Hacker News new | past | comments | ask | show | jobs | submit login
CoffeeScript: less typing, bad readability (ceronman.com)
349 points by dcu on Sept 17, 2012 | hide | past | web | favorite | 218 comments

Most of the bad code examples in this post are taken from the CoffeeScript compiler itself. For that I have to say "mea culpa". There's a lot of overly clever and dense bits that have accrued in the compiler over time -- sometimes due to the tangly nature of the codebase, but also sometimes because we tend to use the self-hosted compiler as a test bed for edge cases of combining different language features together.

If you're looking for more regular style, this style guide is a better place to start:


Perhaps a style guide should be put up at coffeescript.org, or at least linked to.

I was literally coming here to post this response. It seems like a revolution in (usage of) coffeescript is needed in much the same way Crockford's Good Parts paved the way for writing good javascript.

Note: I'm not saying Crockford was singularly responsible for this, but The Good Parts was certainly well-received and very much popularised for it's opinionated views on the language.

Edit: I mean in terms of coding style, CS itself has already cut out a lot of The Bad Parts of JS

Perhaps a style guide should just be encoded into the compiler.

I feel like CoffeeScript is failing at what it's attempting to do if it needs its own style guide.

This is a fantastic post -- I feel like I've run into all these crazy "gotchas" over the past year, and more.

And the worst part of it is, nearly all the confusing/ambiguous/different compilation examples given rely on the undocumented rules CoffeeScript uses for parsing. It's amazing that, still, CoffeeScript's entire documentation is basically a "getting started" guide, and there is simply no reference documentation.

Magic overdose.

Compare with Go, where you can easily read the whole spec and keep the whole language in your head: http://golang.org/ref/spec

Go is an excellent contrast. Not only is the syntax spartan, but--especially if you include the built-in formatter gofmt--it has a fascist insistence on a uniform coding style. I don't agree with all its stylistic decisions, but I'd rather live with them and have consistency than run into corner cases and gotchas.

I would not consider a 62 pages spec possible to "keep in your head", let alone easily.

The "The Smalltalk Language" section of the last ANSI Smalltalk draft is a more reasonable 28 pages.

I think CoffeeScript seems nice in theory, but in the end, it's something made for people who want to code javascript, without actually using the language. It exchanges some of the terse, functional elegance javascript offers for a different kind of elegance, with its classes and inheritance and such like.

I won't say it's bad, as it is nice to work with when you're writing it, and I do quite like it. But I could never recommend it to someone who didn't have a good knowledge of javascript already.

I don't believe it exchanges any of the elegance. You can do the same things in CoffeeScript that you would in JS, just with less typing. That being said, a knowledge of how the browser will execute your compiled Javascript is crucial.

I personally describe CoffeeScript as "stabby". CS has all these little shortcuts and shorthands which seem great and appear to save time, but many of them are ambiguous, and then you get stabbed by their ambiguity, and you waste time debugging. Python, on the other hand, is still quite concise, but is generally very unambiguous.

I always thought CS was a "kinda-python-y" version of Javascript, though I never actually used it. It surprised me, therefore, to see that the OP considers these two items to be principles applied to the design of the language:

  - Implicit is better than explicit

  - There is more than one way to do it
These are the exact _opposite_ of a couple of the principles of the design of Python[1].

1: http://www.python.org/dev/peps/pep-0020/

Indeed but they obviously seem to apply to CS as he has shown in his examples.

Oh, I agree, it was just surprising to me to see them there.

As I wrote in the other recent CS thread here, I think your coding style must be ill-suited to coffeescript. I've had a few situations where I've had to test things out for myself, but nothing like what you put forth. I just write CS as I would JS and that's that.

To that end, the docs on Coffeescript.org are the reference documentation (as they are the only official documents to reference) and they've suited me just fine. I get the sense that you're looking for a document in the style of ECMA-262, rather than one in the style of MDN.

Right. I've started using CoffeeScript for a side project, and I really appreciate how it fixes most the warts in JS, but I wish they had stopped there rather than making gratuitous syntax changes. Parentheses in function calls is not an actual problem.

I'm afraid that in conjunction with significant whitespace, they're not at all gratuitous. For example:

    describe 'Calculator', ->

      it 'can add two positive numbers', ->
        result = calculator.add 2, 3
        assert result is 5

      it 'can handle negative number addition', ->
        result = calculator.add -10, 5
        assert result is -5 
Mandatory parentheses make significant whitespace lose much of its appeal when passing a block construct into a function. This is a large part of the reason why Python limits lambdas to a single line...

> This is a large part of the reason why Python limits lambdas to a single line...

Python doesn't limit lambdas to a single line, but to a single expression. That is a rather different constraint.

Sorry for the imprecision. It's the same thing to Python, in terms of significant whitespace. Language features that use whitespace for structure are not considered to be expressions, and can't be used as part of a larger computation. In CoffeeScript, they are, and can.

Valid point. Although I'm on the fence about significant whitespace in CS as well; I value fixing JavaScript's glaring flaws much more highly than arguably more convenient but unproven syntax. (The boundary is admittedly fuzzy, e.g. list comprehensions).

Optional parentheses wouldn't be so irritating if the rules for omitting them were universal. Currently, when calling functions with arguments you can omit the syntax, but when calling functions without arguments you must use parentheses. Could the 'dispatch vs variable reference' question be decided with some sort of suffix that preserves the value of whitespace? Something like 'f!' 'g! arg1, arg2'?

I think it would be interesting to see the operational semantics for a very small subset of the language. Applying that type of formalism would likely suss out or highlight the corner cases like the ones highlighted in the article.

Maybe coffeescript programmers could just use their brain and choose not to descend the Infinite Staircase of Obfuscation (the carrot's not worth it). Just because Coffeescript can saw your foot off doesn't mean you have to.

I have used CS on 3 projects and Scala in 2 projects. For me, both CS and Scala have terrible readability. Scala is more worse, I'll need at least 3x to read it than write it. Obviously, YMMV (tm)

I don't have a strong opinion on the article itself (it seems well-reasoned to me, but I don't use CoffeeScript personally so who knows)... but the comments attached to it are hilarious.

I think CoffeeScript has great readability, because I’ve taken the time to learn how it works.

If you have to take the time to learn how to read something, it's not easily readable.

CoffeeScript is only hard to read if you try to read it with a JavaScript mindset.

Yes, why would anyone approach a tool pitched explicitly at JavaScript developers, whose "golden rule" is "it's just JavaScript" (see the top of http://coffeescript.org/), with a "JavaScript mindset"? Truly, it is a mystery.

So is mathematical notation not readable? On the one hand, it takes quite a while to learn how to read it. On the other hand, it can condense literally pages of text into a few equations which somebody experienced can understand almost at a glance.

The same is true for programming languages. Sure, anybody can read Python. But it's fairly verbose and attempts to mirror English. On the other hand, Haskell is somewhat terse and attempts to mirror mathematical notation. I've found that, after about the same amount of time with either language (about a year of off-and-on use for both), Haskell becomes far easier to read than Python.

So just because something is easier immediately does not mean it is any more readable. It could, in fact, be less readable and merely more familiar!

Mathematical notation is not readable, no. It requires extensive training to be able to understand it, and even experts will take quite a while to comprehend a document they are not already mostly familiar with. It is not like you cross some threshold and immediately become able to understand all mathematical notation with equal ease. There's plenty of evidence for this; recently we had this quote about the ABC theorem: "It can require a huge investment of time to understand a long and sophisticated proof, so the willingness by others to do this rests not only on the importance of the announcement but also on the track record of the authors" (http://www.nature.com/news/proof-claimed-for-deep-connection...)

Other examples that highlight how understanding of mathematical research involves much more than just reading: https://docs.google.com/viewer?a=v&q=cache:5NM3XjxAWfUJ:... http://math.stackexchange.com/questions/13460/how-to-read-a-...

> even experts will take quite a while to comprehend a document they are not already mostly familiar with

Experts don't take time to read notations. They take time to understand the full scope of consequences embedded in a line composed of such a condensed and readable notation, within the frame of an otherwise complex proof.

> Mathematical notation is not readable, no.

I beg to differ. You mention the recent possible proof of the ABC conjecture, and I happened to browse through various materials on the subject. While I am far from being able to understand the crux of the proof, I can perfectly read the various components of it. In fact the notation system is so readable it actually helps tremendously in grasping concepts previously unknown to me.

Mathematical notation is much harder to read than prose. It does a very good job of serving as a mnemonic for something you already understand, and a very poor job of explaining novel (to the reader) concepts.

It's also as bad as natural language for ambiguity and informality.

I don't think either English text or mathematical notation is a magic wand for readable programming language; It's more about reducing accidental ambiguity and the amount of mental context needed for parsing 'natural' constructs in the language.

