JavaScript: It’s Not Just for Browsers Any More
pragprog.com
pragprog.com
I'm not trying to imply something against this aproach, this really is a question.
I revisited Javascript a few years ago and I was pleasantly surprised. It's come a long way, and it does make event-based programming easy.
Most languages used in web development have closures. A large number of them have event-based frameworks. I found quite a few other candidates poking around in the other languages but lacked the skills or time to evaluate whether they were also structured like Node.js.
I don't mind the existence of Node.js, but the people claiming it's "better than most" make me wonder if they've ever used the "most" they claim it's better than. Erlang in particular. That's something that's "better at the event-driven paradigm than most". Javascript has nothing on Erlang, Javascript's just another C-derived manually-chop-your-code-up-to-work-with-callbacks monstrosity by comparison.
Edit: No, if I were going to pitch Node.js, I would pitch it as "Use the same language on both the client and the server; it's adequate on the client, it's adequate on the server." OK, "adequate" isn't the best pitch but it's honest. The stupidest thing about web development today is needing to know three languages (HTML + browser quirks counts as roughly as complex as a language, client-side JS, server side not-Javascript language) just to get your foot in the door. Having the same language on client and server will probably provide some interesting capabilities, such as the thing mentioned in another comment where your comment formatting code can be run in either place (show on the client exactly what they'll get if they submit, run on the server to validate it), or validation code that is guaranteed identical on both client and server, or several other interesting things I can imagine where you can play games with exactly where something is run. Server-side Javascript just shouldn't be pitched as a "uniquely capable" language in a field full of PHP, Perl, Python, and Ruby... "uniquely capable" is Erlang or a the Seaside framework, not Yet Another (Dynamic) Algol Variant.
Not really relevant in this context, as naming things isn't that big a deal. Lambdas are only needed for one-liners and they serve that purpose fine.
"As far as I know, Python, Ruby, and PHP all suffer from the fact that all functions are not first class."
Definitely false for Python, Ruby actually has multiple kinds of first-class functions (for better or for worse, I am assured that in practice it's not a problem), recently false for PHP (http://en.wikipedia.org/wiki/PHP#5.3_and_newer ), and to save time, also false for Perl, Lua, ${all functional languages} (pretty much by definition, there), and false for C# as of 3.0.
If you want to play "oh but the support is quirky", Javascript has its own quirks on the closure front:
var i;
var funcs = [];
for (i = 0; i < 5; i++) {
funcs.push(function () { alert(i) });
}
funcs[3]();
Mind you, that's only a quirk and I wouldn't crucify the language for it, it's arguably correct and sensible once you understand it. The again, Python's scoping is arguably correct and sensible once you understand it, too, but people don't cut it much slack for that.
(The answer is 5 by the way. Compare with Perl, which can actually close on a variable created inside a scope in the function, instead of having just one function-level scope like Javascript. Compare with Python, which does what Javascript does.)
We of course need to define what first-class is ... if it's the ability to have higher-order functions, than 2.0 fits. C# 3.0 only adds a light-weight syntax for anonymous delegates plus the really kick-ass ability to get a syntax tree of that method instead of the method's reference (on which Linq is based) ... and I really wish other languages add this capability (it was possible in Ruby 1.8.x, but it's not anymore in Ruby 2.0).
Python kicks ass in regard to first-class functions ... and it leverages that ability with decorators. People using other languages go through great pain to have AOP capabilities (for example), but in Python it only takes like 10 minutes to add a couple of utilities to do whatever you want.
The only frustration I have with it is the lack of multi-line anonymous blocks, but Python is so flexible you can add that using "with" block hacks, like described here ... http://billmill.org/multi_line_lambdas.html
Also, once you understand dynamic-scoping ... it's really not that big of a deal to deal with the situation described above. It's just different.
I like the lexical-scoping in Perl as it prevents all kinds of errors ... but in Python I have real exception-handling, and a kick-ass debugger console just by adding "import ipdb; ipdb.set_trace()" anywhere I want.
I'm not sure what makes this a "quirk" since this is exactly what closures should do: they maintain a reference to the variable from the enclosing scope. This is also the behavior of a fairly classical example of closures in scheme.
The first way, which I prefer, is to make the binding of i to its value inside the loop body, this way you get a nice fresh binding on each loop iteration, so if you iterate the loop 5 time you get 5 bindings.
The other way, which I loathe but is still an arguably "correct" way to do it, is to make one binding of i outside of the loop and then modify it on each loop iteration, so no matter how many iterations of the loop you have, there's only one binding.
