I've abandoned it at this point for a very particular reason: operators don't work well in this phrasing. That is, the example case given in the above discussion:
function makeView(templateURI, dataURI) {
var template = readURI(templateURI),
data = readURI(dataURI);
return Mustache.render(template, data);
}
...while it's not bad, also doesn't actually try to "write logic", in a post that's about "write logic, not mechanics."
What you really want to do is much more complicated because the example code starts to look ugly:
Don't get me wrong, I still like the idea and think that there might be a very promising way to do this sort of work in Node. Certainly I prefer those five lines to always writing:
db.getUser(data.session, function (err, user) {
if (err) return callback(err);
if (user.role === "admin") {
db.send(data.request, function (err, reply) {
if (err) return callback(err);
render_file("yay.xml", reply, callback);
});
} else {
render_file("no_permissions.xml", {}, callback);
}
});
But as much as I really don't like that lame error stuff which breaks in a stiff wind, I also really want to be cautious about the exchange of these two lines:
if (user.role === "admin") {
userPromise.attr("role").equals("admin").then(
What we really want, I guess, is some sort of mesoscopic system -- we still want operators and familiar control structures from our "micro" world, but we also want lazy evaluations and so on from our "macro" framework. This is not impossible but it is hard to get right. If my "ducky" project moves forward it will probably look something like this:
You bubble up the lazy constructs with returns, the error handling gets internalized within the lazy constructs, and the "control-method" invocation allows you to build an arbitrary control structure with whatever object you have right now.
it’s possible to make sync function async
it’s not the case other way round
Is this terminology correct? I thought it is easier to make async functions sync by blocking it, and harder to unblock a sync one. Am I confusing this with something else?
I think the OP refers to the syntax/API for each. If you're already using async APIs, adding a sync one that runs in a separate thread won't change much.
On the other hand, you need something like CPS or coroutines to cleanly use an async function in a program mostly using sync APIs.
Of course, if you have the right tools (CPS/coroutines) you can make any async API look sync, as you said.
It's a bit misleading way to put it, but more or less correct when talking about client-side JavaScript. There are essentially no blocking calls, so a "synchronous" function in this context just means a normal function which directly returns a value and "asynchronous" means a function that takes a callback.
What the quoted bit boils down to is that any function that returns normally can be converted into a function that calls a callback, but not the other way around.
Sorry for confusion, I meant indeed a sync / async APIs. You could easily wrap sync function to give it an async API in order to chain with functions that had async APIs, but it's not really possible other way round (not in JS).
I will be down voted for saying so, but "than" and "then" seem to be reversed throughout the entire article. I would like to learn more about promises now.
You're absolutely right! My English is far from being perfect and in addition I was rushing to get this post out making bunch of mistakes. I tried to fix all the "than" "then" issues, thanks for pointing that out ;)
I was thinking about promises, because I just wrote a procedurally generated universe in Javascript with over 2^100 locations. To do the procedural generation correctly, the Math.random() function has to be called in the same order every time for generating each "sector." I had used promises in Smalltalk, but never found an application super for them and was thinking that promises would be a way to code this cleanly.
Oh noes!!! and his library starts with the dreaded exclamation point!!! -- another disgruntled JavaScripter.
I know that people identify strongly with technical choices[0] (and if you feel uneasy, you're probably the victim of this natural inclination of ours[1]), but I'm tired of the semicolon drama here on HN.
Well put. It still puzzles me that promises and/or CPS don't have a greater adoption in the JavaScript community. I guess that what happens is that the concept of a callback is a very easy to understand one, but a system becomes more complicated and cumbersome (exponentially, I would say) with every new callback that is added to it. The real problem comes when creating big async applications in JavaScript based entirely in callbacks, as it is the norm now specially in NodeJS projects. Using callbacks to build complex applications is a sure way to un-maintainability.
The code doesn't actually work though, the very first example is broken:
function sum(a, b) { return a + b }
sum = promised(sum)
var a = defer() // make promise
var b = sum(a, 1)
var b = sum(b, 5)
console.log(b) // eventually prints => 17
a.resolve(11) // fulfill promise
I have already fixed that, it was a typo, I meant following instead:
var deferred = defer() // make promise
var a = deferred.promise
var b = sum(a, 1)
var c = sum(b, 5)
console.log(c) // eventually prints => 17
deferred.resolve(11) // fulfill promise
That, or a majority of the JS coding crowd is too novice to appreciate maintainable code better than copying and pasting "simple" plumbing all over the place.
It's much more useful now, IMO. Now you can chain an error handler onto your view's render method, and the level of abstraction becomes "I couldn't render the view" rather than "I couldn't get some JSON from the server".
Functionally the behavior is the same - you're still going to have to gate an asynchronous stream of actions (make sure steps a, b, c have completed before d, etc.), but with this, you can describe that logic where it matters - renderView().then(attachHandlers).then..., instead of callback nesting hell (attachHandlers would have to be in renderView).
A good example of deferreds can be found in jQuery - every AJAX method returns a deferred that you can chain success, error, etc. handlers to. So it's not like this idea isn't a mainstream one.
The main problem I see is that very few libraries handle receiving promises as arguments, probably because there isn't a standard/widely-used implementation of them. So your logic ends up being backwards, e.g you'd chain Mustache.render off of a jQuery success promise, rather than just passing the success promise in.
What I'm suggesting is that promises don't need any special treatment. If you have a function that you want to feed a promise just wrap it: promised(Mustache.render). That's all it takes.
It strikes me that this developer chooses to write an abstraction on top of JS for solving his recurring problems, over simply using language features readily available.