Personally, most of the benefit I see from mathematical notation is down to pattern matching -- "Oh, this looks like a Fourier transform". The idea seems to be supported by the fact that there is a unwritten convention for which symbols to use for which concept (alpha over beta for example) and the way in which the equations are laid out.


> If you have to take the time to learn how to read something, it's not easily readable.

I don't think that's completely true. If you're used to algol or C-style, Lisps tend to be hard to read. Concatenative languages even more so. Not necessarily because they're hard to read in and of themselves, but because the "shape" of the code is "all wrong" compared what you're used to, and your usual anchors (keywords, binary operators, infix assingment, ...) are nowhere to be found.

I mean Arabic or Korean are not necessarily harder to read than English, but you've got to know them and how they work before you even recognize patterns in the squiggly lines and little squares.

Not really; Lisp is hard to read because the syntax is unusual, but it's not hard to understand because it's complicated. On the contrary, it has the most consistent syntax of any language.

They author's point is that Coffeescript is hard to read because it has too many edge cases which are not easily understood under a smalla and consistent set of rules. I have not programmed with CS, but it sounds like a convincing argument.

Or, to restate this and amplify it a bit, Lisp syntax is hard to read because most people are unfamiliar with it. This isn't -- or should not be -- a problem if you need to learn to read a language in order to use it. (This is orthogonal to whether you want to learn it in the first place, or whether you "ought" to.) It is a problem if learning how to read and write the language isn't enough.

The point TFA raises is that people who're familiar with CoffeeScript syntax may still make mistakes resulting in completely erroneous but valid code. Yikes. And the inherent nature of whitespace or indentation makes the problem harder to detect by cursory inspection.

I have no strong opinions about CoffeeScript, myself. I just thought this was an interesting distinction, and a rare case where syntax makes a significant difference.

Mere consistency is not enough. List is hard to read because humans do not have a natural feel for how deep a set of nested brackets are, and you have to be able to spot the matching open-bracket to understand the code.

Again, it's hard to read if you are not familiar with it. Even as someone relatively new to Lisp, indentation and formatting makes a world of difference for readability.

This isn't specific to Lisp, either. I posit that if you took your average C-like and removed indentation from any reasonably sized function or method (one or more loops, one or more if/else), you would have a hard time counting brackets.

I agree that consistency is necessary but not sufficient, though, which is what I think you're getting at. Most of the time, though, consistency is clearly better than inconsistency. There's less to remember or, if you prefer, less to screw up. :)

Please note that I was not replying to TFA, I was replying to smacktoward's assertion that if you need to learn how to read a language it's not easily readable.

i'll stick to javascript. coffeescript doesn't add a whole lot to the table, javascript is the standard and that's a good thing, re-writing the language is not the answer in my opinion. i'd have to imagine that saving a few keystroke might sound sexy to lesser experienced developers, but that really is all it adds to the table. being able to code on any project immediately off the bat with a standardized language is a huge plus, in terms of flexibility and scalability. you don't want to limit your company to exotic language, only to later find out few months later another new language called "frapscript" is now the new cool thing to do.

i recall the days when i was thrown into a project where the requirement was to use mootools, rather than jquery. even though they are both "javascript" in nature, it severely limited my productivity. i can see the same thing happening to companies trying out coffeescript, it may seem like you are doing the better thing long-term wise, but in reality, it is not so much. it's better to just stick to standards and you'll be more productive down the line, i can almost guarantee it.

I'm not the hugest fan of coffeescript, but I don't know if you can consider it to be an 'exotic' language at this point. It's quite widely accepted.

on which metric do you base it on, to be widely accepted?

Looking at github's top languages, it sits at #11. I thought it was higher than that, but it's still nothing to sneeze at.

Maybe saying it's widely accepted isn't entirely appropriate, but it certainly is gaining a lot of traction.

I'm not sure that this will ever be a fair statistic because coffeescript should never really be used for shared / module code IMO. I love CS for my application code but would never release a package on github in CS, only JS.

Once you are releasing things outside of your controlled/team environment using anything but JS is just a loop for others to jump through.

I imagine a lot of projects have both CS source for development, with compiled JS for deployment. That might skew the stats between JS+CS, but it's a good indicator of CS' popularity compared to other languages.

Considering that, releases would be in JS, as they should be.

Your last paragraph confuses me.

So you had jQuery experience and were inhibited by working on a MooTools project. I get that, makes total sense.

But then you say that switching to CoffeeScript may seem like a better long term choice but "in reality it is not so much."

Surely MooTools would've been fine for you as a long term choice.

Yes, there is a penalty to switching. Whether it's jQuery to MooTools, JS to CS, whatever. But that doesn't make it a bad long term choice.

Also, CoffeeScript is more than just saving keystrokes. That makes it seem like it's all about initial up-front cost savings. Instead, it's about having shorter, more expressive code. That pays dividends for the life of a project.

"Instead, it's about having shorter, more expressive code."

yes, but at the expense of a lot of other negative baggages that comes along with it. by having it's own compiler rather than leaving that to the browser seems like it's opening a can of worms.

and just reading from the comments, i can see how we could soon see a birth of "another language", that's supposed to be solve all problems coffeescript'ers face.

for one, i guess readability seems to be the common problem, which isn't good for productivity.

It sounds like you haven't ever used CoffeeScript?

Don't you find it a little odd that you have such strong opinions on something you've never even tried?

i can see how we could soon see a birth of "another language", that's supposed to be solve all problems coffeescript'ers face

That's called incremental progress a d most people see it as a good thing. Others yell, "Get off my lawn!"

There's a bit of a contradiction in your statement. MooTools is much more idiomatic than jQuery in it's use of javascript inheritance, object creation, etc, so it's "more javascript". jQuery is so efficient to work with because it created it's own nice DSL, not because it is standard. The standard could be C, would that make you more productive too?

it was a point that even if you are merely switching a js framework where both are written in javascript, productivity can drop.

if you now switch to coffeescript, something not remotely close to javascript or any other languages fwiw, you'll experience un-productivity. it's akin to having your top developers walking in eggshells and doing nearly nothing close to what he is capable of.

i think coffeescript is great in the sense that we are trying to create a better language. but i also sense how much damage it can bring to budding startups who can't find talented coffeescript'ers to fill the void when their lead developer moves on to other things.

a one off go project, i'd say it's fine. but anything more complex, i think you'll see programmers wishing they had gone back to what they were familiar with.

What damage? You can drop it at any minute and start working on the compiled javascript. I've done this a couple times.

That's true for any tool/language/platform/whatever, everything has a learning curve (and coffeescript's is quite short). Going by your words we shouldn't ever learn anything.

And you shouldn't switch to a language if your developers are not comfortable with it. Do some experiments, try it out on a smaller project, then you might switch if you see good results.

> If you have to take the time to learn how to read something, it's not easily readable.

What? What language can you read without knowing it first?

If you're used to one C-style language, it's generally not hard to read code written in another. This is not to say that you'll grok it 100% accurately on your first go, but another C-style language will be more readable to you than, say, a Lisp-flavored one would be.

Of course that "readability" leads to devs writing C in Java, or what have you, and naking subtle mistakes, instead of learning the new language. See Dijkstra's "On the cruelty of really teaching computer science"

Though most examples are not something you'd write normally, this line is what kind of flopped the article for me:

    Given that CoffeeScript doesn’t fix any of the
    fundamental problems of JavaScript
same for a comment here:

    coffeescript doesn't bring much to table
There is so much coffeescript adds that I find these laughable:

    string interpolation 
    multi-line strings
    guarded loops (hasOwn...)
    easy constructor/prototype definition (class)
    avoiding global leaks/shadowing
    function binding
    splats, defaults
    safe usage of reserved names
    existential operator
    chained comparisons
    READABLE regular expressions
This just off the top of my head. These make a huge difference to writing code everyday. I've been writing CoffeeScript for 90% of my projects in the past 2.5 years and it pretty clearly affected my productivity for the better. I can only conclude that whoever says "coffeescript doesn't add anything" hasn't really used it for work.

Case in point, if you look at the linked article at the end you'll see an update by the author: I actually love CoffeeScript now that I've been writing it for a year.

I totally agree that CoffeeScript brings a lot to the table compared to old feature-less JavaScript. But the author is talking about readability of the language, and not about the features that it has.

The problem that he points out about "Implicit is better than explicit" and "There is more than one way to do it" are quite serious problems that modern languages shouldn't have.

See quote at the top, last paragraph from the post. His conclusion is that these readability problems swamp everything that coffeescript has to offer, which I think is really not the case - especially considering these are mostly style issues.

I don't consider this article bad at all, because it really makes some good points.

