let's say we go with your approach. That means to get the try/catch construct work well I have to write something like that:

Well, in the above code you would get more correct behavior by not listening to the error event at all and letting Node.js handle it automatically. You would even get better logs and a core dump for the error.

[mscdex]

In general this doesn't feel quite right because we're just promisifying a single method/scenario.

However, even if we added this feature, I too believe we should not be housing it under an existing method. Having it under a different name would easily make this semver-minor I think. Currently it could be viewed as semver-major?

[benjamingr]

In general this doesn't feel quite right because we're just promisifying a single method/scenario.

Well, for event emitters this is the single method/scenario that makes sense to promisify - streams deal with this better (as async iterators) but for event emitters anything other than once doesn't make sense for promises.

I do hopefully we have Symbol.observable (or Symbol.observer) and interop with observables when that comes and when/if that makes sense.

Alternatively, we can also make an on that returns an async iterator.

Having it under a different name would easily make this semver-minor I think. Currently it could be viewed as semver-major?

I'm less concerned about semver-majorness vs. minorness - I'm more interested in figuring out whether other collaborators thing this is a good idea at all since I changed my mind about it.

I don't think we talk about these things enough :)

I too believe we should not be housing it under an existing method.

Do you have any naming suggestions?

[jasnell]

-1 on polymorphic returns, way too much of a usability footgun.

If this is going to happen, a separate method would be preferred but even then I'm really questioning the need for this. I've looked into it several times in the past and could not come up with a good general use case that would justify it having to be in core. It's something that can easily be done in userland, there is no real performance benefit to be realized having it in core, and the use cases appear to be rather limited.

[jasnell]

Marking semver-major because of the addition of the polymorphic return and the change to the error case. It may not be likely, but there is a non-zero chance of the polymorphic return breaking existing code.

[benjamingr]

I've looked into it several times in the past and could not come up with a good general use case that would justify it having to be in core.

Well, users have been asking for it a lot which is why I thought it makes sense to bring it up again for discussion. The most common use case people are asking for is when they need to wait for a "I've started" event inside an async function - basically when event emitters are used both for starting something and for sending data over.

It's something that can easily be done in userland, there is no real performance benefit to be realized having it in core, and the use cases appear to be rather limited.

Well, it simplifies the API usage for every event emitter we ship and aligns it with how users write coded in practice.

As for performance - I agree, it might be possible to motivate this with performance but this is not that PR.

-1 on polymorphic returns, way too much of a usability footgun.

Why do you believe that this is a foot gun? I'm honestly asking - I have no strong feelings either way.

[jasnell]

I'm not going to block it on the grounds that I don't think it's necessary. That's a generally subjective point-of-view. I'll remove my red-x if this were added as a separate method rather than implemented as a polymorphic return.

[jasnell]

Why do you believe that this is a foot gun?

Through hard experience I've learned that not every event emitter used in the ecosystem today extends from the require('events').EventEmitter class. There are roll-your-own implementations in use by the ecosystem today that are specifically designed with isomorphic support in browser environments. Those implementations would not automatically pick this variant of EventEmitter#once up and if we started to see adoption of this those implementations would fail. Further, it is simply not possible to polyfill without monkeypatching the existing method.

By implementing as a separate method, we avoid changing the existing contract with once, make it easier to detect support for the new feature, and make it easier to polyfill... e.g.

EventEmitter.prototype.oncePromise=EventEmitter.prototype.oncePromise||functiononcePromise(){/** ... **/}

[benjamingr]

I'm not going to block it on the grounds that I don't think it's necessary. That's a generally subjective point-of-view. I'll remove my red-x if this were added as a separate method rather than implemented as a polymorphic return.

I'm less concerned with whether or not this particular PR lands or not - I'm more interested in whether or not we should generally do this.

If you believe this isn't useful or that the use cases above aren't interesting (or are but aren't worth it) do speak up. I don't have a horse in this race.

By implementing as a separate method, we avoid changing the existing contract with once, make it easier to detect support for the new feature, and make it easier to polyfill... e.g.

Got it, that makes sense. I hadn't considered this makes it significantly harder to polyfill.