As far as I'm concerned, the first implementation is obviously the better one because the instances where this benefits you come often while the instances where the second implementation benefits you is very rare and unidiomatic. I think the second implementation is used for better performance but that's a mistake.
I had this problem of perceptions when I switched from Common Lisp's Loop to Iterate. Loop, for all its many big shortcomings, had the decency of making a fresh binding on each new iteration, though IIRC the standard specifies that bindings can be established outside or inside the loop context. When I want to make a closure based on the value of an iterated variable with Iterate I have to do an explicit binding...
Javascript's already closely tied to the web, though, and it gets plenty of word-of-mouth since web developers already know it. That will probably beat any technical advantage.
Erlang is different. Node presents a different programming model than Erlang - Node handles thousands of connections per process - and keeps itself close to the metal. Erlang is it's own operating system handling one connection per processes but allowing very many processes. Is Node's model better? Well at the moment it's incomparable since Erlang is very much a solid, production ready system and Node is not more than some dude's hack. But supposing that Node becomes stable and usable at some point, a major advantage of the Node model is that for the first 10,000-1,000,000 concurrent connections the programmer doesn't have to know anything about concurrency - a single process will just handle it. There is no forking or IPC - it's just synchronous events and callbacks. In Erlang you have to think about about IPC on the first connection. The Erlang-model might turn out to be the wrong level of process granularity. Maybe the right level of process granularity is how computers are already designed: 10,000-1,000,000 connections per process, use one per core.
I also wish for less hype around the Node project but I think you're short changing it by saying it's same thing as Twisted but in Javascript instead of Python. The main selling point is not that you can run the same code on browser and server. The main selling point is that you don't have to know what you're doing, at least for the first 10,000 concurrent connections.
I'm tremendously amused that you write this when every other poster on HN is desperately trying to generate hype for their project/company.
Also false. An Erlang "process" is not an OS process. It already works the way you say, except it can use multiple cores simultaneously. Automatically. Node.js has nothing on Erlang except a familiar syntax. Nor does IPC come up like some kind of blocker; it's baked into the core and quite natural, often all wrapped up behind a simple function call.
And let's not even talk about the baked-in clusterable Mnesia database or the OTP library, which Node.js can only dream about. (Erlang isn't just multicore, it's pretty easy to make it multi-system.) Or how Erlang has code replacement (the Node.js complaint of the day) built right in to the core, where it really has to be for it to work. And Erlang laughs at your "non-blocking purity"; in Erlang, you can call sleep and it won't block anything! No special support needed, no magic from the user, no arranging things in callbacks manually (how 1980s). Try that in Node.js.
It's actually impossible for Node.js to do some of these things because you have to start with them.
You don't understand how Erlang works. That's fine, except you started to criticize it. I refer back to my previous comment that the people hyping Node.js don't seem to have used the competition. Maybe my problem is that I have used the competition, quite extensively. Come to think of it, I've yet to hear anyone say "Gosh, I was using Erlang but Node.js is just so awesome I had to switch!:" That probably says something important.
* A shitton of people are already using it on the daily
* All of the existing code written in it is event-driven
* It's got prototypical inheritance, and some people
actually use it instead of bludgeoning in classicism2. There are a lot of people who use erlang.
This is actually a problem. Majority of that shitton use js without knowing what they are doing. It's a copy-pasted code for some client-side validation. Even the so-called framework-users (jQuery and Prototype) depend blindly on plugins for their job.
But the point is that you're not going to get fired for choosing JS on the server. It has curly braces! Your boss isn't so worried about hiring.
Ruby is just a mess in this regard,even if you can arguably do anything you want in the end : http://innig.net/software/ruby/closures-in-ruby.rb
On the other hand, javascript has a simple, no bullshit syntax for functions. If you want a function to be anonymous, just don't give it a name.
Erlang and Lua obviously have true anonymous functions and closure support.
Javascript has java-like syntax. Heck, it even bears half the name even if the language has almost nothing to do with Java. Did you know that the first version of Javascript was actually a scheme variant but they decided to conceal its identity so it would appeal to the masses?
It's stupid that a familiar syntax is one of the foremost considerations in the mass adoption of a language, but that's reality for you.
(PS: I'm a Common Lisp fanatic so I'm acutely aware of this syntax problem.)
It's actually remarkably similar to Scheme [1] with very flexible arrays and real first-class functions.