However a lot of criticism seems to come from the fact that the author is a Python programmer and therefor wants CoffeeScript to be more like Python. No, I am not saying something like the author not understanding it isn't meant to be Python, but for example the implicit vs explicit debate is certainly a more philosophical view. Ask people who prefer other programming languages and they will have a different view. Hey, after all that's why there are so many, even if Ruby, Perl, Python, Falcon, etc. have very similar use cases.

My opinion for example is that dynamic languages are very implicit in first place so saying you don't want that (at all) doesn't make too much sense. I also don't have problems reading CoffeeScript, but occasionally had problems with very explicit languages being too verbose which can make it harder to follow. So looks like my mind works slightly different here.

But back to the article. There are lots of valid points. I think coding guidelines, which one should have working in a team, no matter what language could solve some of them. Some points look a bit like mixing different styles on purpose and at least can't see how you could find something like that in the wild. Just because you can code ugly, it doesn't mean you have to, but again that is more of an opinion. Some people like usually verbose languages, because they say they are easier to read, others like ones with shortcuts or where you have multiple options to express things, making it easier to read (for some people). I for example always enjoyed the fact that Perl has unless and until in place of negated if/while.

But that's more what you prefer. But hey, CoffeeScript is all Javascript, so if your team doesn't like it it's (comparatively) easy to step by step switch back.

It's funny, because as a big fan of Python who has had to do a lot of CoffeeScript lately, I find CoffeeScript very similar to Python, and very readable. It's certainly more readable than JavaScript, when you don't get fancy. It's called "Unfancy JavaScript" for a reason.

Many of the issues raised in this article can be solved by "simply" not doing it (admittedly not always a real solution). Just because the language allows something to be done, doesn't mean it should be done that way. Our team has a styleguide that clearly explains good and bad practice, for CoffeeScript AND Python. (It's possible to do bonkers stuff in Python, too, just harder.) It includes things like: use explicit returns, especially when intending to return nothing; include parenthesis unless it's more clear without them (callback as arguments).

Yes, it'd be great if the language were more explicit and made it harder to do confusing thing. But, its core goal is being "just javascript", which prevents some of that explicitness. Also, the flexibility lets the real goal be clarity.

I find

    my_object =
        key: 'value'
        fn: (response) ->
to be more clear than

    my_object = {
        key: 'value',
        fn: function() {
because it doesn't have all the crap. CoffeeScript's whitespace syntax is even more helpful when those objects start getting nested. Plus, not having to deal with trailing commas is amazing. At the same time,

    someFn arg1, arg2, ->
    , arg3
can be confusing, so parens help:

    someFn(arg1, arg2, ->
    , arg3)

I try to not be "that guy" when it comes to CoffeeScript, but it's easily one of my favorite languages now. All the crap that JavaScript requires is just gone.

I much prefer this:

someFn(arg1, arg2, (-> doStuffInACallback()), arg3)

But far better, if you have any say in it, is making the callback the final argument to someFn.

someFn arg1, arg2, arg3, -> doStuffInACallback()

The thing about coffeescript is that most of the pitfalls being discussed here are very easily avoided with a modicum of experience and sense, whereas in javascript, many of the core issues simply cannot be avoided, and must be worked around in plainly ugly and bug prone ways (see: string concatenation, this/that juggling, "])})]));, just for starters).

both of those look really bad, try reading that sort of indentation when it's already 4+ indentations in, it's terrible

I do every day, both in CoffeeScript and in Python. I find things like braces frustrating, because I already keep it indented, and have to then manage the openings and closings. With proper 'aesthetics', indentation is perfectly readable. Plus, if something starts getting too indented than is readable, it's a sign that it needs to be refactored. (Whereas JavaScript looks messy even with otherwise sensible nesting.)

Sorry, CS and Python treat whitespace as a significantly different beasts. It's a disservice to everyone to say, "I read <insert significant-whitespace-block language> everyday, therefore <insert language that substitutes anything from call points, to auto-insert this or that, infer types, etc> is just fine!"

I think a majority can grok and parse Python, but the same can certainly not be said for CS. It's unfortunate too, because it has some Good Parts.

I didn't mean to say "Python does it, so CoffeeScript is not a problem", but rather that both can suffer from too much indentation. CoffeeScript is a little more prone, given the callback-heavy nature of JavaScript, but similar treatment of surrounding whitespace is helpful in maintaining clarity. (Basic example: I find Python's standard of four-space indentation helps CoffeeScript readability, and prefer it over the popular two-space indentation.) And in both cases, excessive indentation is a useful signal.

Readability is very subjective and depends on the user's knowledge of the language, as well as personal style, or 'accent', if you will. Code written with, for example, leading commas in dictionaries instead of trailing commas just looks bizarre to me and is a little harder to read, to me, even though I like to do something similar and stack colons.

yeah python is still much clearer than coffee. I used to somewhat prefer significant whitespace but coffee has definitely killed that for me haha. Seeing braces after looking at files of coffee is like a breathe of fresh air, it just lines up nicely for the eyes, even from an "artistic" stand-point I think the balance is more appealing

Question: am I the only one who's driven nuts by "if"s that come after the "then" part? E.g.:

    do_something(with, these, args) if im_supposed_to
I mean, the processor/interpreter always needs to evaluate the "if" first, so what purpose does it ever serve to put it after the "then"? To me, it just confuses things because it feels like code is getting executed backwards -- like crossing an intersection, and then checking to see if the light is green.

I know it works "in English" ("do this if that"), but when I scan other people's code I'll sometimes completely miss the "if" (sometimes it's just off the screen).

Are there any examples where this reverse-if actually helps, instead of harming, code intelligibility?

Postfix conditionals are pretty nice for conditional statements that are particularly unlikely to happen. Imagine, in a function body:

    write(file) unless terriblyUnlikelyThingHappens
    return results
... it helps to keep the expected main-line flow of your instructions reading smoothly -- much the same as the way you'd tend to use it in English. Doing it the "normal" way instead:

    if (!terriblyUnlikelyThingHappens) {
    return results
... makes it a bit more difficult to follow what should be going on in the function.

I personally find that I end up reading more lines of code than I need to.

When scanning and I see

    If (condition) { perform_some_action(foo, bar, baz); }
And I know condition is false I can ignore what follows but

    perform_some_action foo, bar, baz if condition
I now have to stop and parse the entire line. But not only that, because it can appear at the end I need to parse every line to make sure I don't get caught by some if or unless. Short lines are easy sure but it's still annoying to me.

yeah I agree here as well, postfix anything is quite annoying to read, especially so when chained, and the absence of parentheses makes this even more terrible

That would make sense if you were executing code in your head, but if you're reading you want to read all of it anyway...

Not at all true, most of the time you are reading code with a specific goal or area of interest in mind, and thus are only interested in the subset of code related to that purpose.

But @OriginalSyn has a pont. If you are reading code because something is misbehaving on a given configuration and you read

  if (some condition that doesn't hold on that configuration) ...
You can avoid everything inside that if statement if you know that's not important for the case you're considering.

But if we want to avoid it, we would be required to already know what's inside the parenthesis. If we already knew what's inside it, putting condition before helps more as in that case (1) Either it's true and you have to parse the block or (2) It's false and you just jump to the end of it.

I find it helps when you use short-circuiting. The returns are the first thing you see, making it more obvious what is going on.

  myfn = (arg) ->
    return 1 if arg is "one"
    return 2 if arg is "two"
    return arg.toInt()
This is obviously a little contrived with such a short method, but helps when you have stuff to do between conditional returns.

Yeah. The example is not very good, as it would probably be better to write as switch expression. But on a complex function, returning early can avoid the need to retain a lot of context in your head while reading that code. I find it especially helpful for returning/throwing on error conditions or breaking out of recursion.

I find they are useful for shortcutting out of functions:

    throw 'error' if badness
    return null unless goodness
Since it doesn't make sense to return or throw unconditionally it is makes it pretty clear that you should look for the if/unless.

Though, what's the benefit of putting the conditional after? I find these:

    if badness throw 'error'

    unless goodness return null
to be more readable

It's definitely nice to see the return at the beginning of the line -- makes the code easier to scan.

And anytime there's code after a return, it had better be connected to a conditional. :)

Those are good points. I'd probably make it multi-line:

    if badness
      throw 'error'
    unless goodness
      return null

If may be strange because of the asymmetry but I'd actually prefer `if` to be come first all the time and `unless` to come last all the time - precisely because `unless` reads like it's for exceptional circumstances:

  value = cache[key]
  if value?
    return value

  return file.readContents() unless not file.exists()

+1 reads poorly

Personally, i write my CoffeeScript with as much syntax as possible, to keep it as readable as possible: as a rule, I always use parenthesis and commas while leveraging CoffeeScript's "good parts".

But that said, the last couple of projects I've written in pure JavaScript and it's actually made things much easier to go back and read. Like the author of this post, I too have had difficulty reading back over my old code and, at a glance, understanding precisely what's going on.