What about waitFor as an alias?

[addaleax]

I’m okay with this – it might be time for us to realize it’s more important that we have this feature in some way than what exactly the best way is.

I think once() is a succinct name and I don’t really understand how this is semver-major, given that what we previously generated is a type checking exception.

[jasnell]

@addaleax ... it's semver-major because it changes the contract of the existing method and breaks compatibility with, for example, things like eventemitter3, which has over 3+ million weekly downloads.

Overloading once() is not a good idea. Implementing as a separate function is cheap and non-disruptive and would make this semver-minor.

[benjamingr]

I've made a PR at eventemitter3 to get some discussion started.

I also have no issue with renaming to waitFor, I think it's slightly less ergonomic but I'm on board if overloading is a pain for libraries in users.

Going to wait for suggestions regarding the name from @jasnell and @mscdex as well as feedback from eventemitter3 (I am in no rush here)

[mscdex]

@benjamingr I don't have any suggestions for a new name, I am just saying we should not be housing this new functionality under the existing once().

[benjamingr]

@mscdex thanks, you have raised two concerns - one regarding the name which I can resolve with waitFor and the other was

In general this doesn't feel quite right because we're just promisifying a single method/scenario.

To which I responded in #20909 (comment)

Could you please let me know if that addresses the concerns raised there?

I’d ideally want users of Node to never have to write new Promise which is harder to do correctly for a lot of people (I'm one of those people) - for example I’ve seen a once with a nasty bug where the promise was resolved in the handler, then the handler threw and the exception was swallowed causing a leak.

async function sample() {
await new Promise(resolve => ee.on('got-item', item => {
resolve(); // finish the await
// this throws synchronously if `item` === `foo` and leaves a database connection open
// after several millions of request, the user who asked for help with this bug last week
// got issues with the server and had a very hard time to figure out why. reportAnalytic(item); });
}

I’m also not sure this PR is the right way to address my concern about it.

[vsemozhetbyt]

Nit: // Verify the promise returning version of `once`. (uppercase and period)?

[kirly-af]

Hi @benjamingr,

Some answers on that comment (#20909 (comment)):

Mostly the expectation and explicitness.

• On expectation: Promises are made to be rejected. I don't think that would be a problem.

• On explicitness: So what if we make the error handling explicit?

try{client.connect()awaitclient.waitFor('connected',{errorEvents: ['serverError','readError']})}catch(err){// handle my server error}

Once again my concern is we need a chance to handle errors using the Promise API.

// what about the async flow wmwhen in a Promise chain or with await ?client.on('error',(err)=>handleError(err))try{client.connect()awaitclient.waitFor('connected')// pending foreverdoOtherStuff();}catch(err){// no way to actually use the catch construct}

How would you handle this kind of cases?
Imagine a much worse scenario were we make multiple attempts:

emitter.on('error',(err)=>{handleError(err)setTimeout(()=>connect(),500)})asyncfunctionconnect(){try{client.connect()awaitclient.waitFor('connected')// pending foreverdoOtherStuff();}catch(err){// no way to actually use the catch construct}}

Promises are not cancellable so we have no control over them once they're created. Let's say we make 10 attempts before successfully connect to the server. That means we have 9 Promises that will be pending forever without any way of deleting them.

[mscdex]

for example I’ve seen a once with a nasty bug where the promise was resolved in the handler, then the handler threw and the exception was swallowed causing a leak.

@benjamingr That sounds like a user error, not something that could/should be solved in core?

[benjamingr]

I would like to wait with renaming and landing this until after the collaborator summit if that's OK with everyone since I'm going to discuss some use cases which are related in the promises session.

Afterwards my follow-up plan is to rename this (addressing mscdex and jasnell's concerns) and add some explanation in the PR from the discussion above motivating why expecting users to do new Promise encourages dangerous code.

---

@DaAitch thank you for weighing in on this issue and sorry for missing your comment earlier. I motivate why I believe this is needed in #20909 (comment) and the discussion following it. I am going to discuss various cases I've seen it implemented dangerously and indicate we need a much better async story in the summit too :)

The fact you love promises but are concerned they are not low-level enough is very interesting to me. I would love to discuss why that is (you can send me an email or open an issue in https://github.com/nodejs/promise-use-cases/tree/master/use-cases ).