Promises are as much a pattern or mechanic as callbacks, but the latter feels far more natural in JS. The `promised` decorator is interesting, but it has problems:
- The application I work on, and have in mind with this, would need just about every function decorated.
- It doesn't account for methods, unless you're default decorating the method, ie.: `Foo.prototype.myMethod = promised(function(/* ... /) { / ... */ });`
- Every decorated function takes a noticable performance hit, because things like `Function#apply` and concatenating `arguments` are relatively slow.
- It doesn't sit well with me that the example rewrites a method on a prototype declared elsewhere: `console.log = promised(console.log)`
Further, the article and library don't even scratch the surface of complicated async flows. Think async versions of common functional-style methods like `map`, `reduce`, etc.
For example, a basic scenario from our own build process is: Scan a directory for template files, read them, compile them, then concatenate the result and write it out.
We used to have a promise library to do all of this, from handling a build process to performing database queries. I discovered Async.js at some point and haven't look back since: https://github.com/caolan/async
Promises are likely to be baked into JavaScript at some point in the future[1], so I would recommend becoming familiar with them now. Even if you don't use the pattern personally you will likely run into libraries that do use it once it is part of the spec.
Just to be clear this library implements just a subset of Q with exact same API, with only addition of `promised` wrapper. I'm convinced that this wrapper is a better way
to deal with promises than dozens of utility functions that you have to learn about.
For example Q.all is promised(Array) I find later more intuitive.
That's not to say don't use Q! Q is brilliant piece of software and I'd be more than happy to see more people using it.
First of all I implement abstraction using language features and there for I take advantage of it.
- I favor maintainability over performance, also keep in
mind that promised function take promises as arguments
and can't do much until they're fulfilled (associated IO
is done) so that small performance hit is insignificant
in most of the cases.
- You can write your own decorators to wrap constructors
and their methods if you need to. That being said, I'd
recommend against, mixing mutable state with logic does
no good in long run. You'll be better of with functional.
- As for map / reduce, promises represent eventual values,
not sequences of them. For that there are streams and I
have explored that area as well:
https://github.com/Gozala/streamer/wiki/stream
I have not wrote about it because I don't think it was
good idea to dump everything in one post.
40 comments
[ 4.6 ms ] story [ 90.0 ms ] threadI've abandoned it at this point for a very particular reason: operators don't work well in this phrasing. That is, the example case given in the above discussion:
...while it's not bad, also doesn't actually try to "write logic", in a post that's about "write logic, not mechanics."What you really want to do is much more complicated because the example code starts to look ugly:
Don't get me wrong, I still like the idea and think that there might be a very promising way to do this sort of work in Node. Certainly I prefer those five lines to always writing: But as much as I really don't like that lame error stuff which breaks in a stiff wind, I also really want to be cautious about the exchange of these two lines: What we really want, I guess, is some sort of mesoscopic system -- we still want operators and familiar control structures from our "micro" world, but we also want lazy evaluations and so on from our "macro" framework. This is not impossible but it is hard to get right. If my "ducky" project moves forward it will probably look something like this: You bubble up the lazy constructs with returns, the error handling gets internalized within the lazy constructs, and the "control-method" invocation allows you to build an arbitrary control structure with whatever object you have right now.On the other hand, you need something like CPS or coroutines to cleanly use an async function in a program mostly using sync APIs.
Of course, if you have the right tools (CPS/coroutines) you can make any async API look sync, as you said.
What the quoted bit boils down to is that any function that returns normally can be converted into a function that calls a callback, but not the other way around.
I know that people identify strongly with technical choices[0] (and if you feel uneasy, you're probably the victim of this natural inclination of ours[1]), but I'm tired of the semicolon drama here on HN.
Could we please stop?
--
[0] http://paulgraham.com/identity.html
[1] or did I miss your sarcasm?
Constructive post though, thanks!
Functionally the behavior is the same - you're still going to have to gate an asynchronous stream of actions (make sure steps a, b, c have completed before d, etc.), but with this, you can describe that logic where it matters - renderView().then(attachHandlers).then..., instead of callback nesting hell (attachHandlers would have to be in renderView).
But my hope is that more libraries will be accepting promises in a future so one won't have to do the wrapping of a world around.
The main problem I see is that very few libraries handle receiving promises as arguments, probably because there isn't a standard/widely-used implementation of them. So your logic ends up being backwards, e.g you'd chain Mustache.render off of a jQuery success promise, rather than just passing the success promise in.
Promises are as much a pattern or mechanic as callbacks, but the latter feels far more natural in JS. The `promised` decorator is interesting, but it has problems:
- The application I work on, and have in mind with this, would need just about every function decorated.
- It doesn't account for methods, unless you're default decorating the method, ie.: `Foo.prototype.myMethod = promised(function(/* ... /) { / ... */ });`
- Every decorated function takes a noticable performance hit, because things like `Function#apply` and concatenating `arguments` are relatively slow.
- It doesn't sit well with me that the example rewrites a method on a prototype declared elsewhere: `console.log = promised(console.log)`
Further, the article and library don't even scratch the surface of complicated async flows. Think async versions of common functional-style methods like `map`, `reduce`, etc.
For example, a basic scenario from our own build process is: Scan a directory for template files, read them, compile them, then concatenate the result and write it out.
We used to have a promise library to do all of this, from handling a build process to performing database queries. I discovered Async.js at some point and haven't look back since: https://github.com/caolan/async
I don't too much like the promise pattern this article gives, Q is the definitive implementation: http://documentup.com/kriskowal/q/
[1]http://wiki.ecmascript.org/doku.php?id=strawman:concurrency
For example Q.all is promised(Array) I find later more intuitive.
That's not to say don't use Q! Q is brilliant piece of software and I'd be more than happy to see more people using it.