Over the past couple of months I've gone from being absolutely devoted to CoffeeScript to on the fence, mostly due to readability.

(I'd love to see a fork of CoffeeScript that forces syntax.)

>I'd love to see a fork of CoffeeScript that forces syntax.

A compiler option that enforces syntax would be nice.

I have recently implemented to quite large project with CoffeeScript, the most recent one for a full export to HTML for http://flairbuilder.com, and I have to say that CoffeeScript is a superb language. It's a pleasure to work with.

The kind of problems that the article points out are easy to spot. If small syntax changes yield significant output code, that will be immediately reflect in the program execution.

Plus, a decent code base will have unit testing in place, which should catch more intricate, harder to catch, unwanted code flavors.

> The kind of problems that the article points out are easy to spot. If small syntax changes yield significant output code, that will be immediately reflect in the program execution.

> Plus, a decent code base will have unit testing in place, which should catch more intricate, harder to catch, unwanted code flavors.

That's rather optimistic.

I've been using coffeescript for about two years - I thought it was the best way to keep the pleasant aesthetic of python, which was my favorite language, and have been happy working with it ever since.

I just want to throw out a positive experience with it and to say that I don't support the author's premise of "here's some misleading things you can do with CS, therefore CS is unreadable". This applies to all languages, and 'fanciful' features and syntax should be avoided almost always in every language for the sake of readability.

There are MANY reasons not to use coffeescript, but JS is all functions all the time, and -> is my best friend.

I haven't used CS as much as yourself, but I don't find anything pretty in "element = =>".

That is a good example, and something that is eased by syntax highlighting (although I rarely implement double-bar function literals, keeping a pythonic 'self = this' reference is preferrable in most cases)

Although it is all subjective aesthetics, CS shines when paired with underscore.js, and really all "callback-as-last-argument" paradigms, ie:

  _.each items, (item) ->
     # do stuff with item
     # ...
(edited formatting, also what alec said below :)

If you use anonymous functions a lot, it's amazing:

    result = _.map object, (val, key) ->
is much cleaner than

    var result = _.map(object, function(val, key) {
        return foo(val)
I think.

I find this review very even-handed. It doesn't really make any bold claims, but it does provide a slew of evidence.

Eh. I think it's pretty silly. It basically boils down to "If you write terrible Coffeescript, it's hard to read". Duh?

Any time you are writing code that is ambiguous, use what you need to in order to clarify it. Don't omit punctuation just because you can if it makes the code less readable.

     option1: 1
     option2: 2
Using empty parameters in a function declaration is even funnier; you're adding punctuation that makes the code more vague. Just omit it!

   doSomething -> 'hello'
   doSomething(-> 'hello')
The third example is totally valid, and one I completely agree with - omitting parentheses on inner nested function calls makes for unreadable code. Don't do it.

The fourth example - complaining that inconsistent indentation produces inconsistent results - is just incomprehensible to me. When you're using a whitespace-delimited language, you'd best pay attention to your whitespace.

Ambiguities with optional curly braces are certainly nothing new to Ruby developers. Convention is that only the last hash in an argument list may have its braces omitted (or more stringently, a hash may only have its braces omitted if it is the last parameter in the list). Makes for finely readable code. Don't write unreadable code.

One of the fundamental rules you should follow when working in any language is "don't be clever". Optional parentheses can make for very clean code, but if you try to get "clever" with them, you end up with ambiguous code. List comprehensions are awesome but if you use them when there's a simpler construct you can use, you're doing it wrong.

You can write bad, unreadable code in any language. Part of maturation as a developer is learning to write readable code, not just working code. A bad workman blames his tools.

> "If you write terrible Coffeescript, it's hard to read"

No, it's "If a language that provides for many ways to write terrible code (incl many ways to write the same code), people will write hard to read code." For example if parens are required, it is impossible to "get clever" with not using them.

> A bad workman blames his tools

Great workmen evaluate and avoid bad tools.

> No, it's "If a language that provides for many ways to write terrible code (incl many ways to write the same code), people will write hard to read code."

A prime historical example being Perl. A second one being PHP.

Perl is a great example because it is an amazing language with a horrible reputation that it mostly deserves.

If writing good, maintainable code requires knowing/using a large number of rules and best practices that aren't part of the language (and whose absence are actually considered to be a virtue of the language), that seems like a flaw in the language itself.

Yes, you can write bad code in any language. That doesn't mean that languages don't have an impact on the ease of writing good and correct code.

If you think that the existence of style guides are a flaw in their respective languages, then I'm going to (as non-condescendingly as I can via text) suggest that you may not be a very mature developer yet.

Style guides and best practices exist for every commonly used language, and not because of flaws in the languages themselves, but because unless there is literally only one way to do things, a dozen developers will come up with a dozen ways of solving a problem.

"as non-condescendingly as I can via text"

Yeah, that didn't work out so well.

There's a difference between style guides offering a well-lit path to follow, and a language having a large number of possible expressions that are off limits in practice because they lead to bad code. The former marks out a path, the latter indicates a minefield. When the language's design actually facilitates the creation of bad code, not just by offering the possibility of dangerously ambiguous expressions, but by touting that possibility as a plus, that's a flaw.

Maybe accepting that flaw is worth it because the expressiveness of the language makes it a net plus; doesn't mean it's not a weakness or flaw or a point to be cautious about.

I do apologize - I didn't mean to be perjorative. I meant to communicate that I've heard the sentiment ("why does it let me shoot myself in the foot? It shouldn't let me do that!") espoused by many, many new developers, and it's an attitude that I've seen many people grow out of.

Coffeescript is a very flexible language. Its flexibility lets you do bad things with it, but the flipside of that attribute is that you can do very good things with it. I don't think that it's either "a path or a minefield"; I think those are two sides of the same coin. Optional parentheses and whitespace delimiting can reduce visual clutter, but if you abuse them and reduce it too far, you increase cognitive load on the person that has to read and maintain the code.

The more stringent the language, the less guidance you need in the form of style guides and community standards, but that's simply because the language itself imposes those standards on you as a byproduct of disallowing flexibility. It's absolutely something to be cautious about, but it doesn't make the language inherently flawed.

Syntax flexibility is not a virtue. Creating restrictive rules about what I can do to an object or how functions work ties my hands when I try to do tricky things, but syntax only limits my artistic urges to do esoteric code layout.

The C style syntax situation with optional if brackets is almost universally considered a flaw:

  if (foo)
This is not the flexibility to do hard things. It does not save a lot of typing or make some other common construct less cluttered. It is just a defect in the grammar that it's too late to fix.

It sounds like Coffeescript has more of these flaws.

I don't consider the optional braces in C-if a flaw. I use them for early return often, eg

  if (somethingBad)
    return NULL;

And I find that the braces are often just unnecessary visual noise. Of course, there is always the issue of people adding extra statements, but I haven't found that to be a problem in practice. YMMV though

The trick to avoiding the problem of extra statements is to only skip the braces if it all fits on one line:

    if (somethingBad) return NULL;
That's a clear visual indicator that the braces have been skipped, but it's basically impossible to look at the next line and not know that it's NOT part of the conditional. Also, if you're having trouble fitting it all on one line, it's a good cue that you're not in a situation where skipping the braces is safe.

I think there ought to be a distinction between semantic flexibility and syntactic flexibility, though. They may be correlated but it's not always clear cut. Lisp is an example of a syntactically inflexible (s-expressions or walk) but semantically rich language (macros).

My own opinion is that syntactic flexibility does have value (e.g. Haskell). I think it's also vastly overrated relative to that value; syntax is incredibly easy to bikeshed. Semantics require you to know something about the languages at issue.

That said, my opinion is that this kind of syntactic flexibility offers dubious value if the trade-off includes not just obviously broken programs but also subtly broken ones with errors which are difficult to spot. Newbie errors are part of the learning curve, but these seem like mistakes even experienced people can make.

> Its flexibility lets you do bad things with it, but the flipside of that attribute is that you can do very good things with it.

The problem is that there is a local maxima in the flexibility cost-benefit curve. After you reach that maxima additional flexibility degrades your experience.

I appreciate the apology, and I'm sympathetic to being quick on the draw with what look like the complaints of new devs.

But surely for a language that is almost purely designed to make syntactic improvements, the designer could have taken a stronger stance on syntax. I wish he could have at least decided on semicolons. The beauty of Python is that it is actually hard to write unclear code that does what you think it does. I'd love to see a PEP-8 for CoffeeScript or some kind fo built in linter.

Not saying that they don't exist, but I've literally never seen a python style guide (other than "don't mix tabs and spaces", which is part of the language now), never seen someone make a commit just to fix code formatting in python (something that happens quite often with every other language I use professionally), and never had any trouble working on other people's python.

People make the same arguments about Perl, but CoffeeScript doesn't have CPAN.