We have something else, JavaScript. It does not have sexprs. It doesn't have continuations. It doesn't have tail call optimization. It has mutable array operations, but immutable string operations (got it backwards.) There are no macros, hygienic or otherwise.
The only thing it has in common with Scheme are first-class functions. If you want anything else, you must implement it yourself out of the Turing Tarpit, and it will perform badly, because optimizing JavaScript is difficult due to its mistaken specification.
By any of these metrics, Python, Ruby, Perl, and C# all have more in common with Scheme than JavaScript. I think the people claiming 'JavaScript is like Scheme but with a different syntax' are mistaken from lack of experience.
1) This approach lets me put together a quick server without a big, complex container to configure (No Apache, Tomcat, Rails, (Insert Python Framework Here), etc).
2) Javascript really is a nice language to program in. First class functions lead to some neat tricks, and aside from a few minor quirks (scoping issues, overloading the "+" operator) it lacks nasty gotchas like Java and PHP, at least so far.
3) Lots of programmers have JS experience and web experience, so there's a large potential pool of users (Subjectively, this may or may not be a good thing, see PHP. But there's demand)
It's probably not the be-all-end-all, but there are some positive reasons to see work in this space. I'm curious to see how the language implementations evolve as people start doing more complex server side things with JS.
My biggest issue is that as far as I know there are no decent tools for it yet. Notepad, hey, sod that crap.
TBH I was hoping it was going to die a death with flash/silverlight, but alas with apple hating flash and the g man sponsoring it, it seems destined to stay.
It honestly feels like a step back.
It reduces the learning curve for new developers because they only have to learn javascript, not javascript plus something else.
It would be nice to write both client and server in one language.
I think this was briancarper.net where - since he's using clojure on the server - ended up running a javascript interpreter (rhino) on the server so that the same formatting code could run on the client and server.
http://github.com/raganwald/homoiconic/blob/master/2010/02/d...
In the rails world, for example, if you've done your validates_* declarations on an ActiveRecord model, it shouldn't be too difficult to generate the javascript required for client-side validation.
Actually found a project that goes a long way towards achieving just that: http://github.com/grimen/validatious-on-rails
But I guess you're referring to complex validations, expressed as code instead of validates_* declarations. It seems like you would prefer either an all-JS approach or an automated translation to JS, as per the blog post you linked. To me, it boils down to whether such a complicated validation is really appropriate for client-side validation and whether it blurs the lines between form validation and business logic to such an extent that it's better off just living on the server.
But I'd love to hear your further thoughts on the matter.
Cultural norms and design patterns in Javascript Land increasingly orbit around projects like jQuery. This sort of influence works out spectacularly-well if you've spent the last several years working countless hours on browser-side code and suddenly stumble on nodejs and its evented/closure style of programming. Perhaps this doesn't describe you, but from my vantage point, this describes a vast number of programmers. There are a lot of people in that demographic who want to be able to rapidly write out web apps in the same frame of mind client and server side and not have to switch to thinking in Python or Ruby.
The fact that V8's engine is competitively fast isn't hurting these efforts either.
http://shootout.alioth.debian.org/u32/benchmark.php?test=all...
By definition they are flawed benchmarks, but you will have a hard time writing faster JavaScript code than equivalent code in most other platforms. In the case of Lua (and Python, and Perl, and many many others) the equivalent JavaScript code will also be as long or longer, and not as clean.
I'm not going to get mad at anyone for their choice to use JavaScript, but the uncritical repetition of sentences like "Thanks to V8 and other modern engines, JavaScript has become fast compared to optimized language/platform X" or "JavaScript offers these N features you won't find anywhere else" will not fly.
http://shootout.alioth.debian.org/u32/which-programming-lang...
http://steve-yegge.blogspot.com/2007/02/next-big-language.ht...
... though he doesn't explicitly say he's talking about Javascript, it's pretty likely that he's talking about Javascript.
1) It's got C-like syntax.
2) It's got the dynamic- and functional-language features that make people happy.
3) We're stuck with it no matter what. Changing the world's installed base of web browsers takes about a decade; such is the lesson of IE6. The only universally-supported client-side language on web browsers is Javascript. Its successor, should it exist, hasn't even been sighted on the horizon. [1] So we're in for at least another decade in which every web developer needs to know Javascript.
4) Because of #3, lots of folks are working to make Javascript fast and reliable. There are several JS interpreter projects in active competition. That kind of focus has already paid off in spades, and is going to pay off even more over time. Those of us who remember the days when Java was considered painfully slow, to the point where people complained about how crippled it was, understand what happens when the bulk of the world's compiler wizards spend a decade optimizing a language's runtime: It tends to become better. Much better.
