Automatic Translation

I find Google Translate frustrating. Why do I need to tell it what language my text is in, or what language I want it translated to? Can’t it work it out?

http://bluebones.net/translate/ is a wrapper for Google’s translator that automatically detects the language of source text or web site and translates to your language (based on browser settings).

You might also find this bookmarklet useful. It translates the page you are currently on into your language. Drag the link to your Firefox toolbar: Translate

I was inspired to do this by Mike Linsvayer’s post on the same subject.

Do Not Reply

I don’t think I will ever stop being annoyed by receiving emails from do-not-reply@ email addresses.

Sometimes they are from companies (like eBay) who genuinely feel they get too much email from customers as it is and would really rather avoid it. I don’t like that attitude, but at least they are conscious of what they are doing.

What’s even more surprising is that you get email like that from startups and other groups who are crying out for feedback and have a spiffy feedback form that’s linked from every page on their site. Or from applications that are otherwise super-usable and super-friendly.

Whoever answers your emails from your feedback form can surely also answer replies to your automated emails. Can’t they?

Data as Procedures with No Data

Further to my post on data as procedures, I’ve translated to Common Lisp the implementation of cons, car and cdr as procedures with no data, not even numbers.

This comes from lecture 5b of the Structure and Interpretation of Computer Programs video series, with the necessary (funcall) adjustments for Common Lisp.

(defun my-cons (x y)
    (lambda (m) (funcall m x y)))

(defun my-car (x)
  (funcall x (lambda (a d) a)))

(defun my-cdr (x)
  (funcall x (lambda (a d) d)))

(print (my-car (my-cons 35 47)))
(print (my-cdr (my-cons 77 22)))

Output is:

35
22

Data as Procedures

I was watching lecture 2b of the SICP lectures and it shows an implementation of cons, car and cdr in Scheme as procedures. So I thought I’d write the equivalent in Common Lisp to test my understanding:

(defun my-cons (a b)
  #'(lambda (c)
      (cond ((= c 0) a)
            ((= c 1) b))))

(defun my-car (x) (funcall x 0))

(defun my-cdr (x) (funcall x 1))

(print (my-car (my-cons 'gold 'silver)))
(print (my-cdr (my-cons 'everything 'nothing)))

Output is:

$ clisp pairs.lisp

GOLD
NOTHING

Crazy. “Constructed out of thin air”, as Abelson says.

The scheme implementation is definitely nicer. See What Is Meant by Data? Scheme has only one namespace so there’s no need for the #' and funcall clutter.

Thresholding del.icio.us/popular

I really like delicious’ “popular” page and I particularly like that it is available as an RSS feed. But keeping up with it could really take over your life.

To get around this I’ve set up a “very popular” feed. To qualify an item has to be on the delicious popular page AND have 100+ recent votes. You can subscribe at:

http://bluebones.net/very-popular/rss/

If you want a different threshold just add ?threshold=n to the end of the URL where n is the number of votes a link has to have had to qualify.

High Level Data Structures

Design-heavy languages like Java try and convince us that what we want is an object model that represents every facet of the problem. But what we really want is to remain nimble and adaptable.

A lesson learned from both scripting and functional languages is that the best high level data structure is something lightweight. Perl’s hashes and arrays or Lisp’s lists. Even Python and Ruby’s objects are skinny.

This object or something like it replaces entire object models in seconds:

class Bag: pass

From PyAmazon.

or, with a bit more syntactic sugar:

class Storage:
    def __init__(self, initial=None):
        if initial:
            for k in initial.keys(): setattr(self, k, initial[k])
    
    def __getattr__(self, k): 
        if hasattr(self.__dict__, k) or (
          k.startswith('__') and k.endswith('__')): # special keyword
            return getattr(self.__dict__, k)
        raise AttributeError, repr(k)
    
    def __repr__(self): return ''

From web.py.

Better, you can create objects from your data store quickly and easily with no boilerplate required:

/**
 * Runs a SQL statement against the specified database and returns
 * the result.
 *
 * @param   $sql    SQL to run against database.
 * @param   $host   Host to attach to.
 * @param   $usr    Username for db.
 * @param   $pwd    Password for db.
 * @param   $db     MySQL database to attach to.
 * @return          true if insert/update succeeded, false on error,
 *                  array of associative arrays (rows) if select succeeded
 *                  (may have 0 entries if no rows to fetch).
 */
function db($sql, $db="bbdb", $host="localhost", $usr="root", 
        $pwd="rootpass") {
    mysql_connect($host, $usr, $pwd);
    @mysql_select_db($db) or die("Unable to select database: " . $db);
    $res = parse_result(mysql_query($sql));
    mysql_close();
    return $res;
}

/**
 * Parses the result of a SQL query and returns an appropriate var.
 *
 * @res             Result to parse.
 * @return          true if insert/update succeeded, false on error,
 *                  array of associative arrays (rows) if select succeeded
 *                  (may have 0 entries if no rows to fetch).
 */
function parse_result($res) {
    if (is_bool($res)) {
        return $res;
    }
    $arr = array();
    $i = 0;
    while ($row = mysql_fetch_array($res)) {
        $arr[$i++] = $row;
    }
    return $arr;
}

# Usage:
#
# $myobj = db("SELECT name, company FROM employee WHERE id = 12");
# print $myobj['name'] . " works for " . $myobj['company'];

You could argue that this couples the database and the program’s representation of data too closely. But for the spring in your step you get when coding, I think it’s a worthwhile trade off. This is approximately how Rails does things, but with a layer in between for exceptions where you can override the database.

Web Design Fashion

~1991

Tim Berners-Lee
“Tim Berners-Lee”: The first page on the web, academics putting up papers in default styles.

~1994

Under Construction
“My Home Page”: Clip Art, Animated GIFs, HTML, “Under Construction”, Guest Book.

~1997

Hotwired
“Webmonkey”: Inspired by Wired.

~2000

Google
“Google”: White is the new black, minimal.

~2003

404 Not Found
“Out of work”: Fallout from the dot.com collapse.

~2006

Odeo
“Web 2.0”: new optimism, new hype, please God just one more bubble (make your own)