> Optional parentheses can make for very clean code, but if you try to get "clever" with them, you end up with ambiguous code.

Seriously? Using extra parenthesis to add clarity is being clever? I always thought it was being safe.

No, sorry, omitting parentheses (because they're optional) can be "too clever". I love optional parentheses, but I feel like a lot of people omit them because they feel that they're supposed to, at the cost of code clarity. I'm arguing that you should omit them when doing so doesn't make the code ambiguous, but if you ever need them to improve code clarity, that you should use them.

The OP had a fine example of "doing it wrong", with the nested inner function call without parentheses. While technically valid, it's bad code because it's difficult to read, even for someone accustomed to optional parentheses. Such a construct is "clever" when it shouldn't be.

But then you have to think everytime whether you should put parantheses there or not. Is the code readable without parantheses? Should I put them there in case someone reads the code in a year and doesn't know about the problem domain ... ? It is not much, but it is extra cognitive load that is wasted.

It's akin to programming C and everytime you write a function, if statement or for loop you have to ponder whether it looks best to put the opening brace below the line or at the end of it. I hate reading code with an inconsistent brace placement style like that.

I agree with you in theory, but in practice when working on large teams, languages as flexible as CoffeeScript do become a problem. That's a major reason languages like Java and even C# are so inflexible.

For my personal projects or small teams I'd have no reservations using CoffeeScript if it felt like the best fit. For large teams, I'd reconsider.

A few years ago I made an email campaign editor using JavaScript, I re-implemented it in CoffeeScript a few months ago. What I now want to do is re-implement it in the good JavaScript I've learned from reading the code that CoffeeScript generates.

There are some annoying parts of CoffeeScript. If your function takes callbacks as the first parameters, and another value as the second (such as setTimeout), you end up with some really awkward syntax. I've seen a lot of people define a delay method that swaps the parameters of setTimeout just so it's easier to use with CoffeeScript.

Like the author of the article, I have a Python background, but I've also written a lot of JavaScript, I love the syntactic sugar that CoffeeScript brings, but I hate reading CoffeeScript code. The project that my re-implemented email editor is used in has a hybrid of JavaScript and CoffeeScript code, and I'm not kidding when I say that everyone (myself included) groans when they have to work on the CoffeeScript parts.

    > If your function takes callbacks as the first parameters, 
    > and another value as the second (such as setTimeout), 
    > you end up with some really awkward syntax.
Hopefully no more awkward than the equivalent call in JavaScript.

    setTimeout (-> alert "later"), 300

    setTimeout(function(){ alert("later"); }, 300);
Or, on multiple lines:

      alert "later"
    , 300)

    }, 300);

But here we're introducing brackets just to avoid the syntax ambiguity. The Pythonista in me is crying out for a single way of doing things, either require brackets or don't. It makes me want to go down the road of always using parenthesis in CoffeeScript, but then it'll feel like I'm not using the language in an idiomatic way (which is something I hate doing).

In my mind, this usage of parenthesis is less ambiguous. Parenthesis are used in C-like languages for both function calls and order of operations. That is to say, they are used in these two ways:

    foo(a, b, c)  // functions
    (a + b) * c   // order of operations
Making parenthesis optional for function calls means parenthesis are used instead for just controlling order of operations.

In practice, CoffeeScript can't always guess when something is a function call (such as a function call without arguments), so you end up mixing in some function call usage no matter what.

I really hope it isn't guessing anything...

Personally, I prefer

    setTimeout -> (
      alert "later"
      alert "later"
      alert "later"
      ), 300

      alert "later"
      alert "later"
      alert "later"
      , 300)

I think it's much more clear and close to how we write JS code:

    setTimeout(function() {
      return alert("later");
    }, 300);

In my experience I would find myself frequently falling back to explicit brackets and more traditional Javascript syntax when I wasn't sure how to construct the statement. Which was okay (I wasn't stumped), but it happened fairly often. Your multiple line example (without "function") is the kind of thing I wouldn't be able to create on my own – both the indentation and comma have to be in just the right place, and the rules that govern that are not very clear. Or rather, it seems to be the one thing that works given a bunch of rules that are interacting in that particular situation. And each time I'd get something like that wrong it'd probably compile, but then I'd have to read the generated source to see what I'd actually done.

As for `delay`... I had to think hard to think why would it be a bad idea and concluded that it is non-standard and can confuse readers. If this is the case then just define a `swap` or `flip` function, along the lines of

    flip f = (a, b) -> f(b, a)
and use it inline:

    flip(setTimeout) 300, -> "it's fun!"
Sweet ;)

Making the callback the last parameter is very common in JavaScript too. It just makes things look nicer when you want to put anonymous functions in there. Underscore.js does this for all their functions.

To me, this says that you should define your callback as a function, and pass the function. If you're used to how to write functional code, this can be a clean way to do things.

That said, I've seen the same problem myself.

whenever I see something like this that tries to rag on coffeescript it seems like they all reference ryan florence's post from a while ago: http://ryanflorence.com/2011/case-against-coffeescript/

for the sake of clarification, ryan is sitting right next to me, writing coffeescript (as with all new code we write here at instructure, see: https://github.com/instructure/canvas-lms/tree/stable/app/co... )--and loving it. whatever "crazy gotchas" he found back then are obviously trivial to how much easier coffeescript makes life. it seems like every third tweet he makes is about how much he loves CS now: https://twitter.com/ryanflorence

Other than that, everything in this post just comes down to "just because you can doesn't mean you have to or should." disambiguate if helps make things readable

Perhaps it's time for Ryan to write that follow-up post ;)

One person has opposed CS and now using it.

Many people (including me) tried CS and back to JavaScript.

It's all about readability, team size, and choices.

I guess I'm lonely in this world, but I really want a statically typed JavaScript alternative. Dart is nice, but no JS interop (yet), GWT is just Java, which is not the most fun thing in the world, ClojureScript is nice, but not static typed and Lisp is a bit too extreme for me, but the new Scala JS DSL seems worth waiting for.

Be careful what you wish for.

I've been using GWT solidly for the past two years (not my choice for the project, but sometimes you just have to play the hand you are dealt) - and IMO GWT's biggest problem isn't so much that it is Java, but rather that it is Java without any dynamic features (i.e reflection, late binding etc.)

I could rant at length about the dark alleys that GWT's overemphasis on static analysis can take you, just a massive hairball of impenetrable compile-time code generation in order to provide features that would be trivial if you could just reflect on things. Ironic, given the fact that your code is actually running in Javascript.

So while I don't believe static typing is inherently a bad thing, you can definitely have too much.

Actionscript 3 was very good for this. I haven't done Flash development in a couple years, but i've been writing reams of Javascript (mostly backend). While I absolutely enjoy writing JS, I do miss the ability to declare types for methods, variables and parameters, which is incredibly helpful - especially as a project grows.

My IDE (phpStorm) helps quite a bit with this, currently - provided I document everything well - but I wouldn't mind having a compiler to smack me in the face when things are wrong.

Nope, you're definitely not alone on that. I've been thinking about experimenting with doing a simple staticly typed language that transcompiles to a readable JS. Roy [1] is really nice for example, but i think the generated JS is quite bloated, mainly because it maintains the lazy semantics of Haskell.

[1]: http://roy.brianmckenna.org/


I would say ActionScript, but FalconJS hasn't been released yet. There are other projects (such as roy) that are more functional in nature as well.

I have to try Haxe, but I can't help but wonder, it's been around for a while, how come it didn't get more traction? I assume I'll have to try and find out.

I was a flash developer way back (2003), I hear you.

Yes! I actually ended up in both places yesterday before seeing this comment, never thought I'd seriously consider Haskell, but now I start to understand why so many people like it.

Using CS for about 1.5 years now. Yes, there're ways to hurt yourself with bad CS code. This is also true for Python.

When developer writes some fancy one-liner in CS or Python and I'm reviewing his commit, I just ask him to come over and explain what this thing is doing. Sometimes it takes more than 20 second to read and explain even by author. Then I say: "You wrote this line an hour ago and it's already hard for you to understand what it does. Imagine you'll need to change it in a month. You'll hate yourself. Go and rewrite this code explicitly to help yourself in the future."

If you force yourself and teammates to write explicit code, you get all the pluses of CoffeeScript avoiding bad readability.

I concur with the author. When I write CoffeeScript I often use a more explicit or more c-like style for things I find ridiculously ambiguous or hard to parse when reading quickly.

Particularly, if one line has lots of commas, I put parens all over the place because no matter what I do, having to manually parse the comma arrangements and figure out what's nested and what's a function does not go quickly. I also find myself writing explicit returns somewhat frequently just so it's much more clear what's being returned.

The overarching issue is that the syntax blurs rather than pops -- there are too many instances where things that are significantly different in function look nearly the same.

