I don't understand why this is downvoted. Does somebody disagree? Why?

It's true, every time I see CoffeeScript vs JavaScript examples, they're often unfair. Some addressed in this older thread[1].

Another thing I don't understand is why make the parenthesis for function calls optional? Such as

    sum_and_difference 5, 2

Especially since one of the most common complaints about JavaScript is ambiguous inconsistency (var, semicolons).

Ok, so parenthesis are optional,

    foo = myfunction()

calls myfunction, but

    foo = myfunction

does not? I suppose this is better than the way Ruby does it where you have to jump through hoops to get a reference to your function, but is this really better than making parenthesis for calls mandatory?

[1] http://news.ycombinator.com/item?id=2542492

To me the most convenient thing about it is the auto-declaring variables and proper loops. Whether the sugar is worth it or not comes down to taste. I used Python as my primary language from 2001-2007 before switching to Javascript/Coffeescript full time so I'll provide a bit of commentary.

> Arrays could use a beard or the varied language syntax is going to cause ingrown pains

The whitespace magic only causes me a problem in a few situations.

Watch indent on wrapped if statement:

    if foo is true or bar is false or
        baz is null
        doSomething() #Error

    if foo is true or bar is false or
       baz is null
        doSomething() #ok

Explicitly returning objects is awkward:

    foo: ->
        if not ok then return
        #lots of code here
        return
              x: 1 #ha ha
              y: 2

    foo: ->
        if not ok then return
        #lots of code here
        return {
              x: 1  #ok but ugly
              y: 2}

I always want to use implicit call parens with property chaining but that doesn't work:

    $('.doAllTheThings').filter 'span'
                        .data 'foo', true
                        .wrap '<div/>'
    #produces filter('span'.data('foo' ...


    $('.doAllTheThings').filter('span')
                        .data('foo', true)
                        .wrap('<div/>')

The result of assigning a comprehension threw me off for a while because I was thinking Python list comprehension instead of postfix loop:

    foo = x + 1 for x in array

    foo = (x + 1 for x in array) #what I expected
    
    (foo = x + 1) for x in array #what happens
    for x in array  #same thing written the other way
       foo = x + 1
    

Finally, the whitespace magic working well tends to depend on object/functional arguments coming last in the call list with the functional arg coming after the object.

These annoyances are relatively rare and are the sort of thing you just get used to if you use the language heavily. The upside is that for an appropriately designed API (the options object argument last convention in JS makes this fairly common), the object literal magic gives a feel like python kwargs:

  animatedShow node, duration: 200, onComplete: -> fire 'doThing'

You can also do DSLish things fairly easily. I'm a fairly heavy YUI user. Compare:

    YUI.add('myCoolModule', function(Y, name) {
        Y.CoolModel = Y.Base.create('coolModel', Y.Model, [], {
            aMethod: function(){return this.get('foo') === 1}
        },{
            ATTRS: {
                foo: {value: 1},
                bar: {value: 2, readOnly: true}
            }
        })
    }, '1.0.0', { requires: ['app', 'model'] });

to:

    YUI.module 'myCoolModule',
      version: '1.0.0'
      requires: ['app', 'model']
      (Y) ->
        Y.CoolModel = CoolModel
        class CoolModel extends Y.Model
            @ATTRS:
                foo:
                    value: 1
                bar:
                    value: 2
                    readOnly: true

            aMethod: -> @get('foo') is 1

It requires a one word patch on Coffeescript (superclass name in generated output) and the addition of a tiny YUI.module to YUI.add shim but I think the result is a lot cleaner.

> It's also bizarre from a Python background

Python's auto-concatenation of runs of string literals is convenient but I don't know of another language that does the same thing.

> if foo.bar? then console.log foo.bar

The ? is an explicit existence check, while doing `if(foo.bar)` will fail if bar is an empty string or zero. I tend to write existence checks using the postfix form:

    console.log foo.bar if foo.bar

This is a personal style that indicates that the normal control flow is to log but there's a simple guard on the operation. I'd use the if...then form if I expected the test to fail a significant fraction of the time (i.e. it's a branch instead of a guard). My other quirk is that I use &&/|| only for short circuiting but and/or only for boolean logic.