I'm not sure I'd call promises a "best practice" or say that they're a higher level primitive than event emitters. I think they're just different, have different uses and different cases where they apply.

I personally find promises useful and they are a language feature - so I think Node.js should try reasonably to behave in a way that doesn't make users take shortcuts since promises are hard and when people use them "manually" with new Promise they tend to make mistakes. Over 90% of people use promises and async/await in recent polls.

[DaAitch]

@benjamingr Thank you for your answer. I think you mixed up my answer a bit.
The first aspect should be clear, but the second was, that all semantics we know about Promise is not quite matching EventEmitter.once semantics in my opinion.

I'm not sure I'd call promises a "best practice" or say that they're a higher level primitive than event emitters. I think they're just different, have different uses and different cases where they apply.

Look at this eslint plugin and rule promise/catch-or-return. It says, that a promise has to be returned, or end with a .catch(...): that is what I also call a best/good practice to not have unrejected promises, but all errors handled.
The fact that the PR only wraps a callback in a Promise but there is no fail-case can be very confusing for those appending their promise chains with .catch(...). And since there literally is not fail-case with EventEmitter events I think semantically they are also not matching.

I think Node.js should try reasonably to behave in a way that doesn't make users take shortcuts since promises are hard and when people use them "manually" with new Promise they tend to make mistakes.

That's exactly why something like an EventEmitter should use primitive types like strings and callbacks and no Promise in my opinion. If we start promisifying core here, where/when do we stop? We'd always have to provide both impls. Is this the right way of the async/await story? In my opinion it is not.

[benjamingr]

Thanks for following up:

that all semantics we know about Promise is not quite matching EventEmitter.once semantics in my opinion.

This proposal is going to change a little and not called once, and we will survey users to make sure it makes sense based on what they are expecting. I've seen this implemented at least a hundred times by now and pretty much all of them had similar semantics. I've also validated this with a GitHub code search but sadly I didn't save that research. If you want to look into it and make a summary it would be appreciated (only if you feel like it :))

Look at this eslint plugin and rule promise/catch-or-return

Well, that's just a userland plugin - I happen to agree with its premise but it's one way to write code. I personally definitely think that a promise should be returned if it is created and we'll discuss it in the summit :)

The fact that the PR only wraps a callback in a Promise but there is no fail-case can be very confusing for those appending their promise chains with .catch(...).

Let me ask you a question - is it OK to have a synchronous function that never throws or doesn't throw?

Promises parallel regular functions, a function that doesn't return a promise that rejects is like a function that never throws. There are lots of such functions (Math.random, Math.imul etc).

If we start promisifying core here, where/when do we stop?

I think we stop when users stop writing new Promise in their code since that has proven to be confusing and error prone - there are several examples I linked to in my previous comment.

Note that we already convert some event emitters (readable streams for instance) to async iterators.

[kirly-af]

@TimothyGu I agree with you that we need a way to catch error events, though IMO errorPromise should be a method rather than a property because otherwise we'll end up having this kind of behavior:

emitter.emit('error',newError())// from now on errorPromise is rejected for ever// ...constresult=awaitPromise.race([emitter.oncePromise('done'),// can't use oncePromise anymoreemitter.errorPromise]);

Instead, what would be more useful is:

emitter.emit('error')// not related to any promise so far// ...constresult=awaitPromise.race([emitter.oncePromise('done'),emitter.errorPromise()// reject on error events fired from now on]);

Possibly it would be even more useful to make errorPromise accept an event name as argument.

However, in the (likely) case no error event is fired, each call to errorPromise will generate a pending promise that will never reject. I guess that could be considered a leak. Does anybody know the cost of pending promises?

[benjamingr]

Does anybody know the cost of pending promises?

The problem isn't the cost of pending promises (they're virtually free) - the problem is that once we've added an error handler then Node.js will no longer crash if the event emitter errors.