Just because you CAN do something in CoffeeScript doesn't mean you SHOULD, or should ALL of the time. Just like in JavaScript where, if you want, you can write all of your code on one line, but you shouldn't.

    action true
      option1: 1
      option2: 2
If you really use CoffeeScript regularly, that isn't confusing. I don't usually write it that way though. I would do this:

  options =
    a: 1
    b: 2
  action true, options
doSomething () -> 'hello' just isn't valid code.

Usually I would write

    doSomething ->
      return 'hello'
or maybe

    doSomething -> 'hello'
If you write it the way he did, the parenthesis are confusing.

For this one:

    action = (token, i) ->
        @tokens.splice i, 0, @generate 'CALL_END', ')', token[2]
-- first of all, you always want to indent two spaces. But the main problem with that is in CoffeeScript you do need to use parenthesis after the first call because your code will be unreadable otherwise. You just can't write it the way he did.

    moveTo 10,

    doSomething 1,

    doSomething 1,
Those are ridiculous examples. No one does that. You just write moveTo 10, 20, 10

    doSomething (->
    'hello'), 1
Indentation is significant in CoffeeScript, just like it is in Python. Just having a parenthesis shouldn't change that. Anyway, what I do is this:

    somethingDo = (ms, func) ->
      doSomething func, ms
and then I can just write

    somethingDo 1, ->

    action(key: value, option: value, otherValue)
You don't write it like that if the function takes two objects. You would probably just write

    action {key: value}, {option: value}, otherValue

    x = {
      key3: value3
This is called destructuring assignment, and its part of the new ECMAScript. Its useful to not have to repeat key1, key2 everywhere when the variable with the value has the same name as the object property. The extra curly braces are just necessary to differentiate from the normal syntax.

Sometimes yes/no or on/off are more readable than true/false. That's an advantage.

    x = 1 if y != 0;
Don't use semicolons in CoffeeScript. I also don't put if statements at the end of a line because I don't believe that is very readable for most people including me. Also, in CoffeeScript its probably better to use isnt instead of !=, although rather than !=, you would want !==.

    if y isnt 0
      x = 1
I'm not that sure about the use of unless, although it probably is a little bit more readable overall.

  break for [tag], i in @tokens when tag isnt 'TERMINATOR'
  @tokens.splice 0, i if i
He says that is supposed to delete TERMINATOR from tokens. I tested it, it doesn't do anything, and the second example of the correct way doesn't work either. I think he meant this:

    filtered = []
    for token, i in tokens
      if not (i is 0 and token is 'TERMINATOR')
        filtered.push token

    i += block.call this, token, i, tokens while token = tokens[i]
Another example of something you CAN do in CoffeeScript but shouldn't. Actually it is a bad idea in general. I think a CoffeeScript programmer would actually write something like this:

    class Parser      
      block: (token) =>         
        @currentNode.push new Token(token)
      parse: =>
        for token in tokens
          @block token
I would never write something like 'mainModule.moduleCache and= {}'

Instead of

    js = (parser.parse lexer.tokenize code).compile options
I would write

    tokens = lexer.tokenize code
    parsed = parser.parse tokens
    js = parsed.compile options

> Just because you CAN do something in CoffeeScript doesn't mean you SHOULD, or should ALL of the time. Just like in JavaScript where, if you want, you can write all of your code on one line, but you shouldn't.

Isn't this precisely the problem? Why make a better Javascript then make the same mistakes as the original? One of the main goals is for Coffeescript to be more readable that Javascript. Having ambiguous syntax and so many different ways of doing things means that people will write code in all of these different ways.

Your rebuttal is a long list of what you SHOULD and SHOULDN'T do, but there are two problems:

1. Not everyone will agree with you.

2. People will write it the "bad" way anyway.

Ultimately this means less readable code, especially when you have to read someone else's code.

> 2. People will write it the "bad" way anyway.

People will write bad code in every programming language you give them. I don't see that as a justification for limiting the expressiveness of a language. If you take that line of reasoning too far you end up with Java.

> I don't see that as a justification for limiting the expressiveness of a language.

That's a straw man. I never said that the expressiveness of the language should be limited. There are plenty of languages that are just as (more?) expressive than CoffeScript without the syntactical ambiguity.

> If you take that line of reasoning too far you end up with Java.

That's the slippery slope fallacy. I am in no way suggesting that CoffeScript should have completely rigid syntax. Surely there is a happy medium between CofeeScript and Java. There are plenty of languages that live there.

Or python, which is where the author is coming from. In python there are fewer syntactic ways to write the same thing (semicolons aren't optional, they're just not permitted. The only ways to write and and or are 'and' and 'or', '&&' and '||' simply don't exist. Map literals always look like {key1: value1, key2: value2}, the braces are non-optional ).

Even python's critics would admit it's a very readable language, and it would be unusual to claim it's not expressive.

A nitpick: you can use semicolons in Python.

    x = 3;

The slope slips both ways. If you don't take that line of reasoning far enough you end up with Perl.

> as a justification for limiting the expressiveness of a language

Is it really more expressive just because there are N ways to do it?

Exactly. Expressiveness refers to the realm of problems a language can idiomatically address, not the number of ways a particular unit of logic can be equivalently written. If that were true, you could simply add any synonym for "if" ("when", "whenever", "assuming", "given" etc) to a language and claim the most "expressive" language.

Agree. When they invent a new language to fix some problems of the old one, if same problems (even more) appear in the new one, then why you need the new one? That's funny they try to fix some traps by introducing more pitfalls. What's the point? Those problems can be avoided by design, but they didn't.

In short, it was badly designed. Pointless to use it.

Also, his rebuttal doesn't defend the inclusion of those language features in the first place. If you simply "shouldn't" use a particular feature (e.g. if at the end of an expression), why should it be in the language at all? It shouldn't, ergo the language is nonsense, which was the original claim.

So, in short...

In CS, assign everything to a var, because good luck parsing CS tokenization.

In JS, assign a few things with a keyword or explicitly to a namespace, because good luck with global state(s).

Unfortunately, only the later is recommended, while the former is NOT and discouraged in lieu of being so verbose like the later. sparkles

The last example I assigned the results to a var because it was doing a huge amount of work, any of the steps could fail, and it was just a lot clearer that way.

You don't always have to assign things to vars.

If you have two or more function calls on the same line, use parentheses after the first one. Or use parentheses in all of your function calls. Its just an advantage for some people to be able to leave the parens out because it is a little more readable for them. Otherwise, you can just write it the same as you would in JavaScript.

Scope is another big advantage of CoffeeScript.

You had more than one example of pulling out "anonymous" objects from a call, because that is "how it's done." The point is that it is ridiculous that one must go to these lengths to help parse a language, especially when it flies in the face of the language's doctrine.

Same goes for parentheses "here or there but not there, sometimes." This may very well be the anathema for those that don't understand the claims that ";" or "function" or "}" are real burdens.

It doesn't go against the language's doctrine. The guy who invented it explicitly recommends pulling out anonymous objects and functions if they're not super trivial.

While I agree with your points,

    doSomething () -> 'hello' just isn't valid code.
Is valid code. Not that you'd ever actually want to use it:

    coffee> doSomething = (fn) -> if fn then fn() else (-> 'hi')
    doSomething = (fn) -> if fn then fn() else (-> 'hi')
    coffee>  doSomething() -> 'hello'
    doSomething() -> 'hello'
    coffee> doSomething () -> 'hello'
    doSomething () -> 'hello'

"coffeescript: the good parts"

Exactly, CoffeeScript isn't an excuse for writing incomprehensible code.

To make the OP assessment of CoffeeScript fair we would have to apply the same logic to Javascript its self. If we looked at any minimised JS code we would have to conclude that the language is unreadable.

> Just because you CAN do something in CoffeeScript doesn't mean you SHOULD, or should ALL of the time.

In 1948, Edward Murphy, Jr. an aerospace engineer, was working for the Air Force to install sensors to test g-forces on humans during deceleration. There were two ways to wire each sensor, and one of them was wrong. Someone had methodically installed every single sensor backwards. Murphy quipped something -- no one is exactly sure what it was -- which was cited in the Aviation Mechanics Bulletin as "Murphy's Law: if an aircraft part can be installed incorrectly, someone will install it that way." A few mutations later and this became the famous "Murphy's Law".

I think the original Murphy's law applies with great force to the world of programming languages. If there are multiple ways of doing something, and some of them are stupid, you still have to know all of those stupid methods because some idiot on your team will do it. The probability of this happening increases as the size of your team increases.

The proper course of action as a language designer is to make sure that foolish methods are not built into your language. What you do not do is design foolish methods into your programming language, and then put up a big sign on Hacker News saying "don't do that".

