[Confusion]

If you pass a function as an argument to a function, you're still separating what from when: the function declaration shows what; the invocation of the function shows when. It doesn't matter whether it's

  def foo(f):
    f()

or

  (define (foo f) (f))

That has nothing to do with annotating a function to show it binds to some URL and is invoked when an HTTP GET request is received with that URL. I don't like having to scroll through a source file to discover such bindings and I actually prefer them to be defined at a separate place. In that sense, the Python 2 solution isn't good either and suffers from the same problem I described earlier: people try to be so concise that they loose sight of other goals.

[steveklabnik]

The Ruby code is not an annotation. It's very close to passing an anonymous function to a function. The definition of get is

    def get(path, opts={}, &block)

See that &block? Here's the actual source of get:

      # Defining a `GET` handler also automatically defines
      # a `HEAD` handler.
      def get(path, opts={}, &block)
        conditions = @conditions.dup
        route('GET', path, opts, &block)
 
        @conditions = conditions
        route('HEAD', path, opts, &block)
      end

Think of it as "I'm registering this block as a callback, to be called when this url is processed." I'm not sure how else you'd like to have a web DSL written; every web library I've used has worked this way.

[Confusion]

The Ruby code is not an annotation.

Which is exactly the problem: it should be, because it is meta-information about when the function is supposed to be called. It's not a good idea to wrap that into a function, together with the code that is supposed to be executed, because you are mixing two completely different kinds of information into some new function. This code is being too clever for it's own good: it condenses information so far that you are forced to think harder about what each piece means. I consider this is one of the main dangers of languages like Ruby: it's proponents lose sight of the fact that they are coming full circle, back to writing code that's so clever that no one in the future will understand it.

every web library I've used has worked this way.

I doubt that we've used a disjoint set of web libraries, so my only answer can be: no, they haven't worked that way. Most libraries map paths to named functions separately from implementing the functions.

Conciseness is not an ultimate goal; in fact, it is rather the opposite, as those that suffered under the clever C hacks of others will tell you. To paraphrase Santanyana: people have forgotten the past and are repeating it.

[steveklabnik]

I guess we'll just have to agree to disagree on this. By keeping the function definition near the route instead of just defining a name, I feel that it's much, much cleaner.

> I doubt that we've used a disjoint set of web libraries,

I guess I mis-spoke slightly. I meant assigning a function to a route, because I consider that (in this case) it's an anonymous function to be an irrelevant detail. Basically, to me

    get "/hello", hello
    def hello
        puts "hello"
     end

to be the as clear or less clear than the way Sinatra does it. Rails and Zend (the two frameworks I've used the most) put that 'def hello' in another file, elsewhere.

Anyway, thanks for the discussion. We've apparently boiled it down to taste, so it's a good thing there are multiple languages, I guess.

[Confusion]

Thanks to you as well :). I even may come to see it your way in the near future, as it seems I'm going to be doing some Ruby coding for my new employer.