Faking Co-Routines, or Why Callback Hell Is Over (2014)
pandastrike.comI have yet to see an "improvement" on callbacks, whether it's promises or fibers or generators, where the benefit in readability is worth the havoc it wreaks on my ability to debug the program.
These days I write JavaScript using only functions, literals, variables, and the occasional prototype and it's amazing.
I think many programmers have a desire to believe they are working on complex problems that demand sophisticated tools. I remember learning Ruby and being excited whenever I found a reason to write a DSL. In retrospect it was unnecessary every time. In every case the code would've been clearer if I had just stuck with functions and kept refactoring until I had the right interfaces and data structures.
It helps to remember
function a() {
b(function() {
//etc
})
}
function a() {
b(c)
}
function c() {
//etc
}
> It helps to remember
> function a() { b(function() { //etc }) } is equivalent to
> function a() { b(c) } function c() { //etc } which is not particularly more verbose. And as a side benefit, refactoring that way gives you an opportunity to make c() self-documenting.
It's not always equivalent, since you can have closures.
That's true, and I use closures sometimes. But they are a performance and readability anti-pattern, and it's often better to either pass in or bind the data you actually need.
In some sense closures are globals and globals are bad.
Unfortunately, Node.js decided that every callback should accept an error as the first argument to every callback. If you're chaining a bunch of callbacks, it's tedious and error-prone to add boilerplate to check for an error and handle it consistently in every callback, and violating the DRY principle. It's not easy to simply propagate the error to a higher-level handler.
I find the `async` library excels at this:
There are all sorts of helpful async primitives in there. I don't write Javascript without it!var fs = require('fs'); var async = require('async'); async.waterfall([ (cb) => fs.readFile('foo.txt', cb), (data, cb) => my_function(data, cb), (result, cb) => myDb.lookup(result, cb), ], (err, finalResult) { if (err) { console.error(err); } else { console.log(finalResult); } });Here is a link to the documentation: https://github.com/caolan/async
I would have to look at your code, but generally if you have a long callback chain and you're trying to propagate errors up it you are doing something wrong. Are you really working on 10 nested activities simultaneously, or are you just nesting things because it's convenient and you don't want to think about how to do things in stages?
If you're trying to convince me that "callbacks become painful in situation X" you really don't need to. I know that. What I'm saying is that 9 times out of 10, the answer to the question "do we really want to be in situation X in the first place?" is "no". But instead of getting out of a bad situation by refactoring, people just write crazier and crazier control structures (i.e. promises) to make those bad situations workable.
Its a problem yes, but the nature of javascript makes this a difficult problem to address. Node tried addressing this with the 'domain' module to help with propagating errors upward more generically:
https://nodejs.org/api/domain.html
But it was deprecated: https://github.com/nodejs/node/issues/66
promises are a very nice way of drying out error handling and cleaning up callbacks for now.
If you're repeating yourself, then you could possibly abuse higher class functions to automatically generate error handling?
Yes, there are libs like errTo [1] and iced-error/make_esc [2]. Not to mention Promise.promisify(). Long solved problem.
> I have yet to see an "improvement" on callbacks, whether it's promises or fibers or generators, where the benefit in readability is worth the havoc it wreaks on my ability to debug the program.
Chrome has support for debugging async code - for example you can step from one code block to the code in callback as if it would be executed sequentially. So the 'debuggability' is a problem which could be solved on the side of the tools. Of course the case where there would be one single standard for handling async would vastly simplify situation for the tools developers.
Please: Do not use CoffeeScript in your examples.
It is not common to have "readability" in CoffeeScript and it restricts a significant number of readers from being able to understand the code sample.
> restricts a significant number of readers from being able to understand
Can you seriously say you can't read it? It seemed easy to me.
I don't understand why he talks about a javascript problem being solved, then goes on to show us examples in coffee script.
There really shouldn't be an difficulty seeing how it maps to JavaScript. CoffeeScript is pretty clean for code examples.
It's not that it's difficult to understand, it's about comparing apples to apples.
Paste the code into here if you’re having trouble following: http://coffeescript.org/#try:%7BliftAll%7D%20%3D%20require%2...
I still don't get why async/await is inherently better than using Promises/Streams. Is it purely a syntax difference or is there a semantic difference as well?
I believe async/await uses promises behind the scenes. It just means invoking code looks nicer, i.e writing "var res1 = await prom1(a); await prom2(a, res1)" is marginally cleaner than "return prom1(a).then((res1) => prom2(a, res1));", in the sense that you're just writing statements rather than chaining method invocations.
https://github.com/scriby/asyncblock-generators
That's a control flow solution I wrote on top of generators to make it a little easier to manage parallel tasks, timeouts, error handling, and so on.
I originally made asyncblock, which was based on fibers a few years ago. This module uses the same underpinnings as asyncblock, just based in generators instead of fibers.
The problem with javascript coroutines is that you can only yield from within the generator itself, not from a called function.
This makes it impossible, for instance, to write a nice I/O library that is to be called from within a generator (the library is supposed to yield on a blocking situation).
What’s with the HUGE FONT SIZE?
Use ClojureScript + core.async and get some work done.
Absolutely, cljs is a top-shelf way to work in JS. And core.async is fantastic.