The Original Murphy's Law is so critical to the design of systems -- languages, APIs, libraries, etc., that after your post I feel compelled to ask you to put down your Programmer's Card and step away from the keyboard at this time.

Right because programming is just like installing parts into an aircraft.

Since you are obviously an expert in programming language design, I am curious to know, what programming languages have you designed?

He made it flexible in order to appeal to people who want a functional programming style and like to be clever. I don't want to deal with that type of code especially but some people actually prefer that way if that is the way they have been programming for a number of years or if they are just looking for something new.

"Put down your programmer's card" -- very uncalled for.

Any language can be abused or used in styles that are difficult to read or maintain. Almost all languages that I have seen have more propensity for that than CoffeeScript.

This one of the best language designs ever because of the significant whitespace (which to me is logically the most obvious rebuttal to your comment), small amount of required punctuation, intuitive and practical keywords, solving many of JavaScript's problems, appeal to Ruby enthusiasts and functional programmers which represents a sizable chunk of the best software engineers in web development, and compatibility with JavaScript. In order to do all of that it needs to have the flexibility that we are discussing.

CoffeeScript allows for the closest thing to my pseudo-code that I have seen.

Anyway I have spent way too much time on this discussion already.

> Right because programming is just like installing parts into an aircraft.


> Since you are obviously an expert in programming language design, I am curious to know, what programming languages have you designed?


> Any language can be abused or used in styles that are difficult to read or maintain.

These were three classic fallacies in a row.

As to the second one (ad hominem demands of authority): an in argument, it is never classy.

That being said:

> "Put down your programmer's card" -- very uncalled for.

That was meant as a joke but I can see that it might deservedly cause offense. And so I rescind it and apologize.

  > doSomething (->
  > 'hello'), 1
> Indentation is significant in CoffeeScript, just like it is in Python. Just having a parenthesis shouldn't change that.

The point, I think, is that the example above should be an error instead of compiling into something you wouldn't expect.

Here's a syntax I've seen a lot:

  doSomething ->
  ,   1
or even

  doSomething ->
That style is also handy to construct a long list of object literals.

> This is called destructuring assignment,

I think destructuring assignment is something else entirely. http://coffeescript.org/#destructuring

> its probably better to use isnt instead of !=, although rather than !=, you would want !==

'!=' is equivalent to 'isnt'. And '!==' is not CoffeeScript at all.

> Indentation is significant in CoffeeScript, just like it is in Python. Just having a parenthesis shouldn't change that.

Er... in Python, having a parenthesis does change that, within parens Python uses continuation lines which are not subject to the same restrictions as free-floating code.

Exactly this. Thanks for doing a nice writeup.

wow thanks for writing this all out. i was going to tackle a few of the examples but it was going to take some time. my thoughts exactly. a lot of the examples no one would do and not even the coffeescript page documents it like so. some other people are right though in that there isn't any discouragement so far in writing things certain ways.

I haven't programmed in Coffeescript, but does it bother anyone else that new variables are introduced without "var"?

When reading js, seeing the "var" really helps me to know the programmer's intent that this is a new variable being introduced, not an attempt to reassign one that should already exist. Does CS have features that make this point mute?

> I haven't programmed in Coffeescript, but does it bother anyone else that new variables are introduced without "var"?

Yes. I've come to see it as a mistake in Python and Ruby both, and I'm definitely not pleased by its usage in CS. Even less so as it uses the same rules as Ruby's lambdas without the stops of its methods and global scopes.

Actually I think this is a big plus, as it's much more common to want to declare a new variable than to want to reassign a global, and hard to notice when something is missing like "var". It would be nice if there was an explicitly different operator for reassignment. (p.s. I think you mean "moot")

For more discussion of this feature -- see yesterday's conversation on Reddit: http://www.reddit.com/r/programming/comments/zx137/coffeescr...

Nope, and I don't miss having to have all of the var statements everywhere.

Can you provide an example in js where knowing whether it's a new var vs. reassignment is useful? I can't think of a case, but it could just be me.

This doesn't bother me at all. Mostly because this feature comes from Ruby and so I'm already used to it, but also because it takes away the nasty surprise that JavaScript gives you if you leave 'var' out. I find this invaluable. As for knowing whether or not a new variable is being declared, well, that's why you keep your functions (and therefore your scopes) short.

> This doesn't bother me at all. Mostly because this feature comes from Ruby and so I'm already used to it

This and that are completely orthogonal though, I've been using Python for 8 years and Ruby for 6 so I'm used to implicit scoping. Doesn't mean I like it, or find it a good idea to use it in new languages (I don't).

And the issue is actually bigger in coffeescript than in ruby and python both, interestingly because (through javascript) coffeescript's scoping is much more regular than Python's or Ruby's.

The only language in which I don't know implicit scoping to be an issue is Erlang, because it uses bind-once and immutable structure semantics.

> also because it takes away the nasty surprise that JavaScript gives you if you leave 'var' out.

That can trivially be fixed by linting, or using strict mode. That's really no argument.

I think it depends on what you are used to. I find "var" to be noise.

The big point here: less characters for the same logic is not always better. Saying that you need less characters to write function x in coffeescript compared to javascript doesnt mean it is better.

Its all about syntax and semantics. Code from languages with a small and consistent syntax, one-way-to-do-it, and a easy semantics are easier to understand than languages with a lot of semantics.

For example, Scala is cool. But also overly complex. The same holds for C++. But languages like clojure are very simple to grasp and the semantics of these languages are very small.

    Coming from Python [...]
    For example, in C-like languages, you can omit curly brackets 
    after a conditional expression if you only have one statement:
      if (condition)
    But what happens if we add a new statement:
      if (condition)
Of course, that doesn't seem to be an issue in Python, the language he's coming from.

Most of his points are similarly contrived and could be equally applied to Ruby, a language which doesn't require parentheses and quite a lot of developers seem to like, or Python which is similarly whitespace-dependent and quite a lot of developers also seem to like. But inflammatory headlines do drive traffic from HN...

    Given that CoffeeScript doesn’t fix any of the fundamental problems of JavaScript
I think most developers who are well-versed with both JavaScript and CoffeeScript would disagree that CoffeeScript doesn't fix any of the problems with JavaScript. It certainly makes it easier to loop through object properties. It makes code more readable in a number of ways; post conditions, list comprehension-like syntax, simpler loops, the existential operator (`foo?`), and the maybe object-like syntax (`foo.bar?.baz?.quux`) are definitely improvements.

You can of course write bad code in any language, no matter how structured it is. I myself am about to embark on refactoring a massive, terribly-constructed Python system that processes millions of requests each day. The author didn't follow the strictures of the Python community at all. How is that any different from someone who exercises poor judgment when writing CoffeeScript? It's not.

The issue with the optional curly brackets is that for new users, or even experienced users reading through code quickly is that at first glance:



Both look like they should execute the same way - running the indented code if the condition is true. The reason this is a problem in C and not in python is that in python, this code operates as it intuitively looks, while in C, it does not.

(It can trip up experienced users if they're not paying enough attention and their mind just assumes the curly brackets are there.)

The issue is that:

  * Humans use indentation

  * The Compiler ignores indentation

  * The indentation and the the compilers interpretation are different

Can you give examples? I'd really like to see some unreadable python, and other than nested lambdas I can't see any real ways to screw it up at the basic syntax level. Even golf-style python tends to have a structure that's easy to follow (though what the code's actually doing can be another matter), and I feel it's significant that there's no obfuscated python contest.

This is exactly why I stopped using CoffeeScript. I hate Javascript so I was excited when I saw CS for the first time but after a while I realized I was doing two calculations in my head:

1: What is the JS that CS was generating?

2: What is the JS doing?

Maybe I just didn't give it enough time, but it seemed quicker and more reliable for me to just suffer through the JS syntax.

Both statements do completely different different things, although they look very similar.

This is true for both CoffeeScript examples he gave and their resulting JavaScript.

I agree, it is too bad that CS has so many optional ways of doing things. That said, it is possible to write unreadable code in pretty much any language. If you have a good style guide and stick to it, it is easy to avoid all of these issues.

Great article, totally nails it. No "good" language design should be so ambiguous

After reading this, I felt the urge to classify key aspects of write and readability of the programming languages I've dealt with over the years.

How easy or fast it is...

  - to write what I want to achieve
  - to write something that makes me feel creative
  - to write something that makes me feel tidy, organized
  - to write a syntax error
  - to write a logical error
  - to step into a "gotcha"
  - to find that bug
  - to grasp a short snipped
  - to grasp a 20K loc app
  - to read my own code 1 week later
  - to read my own code 3 years later
  - to read somebody else's code
Now trying to rate languages by these parameters is a tough job that I can't even imagine how to tackle. In many instances it doesn't even has to do with the language itself, but rather with my experience and relationship with it. But any exercise in this direction gives some food for thought while trying to find the right tool for the next job. Or something to think about when looking back at some of the choices I've made through my programming lifetime.

