Update We’ve moved the date of our first meeting to June 17th, so you’re still in time to join us! If you want to follow our adventures, you can also ask for an invitation to our mailing list.

The other day, Andy and I met Jos, an experienced schemer who lives near Barcelona, with the idea of having lunch, talking about Scheme, and create a Scheme Users Group. After a bit of discussion, we agreed on widen the group’s scope, and start what we’re calling Fringe Languages In Barcelona (FLIB). The plan is to conduct periodic meetings with a main presentation followed by some lightning talks (the latter were a complete success at ILC, and we’d like to try and see how they work for us), with as much discussion interleaved as we see fit. We’ll have some refreshments available and, since we’re meeting in the very center of the old city, visits to pubs or a restaurant for dinner and further socializing are to be expected.

As i said, we’re expecting much discussion about Scheme and Lisp, but we’re not ruling out by any means other fine languages. For instance, the talk for the inaugural session (scheduled June 10th17th, 7:30 pm) is entitled The implementation of FUEL, Factor’s Ultimate Emacs Library, and it will include a short introduction to Factor (yes, i am the victim speaker). Jos will come next, the same day, with a lightning talk about PLT Redex. We have free slots for more lighting talks: you are invited not only to come, but to give one if you’re so inclined. This being our first meeting, there will be also some time for logistics and organisation.

So, if you’re near here by then, by all means, come in and join the fun:

Calle del Pi 3 Principal Interior (first floor)

Not really needed, but if you’re thinking about coming, sending me a mail beforehand will help us to be sure that we’ve got enough food and drinks.

We’re looking forward to getting FLIB started, and we’re sure that at least grix more fringers are coming! Don’t miss it!

Erlang Now!

I don’t know whether Monty Python ever wrote a gag on programming languages, but if they did, this Erlang video must be it. The funniest thing is that it is pretty serious, and does a great job showing one of my most cherished abilities when using dynamic languages, namely, adding new functionality to a running system on the fly. As for the Monty Python bit, well, you have to see the to know what i mean: i kept laughing out loud during most of its twelve minutes (those Ericsson engineers seem to be taken from The Larch, but then maybe it’s just my sense of humor).

Update: Mike, one of the engineers in the film, has been kind enough to post a comment about the experience, which I’m reproducing here for your convenience:

We gave a well received “demo” in 1990, conjunction with ISS90, a big telecoms conference in Stockholm. We made this movie to record the demo. We actually used a professional company to do the filming, but I won’t mention their names as they would probably sue me for libel.

The worse of it all, is that we were deadly serious at the time. The Monty Python aspect must be due to our backgrounds. Of the people involved, Joe is English, Robert is Swedish – but brought up in Australia, Bjarne is also Swedish but spend some formative years in Scotland and I’m half Welsh half Scottish with some Irish thrown in somewhere.

In 1990, when we made the movie, the very idea of using anything other than C, Plex, assembly language etc to design embedded concurrent systems was heresy, we expected to take the world by storm. It seems that the cheap communication and multi core processors are giving Erlang a boost 16 years later. Well at least in the intervening time we have tested the hell out of Erlang and its implementations!


PS. If you look carefully at the film, you can see that Erlang at that time had a Prolog like syntax.

PPS. I can’t watch the movie without laughing (at) myself

While we’re at it, let me mention that i like many a thing of Erlang. It’s a curious mix of good stuff: a simple syntax and kind of minimalist flavor that reminds of Scheme, pattern matching and functional variables like Haskell’s, and what amounts to a programming paradigm of its own based on mailboxes and processes (which, as you surely know, are amazingly cheap). Also worth mentioning is Erlang’s error handling philosophy, which is, at first sight, a bit startling (i’m not still sure if it makes perfect sense, but after playing with the language a bit, it looks like it does–this is an interesting post on those matters). Definitely worth a look: see for instance Joe Armstrong‘s thesis, or, if you have some bucks to spare, the forthcoming Programming Erlang.

Erlang R11B-5 in Fink

My good friend and Fink packager extraordinaire Aleix has kindly heeded my request for updating Fink’s erlang-otp port to the latest Erlang release: you can find it (with installation instructions) here. It’s working like a charm on my machine. Sweet!

Scheme on the Erlang VM

Recently, someone in the Erlang-questions mailing list wondered about the possibility of writing a Scheme-to-Erlang compiler, possibly inspired on earlier work by Marc Feeley and Martin Larose on Etos, a Erlang-to-Gambit compiler (don’t miss the great article Compiling Erlang to Scheme about Etos’ implementation). Well, Marc has risen to the bait and produced a prototype implementation of Stoe: join the fun and get it here!

Erlang, Termite and a Blog

In his soon-to-be-published book, Joe Armstrong proposes the following exercise:

Write a ring benchmark. Create N processes in a ring. Send a message round the ring M times. So that a total of N * M messages get sent. Time how long this takes for different values of N and M.

and he adds

Write a similar program in some other programming language you are familiar with. Compare the results. Write a Blog and publish the results on the Internet!

Since i happen to be learning Erlang and, besides, know of another programming language providing Erlang-like primitives, i thought i’d take Joe’s bait and give you a little taste of how Erlang feels to a newcomer.