[kirly-af]

@benjamingr Oh yeah I forgot that. It seems clear we only have 2 options:

1. we don't care about rejection => straight forward promisification of .once()
2. we want to handle error events => wrap the event and error event .once() in the same promise. I describe what in my mind in the last snippet of events: support once with promise #20909 (comment) (without the default error event value)

I wonder what's the general opinion about these two options (@benjamingr is not a big fan of number 2). Anybody has an alternate version in mind?

[benjamingr]

@kirly-af thanks for your comment.

I'll try to clarify my position. I am trying to do something that's more useful for users. We have promises (awesome) but users use them with the promise constructor (lame) which they should never do.

I don't think the promise constructor is to blame here - a different construction API would have different woes. I think manual construction of promises is inherently hard and that not many people get it right.

I'm trying to solve what's most painful to users - I want to avoid cases where users have hard-to-debug surprises in their APIs.

My objection is to any solution that interrupts with the fact event-emitter errors default to crashing Node.js. If I .once to one event and another unrelated event causes an error - I'm scared users will assume the error is related to the event onced and handle the error erronously leaving the server in an indeterminate state.

I would be interested in exploring paths that improve the error handling and debugging experience of our users.

By the way - thanks for opening the issue here (and in bluebird before) and participating in the discussion, I appreciate it.

[dead-claudia]

BTW, I just found/read this stuff, and the ensuing related discussion evolved basically into my #21339, mod name only. (I filed the bug before finding this.) Oh, and also, this code very clearly leaks:

emitter.once("event",resolve).once("error",reject)

Assuming only one of them is called (the common case), this will resolve/reject with that first one, but the second listener will still remain strongly linked to by the emitter, and thus the promise state will have to be retained, so the other function knows to ignore its input.

---

@kirly-af With your two options, my #21339 is basically your second: combining the two into the same promise. I chose there to use on rather than once as the basis for mine, since the once boilerplate basically triples the number of closures that need allocated (one for each listener and one to deregister the other's listener in each), with no real value to it.

@benjamingr Most cases where I've actually needed to await "error" with another event, there's no chance for overlap. The most common case for me by far is with fs.createReadStream and fs.createWriteStream, where you have to register both "error" and "ready" to know whether you can write to it or not, and if so, when. It's a crappy API IMHO, but it's easier to do this hack and just wrap the API, since fs/promises doesn't offer anything to alleviate this issue at all. (If you decided to add promisified versions of those two methods that address those two issues, I wouldn't complain, but there's still problematic libraries.) There's also the drain event on writable streams - if you have a lot of sequential data (like with a generated file or a massive amount of console output), but you don't want to risk memory issues or truncated output in the case of console output, you need to await this event before writing more. And the promise wrapper for this ends up looking very similar to the standard one you would write for opening a stream, too.

[benjamingr]

The most common case for me by far is with fs.createReadStream and fs.createWriteStream

That can be done with Symbol.asyncIterator or for await, we already have streams covered :)

This is explicitly for non-stream event emitters.

[dead-claudia]

@benjamingr I'm aware that's possible for readable streams (it doesn't flow, missing out on that optimization), but that's not so much the case with writable streams.

And I'm specifically talking about "ready", which occurs before you can do any meaningful writing to it. So even though it's about streams and certain stages, my issue isn't generally with the streaming itself. If this ASCII chart helps explain what I'm talking about:

 /--> error
|
create --+--> ready <------- I can't write before this fires, so I need to await
| it. I also can't write if it errors.
\|/
write <--+--> error
| |
| |
/-----> | | <------- If I write here, it's unnecessarily buffered.
| | | I also can't keep writing if it errors.
| \|/ |
| drain ---/
|
|
Normal streaming, outside the scope of
this bug (and my feature request).

Awaiting at those two points just makes sense to me, and I'm already rolling my own abstraction repeatedly which is basically this proposal.

[benjamingr]

I would be very interested in an abstraction that would improve the async story for writable streams - looking forward to see what you come up with!

[dead-claudia]

@benjamingr My now-closed-as-a-duplicate #21339 was pretty much that. A pragmatic solution to it I'm already using.

