[bilalq]

I disagree. I'd take JS-style promises over trying to manage Futures in ForkJoinPools or thread pools any day. Being able to write async expressions in parallel by default means even junior devs take advantage of parallelism. I've seen plenty of code written in Java and Ruby where multiple network and DB requests are made in serial despite having no dependency on each other. The usual reason is that there's just a lot more friction to have it be parallel there.

[arp242]

I have also found Promises very hard to reason about at times: "okay, so I have some code here, and when exactly is this run?" can be a difficult question to answer with Promises. Part of that is inherent in async code, but part of that is also because IMHO Promises make it harder than it needs to be.

I never really used Java, but I have used Ruby and Python (IIRC Python's APIs were modelled on the Java ones) and I agree it can be painful. The thing is, even with an awkward async implementation it's something you have to deal with relatively infrequently when synchronous is the default (as it is in most languages). When you do it can be a pain, but I'd rather have this "occasional pain" vs. "pain every time I want to do any I/O operation".

Personally I like how Go does things.

[ttfkam]

My code is running <- no await

My code is accessing another resource (storage, network, etc.) <- await

It's really not that complicated. If you're surprised by the presence of absence of a Promise, you might want to take a moment to understand what is being processed. There's a good chance there's a gap there that extends beyond a simple keyword in JS.

[peferron]

> I have some code here, and when exactly is this run?

If there are `await` keywords previously in the function, then the line you're looking at will run after these async calls are done. Otherwise it'll run ASAP. Is there something else to it?

[consilient]

People often get confused because they expect `await` to sequence promise resolution too. For example

    const example = async () => {
        const ifError = Promise.reject("something went wrong")
        const value = await someOtherPromise()
        await (valueIsOk(value) ? runNextStep(value) : ifError)
    }

will always throw.

[bilalq]

I don't think I follow. Your example left out all the definitions of these functions, so you can't really deterministically say what will happen. If `someOtherPromise()` fulfills, `valueIsOk(value)` evaluates to `true` or truthy, and `runNextStep(value)` fulfills, `example` will fulfill and not reject. If any of those conditions don't hold, `example` settles as rejected.

[consilient]

The issue is that `ifError` throws whether `example` fulfills or not. Promised values are sequenced by `async`, but promise side-effects are sequenced like side effects of any other javascript statement.

[bilalq]

Sure, ifError rejects, but I don't think the behavior here is surprising or strange at all. This is exactly how one would want it to work. If you wanted to await it, you could do that.

Is the concern you're raising that people may accidentally orphan floating promises? That can be addressed with linter rules. [1][2]

[1]: https://github.com/typescript-eslint/typescript-eslint/blob/...

[2]: https://github.com/typescript-eslint/typescript-eslint/blob/...

[sajex]

Async doesn't give you parallelism by default though, you just get concurrency. You don't get parallelism without using Workers.

[hombre_fatal]

Well, I wouldn't put it like that. Async can trivially cash out into parallel work like if you're just sending queries to a database or ffmpeg or ImageMagick or, frankly, most common use cases which are going to do work in parallel out of process.

All of your I/O work could even be happening in parallel in a run of the mill JS app. Workers just give you parallelism with your sync JS code which is a more narrow claim.

[bilalq]

Fair, I should've been more careful with the way I worded that. The common example I was highlighting was concurrent network IO requests that effectively resolve to parallel work that runs on different nodes. With a service oriented architecture, this can be the norm rather than the exception.

[ttfkam]

And workers get you isolation, no shared memory. You must explicitly pass data ownership from one thread to the next. (And I consider all of that a good thing.)