Discuss
  • Do we really need async generators?

Do we really need async generators?

# Axel Rauschmayer (8 years ago)

I’ve thought some more about the async iteration proposal [1] and my thinking has evolved:

  • I find the async iteration protocol and for-await-of useful.
  • But I still suspect that Communicating Sequential Processes (i.e., async functions plus library code) are a simpler solution than async generators.

I’m genuinely interested in finding the best possible approach, so any kind of feedback is welcome – especially if CSP have any kind of deficiency that I have overlooked.

Details: gist.github.com/rauschma/4dc86ea81585fcfe056de3caa19aa38f, gist.github.com/rauschma/4dc86ea81585fcfe056de3caa19aa38f (I’ll probably publish this blog post on Monday or Tuesday)

[1] tc39/proposal-async-iteration

Axel

# Isiah Meadows (8 years ago)

Missed the list...

---------- Forwarded message ---------- From: Isiah Meadows <isiahmeadows at gmail.com>

Date: Mon, Mar 13, 2017 at 12:14 AM Subject: Re: Do we really need async generators? To: Axel Rauschmayer <rauschma at icloud.com>

From my basic reading, this is not that far off of how iterators work

without for ... of. When I don't have those loops available (e.g. working in ES5 code), I've found this pattern to be really helpful in interoperating with ES6 collections:

for (var next = iter.next(); !next.done; next = iter.next()) {
  // use `next.value`
}

That's very similar to your idea of iterating CSP collections:

for (var next = await channel.take(); next !== Channel.DONE; next =
await channel.take()) {
  // use `next`
}

It's also similar to how C++'s iterators worked before C++ 11:

for (std::vector<int>::iterator iter = vec.begin(); iter != vec.end(); iter++) {
    // use `*iter`
}

This will get unwieldy in a hurry due to its verbosity, though, and even C++ added a sugar version to abstract this. Hence the async iterator proposal.

I do feel CSP channels are closer to Node streams and observables than async iterators, though, and your proposal seems awfully similar to this. I do feel that your CSP ideas could really carry over with my non-linear control flow strawman, though (Note that I'm about to redo much of that strawman, to clean it up substantially).

Isiah Meadows me at isiahmeadows.com