This is my implementation of the ring in Erlang. If you’re an Erlang hacker, be prepared to be scared: this is my first longer-than-ten-lines program–but, please, do not hesitate to criticize it as deserved!.


make_ring(N, M) -> spawn(fun() -> sender(N, M) end).

sender(N, M) ->
    FirstPid = self(),
    NextPid = spawn(fun() -> tunnel(N, 2, FirstPid) end),
    do_times(M, 0, fun(I) -> NextPid ! I end),
    NextPid ! done,
    receive done -> done end,
    {_, Time1} = statistics(runtime),
    {_, Time2} = statistics(wall_clock),
    U1 = Time1 * 1000,
    U2 = Time2 * 1000,
    io:format("Total time=~p (~p) microseconds~n", [U1, U2]).

do_times(N, N, _) -> done;
do_times(N, J, Fun) -> Fun(J), do_times(N, J+1, Fun).

tunnel(N, N, FirstPid) -> tunnel(FirstPid);
tunnel(N, J, FirstPid) -> 
     tunnel(spawn(fun() -> tunnel(N, J+1, FirstPid) end)).

tunnel(Pid) ->
        done -> Pid ! done;
        Any -> Pid ! Any, tunnel(Pid)

Here you can see some of Erlang’s niceties: first class, anonymous functions (using the fun() -> [body] end lambda constructor); function definition by cases (a la Haskell: guards are also admitted) and pattern matching; message mailboxes using receive (also using pattern matching) and the send operator !; easy process creation with spawn; and the pretty no-frills but extremely convenient module system (one would write ring:make_ring(A,B) to call our ring launcher: the Erlang shell will find and load the ring module as long as the file is in the load path). You may also have noticed that there is no variable mutations, or that lower case atoms (like done) are automatically interpreted as symbols. Tuples also make an appearance: they are constructed using braces (as in {_, Time1}) and you assign values to its components, again, by pattern matching. And, oh, those tail recursive calls to tunnel and do_times are properly optimized by the Erlang interpreter.

Although the above is just a subset of the language’s features, it’s enough to show why i like Erlang: it reminds me of Scheme in many ways, while borrowing some interesting bits from ML languages and even Prolog. On top of it, it fulfills Alan Perlis desideratum on programming languages: it changes the way you think, as recently noticed by one of my favourite bloggers. (It is also worth mentioning that Erlang comes charged with a pretty nice library, excellently documented, that has recently gained a slick search interface.)

But back to Joe’s problem. As it happens, there’s ‘one other language i’m familiar with’ (Scheme) that provides and Erlang-like library (Gambit‘s Termite), and rewriting the program was a breeze:

(define (make-ring n m) (spawn (lambda () (sender n m))))

(define (sender n m)
  (define (display-time msg start end)
    (display (list msg (- end start)))
  (let* ((fist-pid (self))
         (next-pid (spawn (lambda () (tunnel n 2 fist-pid))))
         (rt (real-time))
         (ct (cpu-time)))
    (let loop ((m m))
      (cond ((= 0 m)
             (! next-pid 'done)
             (recv ('done
                    (let ((nct (cpu-time))
                          (nrt (real-time)))
                      (display-time "CPU time: " ct nct)
                      (display-time "Real time: " rt nrt)))))
            (else (! next-pid m)
                  (loop (- m 1)))))))

(define (tunnel n j first-pid)
  (cond ((= n j) (send-recv first-pid))
        (else (send-recv (spawn (lambda () 
                                   (tunnel n (+ 1 j) first-pid)))))))

(define (send-recv pid)
   ('done (! pid 'done))
   (x (! pid x) (send-recv pid))))

If you’re a Schemer, you’ll readily understand the code above, which uses the same primitives as Erlang: !, recv (since receive is taken in Gambit) and spawn.

What about the comparison? Well, i love Scheme and, to my biased eyes, Gambit’s version is more beautiful. So let’s move to something that is not in the eye of the beholder: speed. Running ring:make_ring(5000, 1000) in my laptop gives the following output:

Total time=970000 (1520000) microseconds

while typing (make-ring 5000 1000) at the Gambit’s interpreter prompt yields:

CPU time: 19.442507999999997 secs
Real time: 19.780327081680298 secs

So, the Scheme interpreter is about 20 times slower than the Erlang interpreter (the factor seems to be more or less constant when you change the number of processes or messages sent). But Gambit has a secret weapon: it’s a Scheme to C compiler (why, it’s actually called Gambit-C!). So I compiled my program and ran it, with the following result:

CPU time: 2.3031090000000005 secs
Real time: 2.3484270572662354 secs

The interpreted Erlang code is still more than twice quicker than the compiled C code. Not bad: the more i look at that Erlang program, the more beautiful it looks!

Of course, this is just a toy benchmark which says nearly nothing about the real performance of either Erlang or Termite. I just thought it could be a way of whetting your appetite and get you started in these to systems that i’m just discovering. Googling for Erlang and Termite will direct you to several recent blog entries discussing them and helping you to join the fun, but, lest you miss it, let me recommend this Termite mini-tutorial by Marc Feeley.

And again, if you discover any of the silly things i could do better in the programs above, please leave a comment: i’m just learning!