---
[1] The only serious challenger, Flash, is not only proprietary, widely loathed, suffering from a PR slump, and under direct attack from Apple but is also... based on ECMAscript, a.k.a. Javascript. Or so I understand. So it's more of an alternative runtime than a real alternative language.
package com.example
{
import flash.text.TextField;
import flash.display.Sprite;
public class Greeter extends Sprite
{
public function Greeter()
{
var txtHello:TextField = new TextField();
txtHello.text = "Hello World";
addChild(txtHello);
}
}
}
IE6 continues to be a pain for sure, but the mindset behind this is changing as the web and it's users evolve.
You may be right. And then again maybe not. It's going to be fun to find out!
I would note that IE6 persisted until now not because anyone wanted it to -- Microsoft hates IE6 more than anyone, since every IE6 user is both a support timesink and a potential paid upgrade that hasn't happened yet -- but because it sits in a niche. There are people who bought old machines that haven't worn out yet. And there are businesses that built a bunch of stuff on IE6 and see no reason to upgrade.
And the nature of the universe is: Things diverge over time. (Yet another side effect of the second law of thermodynamics.) The number of niches and their diversity is growing now, and they will continue to grow. In ten years there will still be people, somewhere, running IE6. There will also be people running all sorts of other things: Old iPod Touches running obsolete versions of Mobile Safari, eleven-year-old car navigation systems, smart phones and book readers of every description, obsolete gaming hardware (have you noticed that you can still buy a brand-new Playstation 2, ten years after it launched? Have you considered the implication: There will certainly be many PS3 units in use ten years from today?)
The good news is that, thanks to ubiquitous Internet, companies are increasingly able to force customers to use the latest-and-greatest browser technology by pushing upgrades out over the net from on high. But is (e.g.) Sony really going to officially support the PS3 forever? What about the manufacturers who go bankrupt? And, if the market ends up demanding that manufacturers support multiple decades worth of their legacy hardware, is that really going to tend to encourage innovation in web standards, or is it just going to make such innovation that much more costly to implement, encouraging the tendency to embrace the status quo because it is "good enough"? We'll see.
1. Future browser upgrade timelines should not be based on the IE6 fiasco. It's too small a sample size to use as corroborating evidence in these predictions. It may end up being an anomaly.
2. In my own opinion I think the upgrade timeline will grow shorter. I don't have any hard evidence I can cite, it's just my impression as someone who's been involved in web development for the last 12 years or so. The points you mention are all valid, and I'm not saying there won't be new and different pains we'll enjoy ;)
I don't disagree that syntax matters, but C's is hardly a shining example. There's a network effect with programming languages, though (libraries, documentation, community, etc.), and there's a certain amount of realpolitik in designing with "C-Like Syntax" in mind.
I'm more worried about languages with ugly semantics. Odd syntax is only skin deep, but bad design goes straight to the bone.
Yeah, he's a good writer and I've read all of its blog-posts ... but he seems like the kind of programmer that's always discontented with the status-quo, always looking for that next big language, always relentlessly looking for tools that will make programming fun again ... and that particular article seems to me a lot like wishful thinking.
And guess what ... you won't find much happiness in switching languages, unless the culture around it favors simplicity and getting things done (and your previous community didn't). That's why Lisp is almost dead.
Also, Javascript may become popular on the server-side, but that's only because browsers are stuck in the past, while becoming more and more relevant to the future.
But give developers the choice of multiple languages ... and I'm pretty sure Javascript will fade away to a niche. Incidentally, you can play with Ruby and Python in the browser right now ... it requires Silverlight/Moonlight, but it's pretty cool and gives some glimpse in what the future should look like.
It's simpler than both regarding the core language features, and is also the only one of the three to have real anonymous functions.
I'm not fond of C syntax, but i can live with it
What are you talking about?
http://news.ycombinator.com/item?id=1171471
Arguably i took a bold step regarding Ruby, which has them, but is confusing in this regard since blocks are not anonymous functions, and are the idiomatic and preferred construct in ruby. The problem is deeper than that , but Paul Canter does a much better job than me explaining why
What differentiates Ruby from Javascript (or Python for that matter) is that when defining a function using regular syntax, its name is not a variable referring that function in the scope it was declared ... what Python and Javascript are doing. And that's a trade-off that some people like it, because in Ruby you can call a function without parens and it doesn't really get in your way when wanting to do crazy shit with already declared functions.