- to grasp a 20K loc app

Although, 20k lines of Python is probably doing much more than 20k lines of Java.

Yeah, well that's just, like, your opinion, man. -- The Dude.

Personally, CoffeeScript has been my favorite thing to come about in the past few years.

CoffeeScript readability > Javascript readability

and that's all that matters to me

But the whole point of the article was that that isn't true, because CoffeeScript makes more things optional (e.g. in Javascript if you don't see brackets then you know it's not a function)

Yeah that's obvious, but I disagree with "the whole point of the article" hence my previous statement.

Thanks for this post. This is actually how I feel about CS but could never actually explain it correctly. I find it very hard to read other people code because of all these implicit rules that are more often than not ambiguous. Yes, sometime it's shorter but not necessarily faster or easier to read.

Some of the complaints in the OP aren't mistakes that I make often, so I guess I don't see them. Perhaps for someone that makes those types of mistakes often, it'd be a detriment, but so far, I've been happy with my coffeescript experience.

The moment you try to do simple things like concatenating a string you realise that javascript is a bit more tedious than it needs to be.

While nowadays I prefer coffeescript to js, I would rather have some syntactic sugar like scss is for css. I've looked into it but nothing has enough momentum to make me sure it will still be around in two years or so.

As a sidenote, while I mentioned I prefer CS, have you ever tried using a ternary operator? or pass a callback function in jquery (ex. the hover out event)? or just return "this". There are so many things that feel so wrong.

Feel so wrong in... CoffeeScript? Or JavaScript? I think ternaries and callbacks are both much cleaner in CS (with the exception of "callback, something-else" which does end up a little funky unless you add parens). Just naming a variable to return it seems a little wrong to me but you can always leave the "return" in. "Optional" is very different than "never use".

Those last things feel wrong in coffeescript. Honestly, what's more readable:

  x = a ? b : c

  x = if a then b else c
And for the last one:

  function() {
    // doing something
    return this;

    # doing something
Don't get me wrong, I love the fact that I can do nice stuff such as callback?() and some other nifty tricks, but while I'm still having fun using coffeescript, in my experience it's not that great as most people make it up to be.

Just FYI, using standalone `@` isn't considered to be good CoffeeScript style. It's discouraged for this very reason -- just use `this` instead.

The reason why it works in the first place is for consistency ... in every other context, you can use an `@` to replace a `this` or `this.`.

FYI, I usually use this instead of @ myself, but in this case I used @ to empathize a grander issue.

And it's not like there's a warning about it being removed on coffeescript.org where it's being mentioned as an alias for this in the first place.

I prefer words to symbols any day. I think coffee's ternary is better. Also, I have no problem throwing singular @s around cuz they're short and get used so often, but typing out 'this' would be fine too.

I myself enjoy using CoffeeScript, but I understand it might not be for everyone.

It's always good to look at the pro/cons before choosing a language, so I think discussions like this is good.

That being said, some of the problems this article points out can be addressed by going back to more JS style.

e.g. Not relying on implicit parentheses/commas/braces

I like the "Fancy" for loop because it's closer to what I do in Python (yes I know they are not exactly the same). Same thing with the "Tricky" if statements.

Also the redefinition shortcuts provided in CoffeeScript is pretty much what I do in plain JS anyway.


JS: foo = foo || 'bar';


  foo or= 'bar'

CS 1.3.1:

"Conditional assignment of previously undefined variables a or= b is now considered a syntax error."

Yeah I know. It was just a quick example.

It's useful for something like:

  f = (options) ->
    options.foo or= 'bar'
    # ...

Here's my own anecdotal example of CoffeeScript's readability.

Once, before banging out a bit of complex code, I thought I'd write out some psuedocode to help wrap my mind around what I wanted to do. I type out my psuedocode algorithm, realize I've written valid CoffeeScript, then uncomment it.

That's the experience that made me think CoffeeScript is MORE readable than JavaScript.

I'm meh on significant indentation, but significant space characters is just a terrible idea. I blame Haskell.

ParenthesEs is the plural, parenthesIs is the singular.

Normally I don't do the spelling/grammar thing, but it was consistently wrong in the article, and at least a half dozen comments here were getting it wrong, too, so I figured a public service announcement was in order.

wow readability is about API design like requests vs urllib2: https://gist.github.com/973705

except when APIs get 100x bigger, API design becomes 100x more important. and using a language ecosystem that encourages good API design, over a language that is cobbled together and has no particular opinions about what good API design looks like[1] (or, even, bad opinions[2]), so that less experienced developers end up with decent output, is that much more important than concerns like parens, commas and braces. curly braces never made a project fail. shitty API design does.

[1] javascript! [2] java!

sheesh. there are bad things about coffeescript, but this stuff hardly matters.

Oh dude, just learned something new from this article. options = { a b c } generates options = { a:a, b:b, c:c, } I have a lot of maps in my app that look just like that. It'll save me a lot of repetition.

May I ask what are the use cases for maps like that? I don't think I have ever needed one myself.

I have situations like this quite frequently where I need to turn vars into members of an object:

    title = getTheTitle()
    img = (shouldWeGetImg) ? getImg() : "placeholder.jpg"
    body = ...
        # 8 lines of code to define body, some that reference title and img

    # this:
    obj = { title, img, body }

    # instead of:
    obj = {title: title, body: body, img: img}

Pretend there are newlines after a, b, and c. HTML ate my whitespace :P.

while the points in the article make sense and I agree, if you keep your Coffeescript code consistent I really don't suffer a readability problem, especially relative to JS where there's like 5x more code to go through

> you keep your Coffeescript code consistent

That is not enough, because unfortunately, there are more programmers in the world than "you", and I have to use their code and / or work with them.

When language has X options their is X! opinions on "consistent".

That's an unfair hyperbole. CS saves you at most ~30% characters, and that's when omitting every parenthesis possible.

I personally don't get excited by that prospect. Generators would get me excited, though!

well, personally when i converted over i never had to stare at these blocks anymore: } } } });

made code a lot easier to debug. and yes, consistency wasn't the only thing. i would say that converting over has a learning curve sure reading other people's code. but sacrificing readability long term compared to vanilla js? i wouldn't say that. i'm not saying there isn't ambiguity in the beginning but eventually you learn how to not do things in ambiguous ways.

I had this same feeling programming in ruby few years ago when no formal language grammar was defined (with really weird problems with : vs [space]: AFAIR). Is it the same for CS ?

on one of the slide in this presentation i found this interesting ( https://speakerdeck.com/u/clkao/p/livescript-tax-free-javasc... )

   Javascritp + ruby  = coffeescript
   coffescript + perl = coco
   coco + haskell     = livescript
So i think coffeescript is not the end of the line, and i think this mean, there is a serious need for a better javascript

Badly written code is badly written code in ANY language.

I wrote Dashku entirely in CoffeeScript. It was a productivity boost, and for that I owe it my firstborn child.

Seems like there are two types of developers: those who don't want even to hear about CoffeeScript and trying hard to find deficiencies in it and those who embraced it and feel unhappy every time they are forced to work with vanilla javascript again.

Or, there are different opinions on how things work, each finding their language superior in writability and readability.

There are no absolutes. This is all subjective.

Extremists are generally the noisiest. I use CS in a Rails project along side JS without a problem. It's not a massive web site but I haven't hit any big snags mixing the two.

Sounds pretty accurate to me. It's the "unreadable" problem that boggles my mind.

"OMG it's change, run away!"

"readability is more important than writability for a programming language"

Couldn't agree with you more. Readability is measurable and considering that a lot of programmers come from C-type languages, I really don't see why people try to modify an existing formula (as badly implemented as that formula is in JS).

tl;dr: Let's face it, if you're using Coffeescript or similar because there are too many braces or parenthesis, perhaps you should quit development right now and take a class on basic algebra, quit that and save the world your horrific code. Seriously.

Readability is measurable, that's true, but the second part of your sentence makes me wonder if you really understand what readability is.

Do not confuse readability with familiarity. Also, don't forget that readability is a function of both syntax and semantics.

Also, I really don't know what connection there could be between aesthetic preference and quality of code produced. Are you aware of some study that show that kind of connection? With conclusion which says that if I like line noise then I am a great programmer, and if I like whitespace I am bad? If not then your 'tl;dr' is just stupid, sorry.

This isn't about aesthetics, it's about grammar. Just as in written human languages, computer languages have grammar that make things easier to read and understand. How well that grammar is done depends on the language. My point is that if aesthetics is one of the reasons you choose a language, you really don't understand software development or the simple idea of "use the right tool for the right job."

My point is that if aesthetics is one of the reasons you choose a language, you really don't understand software development

Some would disagree: http://paulgraham.com/pypar.html

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact