Alternative to Promise
韩冬
handong05 at meituan.com
Sun Oct 4 16:03:13 UTC 2015
Sorry i messed up with the code format in previous mail, this one should work:
> There are no problems with reusability that I can think of as a result of the internal state machine. The functions passed to `.then` are just plain functions, so they are perfectly reusable.
Sorry, i should be more clear about reusability i’m referring are not the functions you passed, but the Promise itself, suppose we have an async operation readDB, we can construct a web api using Promise like this:
var express = require('express');
var testApp = express()
testApp.get(‘/test', function(req, res) {
new Promise(function(resolve, reject) {
readDB('test', function(data) {
resolve(someProcess(data));
});
})
.then(function(data) {
res.send(data);
});
});
So you’re creating a Promise for every request, using my solution you can reuse Action like this:
var express = require('express');
var testApp = express()
var testAction = new Action(function(cb) {
return readDB(‘test’, function(err, data) {
cb(process(data));
});
});
testApp.get(‘/test', function(req, res){
testAction
.go(function(data) {
res.send(data);
});
});
Every time you call go, the readDB run again, you won’t waste time on recreating whole callback chain.
I even benchmarked such situation, where Action can get close performance to raw callback, Promise will pay about 10~15% punishment depend on readDB cost(this high readDB cost, the lower Promise creation cost), but i’m not going to attack on Promise performance problem anymore, since it’s not meaningful without statistic.
> The state machine approach won't lead to race conditions in a multi-threaded environment because the promise state machine will always live on a single thread. If JavaScript ever gets shared memory multi-threading it will be carefully restricted to prevent this kind of problem.
No doubt on that.
> By contrast, systems built without a carefully engineered state machine (e.g. Thunks and your continuation based system) tend to lead to race conditions in user land code. Once people try and write parallel code or caching/memoised using lazy continuation based systems they quickly end up needing to convert them into some kind of eager data structure. This is often done badly (it's very difficult to get right) and can lead to very hard to debug race conditions. By contrast, promises are very carefully designed and tested implementations of exactly this functionality.
Well, i understand that Promise are javascript’s choice, and i think i understand the reason now, but i really don’t think lazy data structure lead to race conditions ; ), but again i won’t going to defend for lazy data structure without any statistic either.
~winter
> On Oct 4, 2015, at 6:55 PM, Forbes Lindesay <forbes at lindesay.co.uk> wrote:
>
> There are no problems with reusability that I can think of as a result of the internal state machine. The functions passed to `.then` are just plain functions, so they are perfectly reusable.
>
> I don't think there is a significant overhead to promises once properly optimised. I don't see how your solution would lead to lower overhead.
>
> The state machine approach won't lead to race conditions in a multi-threaded environment because the promise state machine will always live on a single thread. If JavaScript ever gets shared memory multi-threading it will be carefully restricted to prevent this kind of problem.
>
> By contrast, systems built without a carefully engineered state machine (e.g. Thunks and your continuation based system) tend to lead to race conditions in user land code. Once people try and write parallel code or caching/memoised using lazy continuation based systems they quickly end up needing to convert them into some kind of eager data structure. This is often done badly (it's very difficult to get right) and can lead to very hard to debug race conditions. By contrast, promises are very carefully designed and tested implementations of exactly this functionality.
>
> _____________________________
> From: 韩冬 <handong05 at meituan.com <mailto:handong05 at meituan.com>>
> Sent: Saturday, October 3, 2015 1:04 p.m.
> Subject: Re: Alternative to Promise
> To: Benjamin Gruenbaum <benjamingr at gmail.com <mailto:benjamingr at gmail.com>>
> Cc: <es-discuss at mozilla.org <mailto:es-discuss at mozilla.org>>
>
>
> Hey Benjamin
>
> I want to know more about the implementation about Promise after two day of research, there’re two different ways of implementing a chain style control structure, one is based on an internal state machine, which save the callback for a moment, and resolve them after async function finish, the other one is based on continuation, every node on the chain are a new continuation contain the computation on the chain, some kind of porting the ConT monad from haskell to javascript, i’d like to compare them and get to know why the state-machine based solution eventually won?
>
> Here is my summary:
> Pros for state machine based solutions:
> Auto memorization.
> Easy sementics.
> Cons for state machine based solutions:
> Bad reusability.
> Larger overhead.
> Pros for continuation based solutions:
> Good reusability, since continuation are just functions.
> Lower overhead.
> Cons for continuation based solutions:
> Complex sementics.
> No memorization(can be done by other ways).
> Do you agree with me on this summary? and suppose in future javascript will get multicore support, will the state-machine based solution subject to race condition?
>
> Thanks again for giving me lots of detail about the history, now i need more : )
>
> On Oct 1, 2015, at 4:42 PM, Benjamin Gruenbaum < benjamingr at gmail.com <mailto:benjamingr at gmail.com>> wrote:
>
>
> > Where do you get the courage to challenge every inventor that they have to learn everything you've learned before they making decisions?
>
> Can we please keep it civil?
>
> > the question is why not check other languages first, when there’re nice solutions already there.
>
> Promises are rooted in the 1980s and have been pretty much adopted in every mainstream programming language one way or another:
>
> - Task - C#
> - Future - Scala
> - Deferred - Python
> - CompletableFuture - Java
> - Future - C++
>
> And so on and so on. The technical committee also includes people who pioneered the concept. Practically everyone on this list knows Haskell, and ConT isn't really anything new to any of us. We can all explore various alternatives that are the continuation monad (http://blog.sigfpe.com/2008/12/mother-of-all-monads.html <http://blog.sigfpe.com/2008/12/mother-of-all-monads.html>) all day long - but JavaScript already has continuations through promises and they are already in the standard so there is absolutely zero chance they'll be "swapped" for something else at this point.
>
> There are about 3 years of discussions to read about the choices and the rationale for why promises behave exactly the way they behave and you're welcome to read those on the specific choices.
>
> If you're interested in current proposals for other async primitives for the language - there is currently an observable proposal and an async iterator proposal - both solve different issues than promises (multiple values over push/pull) and are currently in design stages.
>
> In general, the list frowns upon people who "plug their library" in the list - so I suggest that in the future you start your email with the specific problem you want to address and what you do differently. The more concise you write (and external links aren't that great for this) the better chance you'll get more responses from people involved.
>
> Cheers and good luck,
> Benjamin
> _______________________________________________
> es-discuss mailing list
> es-discuss at mozilla.org <mailto:es-discuss at mozilla.org>
> https://mail.mozilla.org/listinfo/es-discuss <https://mail.mozilla.org/listinfo/es-discuss>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.mozilla.org/pipermail/es-discuss/attachments/20151005/b34267ba/attachment.html>
More information about the es-discuss
mailing list