[ChALkeR]

I have to use something like this in userland for something that uses event emitters with promises:
https://github.com/ChALkeR/Gzemnid/blob/daf4e2580/src/helpers.js#L34-L48

It's basically obj.promise('end', 'error'). It could be trivially extended to support arrays of events, like: obj.promise(['end','close'], 'error').

I think that promise-based .once alternative should support specifying events for accept and reject, else it won't be much useful — what's the point in awaiting a promise that could never fire (and emit an error instead)?

@isiahmeadows I believe that the impl linked above doesn't leak.

That doesn't fix the streams usecase, though — drain events are best consumed as they are, without promises, imo.

[ChALkeR]

Making it explicit, see concern above about this approach not supporting rejection (which could hang).

I like the direction where this goes, but would prefer to see something closer to .promise('end', 'error') (possibly with support for specifying multiple events in an array).

[benjamingr]

@ChALkeR what sort of API would you like to see? What userland prior work can I research?

[ChALkeR]

@benjamingr I left a link in my comment above for an example. I am not aware of major userland right now.

Copy-pasting here with some changes to support arrays (sorry for the poor formatting, typing it directly in a comment, also not tested):

obj.promise=function(finish,error){if(!Array.isArray(finish)){if(finish===undefined)finish=[];elseif(typeoffinish!=='string')thrownewError('invalid arguments');elsefinish=[finish];}if(!Array.isArray(error)){if(error===undefined)error=[];elseif(typeoferror!=='string')thrownewError('invalid arguments');elseerror=[error];}if(finish.length+error.length===0)thrownewError('invalid arguments');returnnewPromise((accept,reject)=>{letacceptWrap=null,rejectWrap=null;constcleanup=()=>{for(consteventoffinish)obj.removeListener(event,acceptWrap);for(consteventoferror)obj.removeListener(event,rejectWrap);};acceptWrap=(...args)=>{cleanup();accept(...args);};rejectWrap=(...args)=>{cleanup();reject(...args);};for(consteventoffinish)obj.once(event,acceptWrap);for(consteventoferror)obj.once(event,rejectWrap);});}

[bajtos]

It seems clear we only have 2 options:

1. we don't care about rejection => straight forward promisification of .once()
2. we want to handle error events => wrap the event and error event .once() in the same promise.

In LoopBack, our primary usage of promisified once is to wait until an http/https server is listening (emits 'listening' error). It's all too easy to forget that listen can also fail and error event needs to be listened for too. To us, the second mode is more important. So far, we are pretty happy with the user-land module p-event.

constserver=http.createServer(/*handler*/);server.listen(80);awaitpEvent(server,'listening');// ^^ resolves on success, rejects (throws) on error

Similarly, one may want to wait until the server is stopped:

server.close();awaitpEvent(server,'close');// in most cases, we want to log and discard errors// because there isn't much we can do here, right?

IMHO, waiting for a server to start/stop listening is a common use case that should be kept in consideration when discussing a promisified once API. Of course, a promisified version of httpServer.listen() and .close() would be even better, but until we get that option, I would expect most people to await server.once('listening') because it's the easiest thing to do. If promisified once don't reject the promise on error, this easiest thing to do will also become the wrong thing to do.

[jasnell]

What's the status on this one?

[benjamingr]

Stalled is pretty accurate - I intend to revisit this in about a month

[dead-claudia]

What's the status on this?

Edit: I guess that answers that.

[kirly-af]

Hi @benjamingr
Can you elaborate on why this has been closed ?

[dead-claudia]

@kirly-af Not him, but I suspect it's probably just because it's been stalled for well over 6 months and the API proposal has itself changed somewhat since. The original bug, #20893, is still open, BTW.

[benjamingr]

@kirly-af in all honesty I have a lot going in my personal life (new job, sick family member, buying a house) and working on this is too much in terms of the emotional investment landing it would require.

Doing this isn't too much work, I encourage other people to give this a shot and I can help guide people with it.

Alternatively, I plan to get a lot more active again (coordinated with the job) around the summit (May) and I might tackle this again.