On Python, it may not have anonymous functions, but it has true first-class functions and true closures. It's only a problem for async code because you want to have the declaration of a closure after you see the call-site that's using that closure, not before it. But in practice that's not so awful ... and if you really really want multi-line anonymous blocks there are hacks you can do with the "with" block.
In this regard Javascript is cleaner, but it is messier in other places. I don't like that its a loosly-typed language for instance (haven't we had enough of that?).
I don't like that it doesn't provide class-based OOP ... and while you can build it on top of prototypes, it's a PITA not having a standard. And class-based OOP is mapped better to the way we think, having a bad reputation only because a couple of popular languages have plagued us with bad implementations ... but people should really take a look at Smalltalk before throwing the baby out with the bathwater.
That's something i don't like about ruby. You can do crazy shit but it leads to code wich is opaque for somebody that doesn't truly know ruby inside out. The language is full of tricks. But i don't want a full of tricks language. I want a simple language that doesn't get in my way, and doesn't provide 7 ways and 3 different mechanisms to do closure like constructs (numbers are not made up)
> On Python [...] if you really really want multiline anonymous blocks, there are hacks you can do with [...]
Well yeah i really really want them, and no i don't want to resort on hacks to use them.
I really have trouble understanding the rationale behind all those half baked constructs. The way Lua/Javascript/Scheme does it is the way to go IMHO : provide one mechanism that is simple and works well. You may loose some convenience or conciseness in the process (compared to ruby blocks for example), but what you gain is one consistent way to do things.
About javascript OO, i admit that it doesn't bother me as much as it seems to bother you, and that it may be linked to the uses i have had for javascript yet, but i have to say i indeed had more pleasure working with python/ruby object models than with javascript at the moment. I can't say if that is due to a superior mechanism or to familiarity with class based models though.
edit: It also makes me wonder how many people would give lisp, basic, and etc a second chance if those languages had the equivalent of "Wow" that jQuery provided to JS for me and others.
do_thing_A_and_after_that_call(function() { do_thing_B_and_after_that_call(something); }
do_thing_C_and_after_that_call(function() { do_thing_D_and_after_that_call(something); }
what should I do to execute some E after both B and D is finished?
I think that non-blocking effectively means introducing easy syntax for parallel execution while complicating syntax for sequential execution.
I think parallel and sequential execution should be easy: on T1 and T2 do E where T1 = (A; B), T2 = (C; D)
If you can guarantee that a given variable can only be modified by one process at a time, it is easy to do something to take care of this for individual processes, and wouldn't be hard to generalize.
(note, I was last using prototype.js, and my javascript is a bit rusty)
wait_list = [];
function wait_for(entry) {
wait_list.push(entry);
};
function done(entry) {
wait_list.splice(wait_list.indexOf(entry), 1);
if (wait_list.length == 0) {
do_C();
}
};
function do_A () {
wait_for("A");
// do stuff
done("A");
};
function do_B () {
wait_for("B");
// do stuff
done("B");
};
function do_C () {
// Stuff that comes after A and B
};All parallelism is cooperative in javascript, meaning that you don't have to worry about control flow switching out from under you at arbitrary points, just when you explicitly yield it by returning from an event handler or from the top level of your program.
Finally, to answer the grandparent question, the easiest answer is a promise group. A promise group is basically a promise that emits a success event once all of its child promises have emitted success events. There are a few implementations floating around, though it's just a handful of lines of code to write and makes for a good exercise.
So assuming that A and B are rewritten to return promises (or wrapped, I believe that isaacs has a library which will wrap functions which take a callback to return a promise instead), you can do the following:
var promise_a = do_A();
promise_a.onSuccess(do_C);
var promise_b = do_B();
promise_b.onSuccess(do_D);
var promise_group = new PromiseGroup([promise_a, promise_b]);
promise_group.onSuccess(do_E);I don't event have a kid! ;-)
Thank you, that was the thing I was looking for. My question now: What happens if one of the promises fails? Is a promise group disposed then? Solution offered by Periodic would wait forever.
And two more that aren't Node-specific:
For example a simple comet style notification system can be created less than 100 lines of code.
With mod_php even if you kill or freeze single apache process your app is still up and running for other users. What happens if Node.js crashes or hoards resources?
> During development, one small, yet important, feature that Node lacks is the
> ability to auto-restart when a change is made to the server’s source code.