Archive for February, 2012


You’re only using PHP because you don’t have Guile and (sxml simple)

In functional programming,Scheme,Web development on February 19, 2012 by gcbenison Tagged: , , ,

OK, maybe you would still use PHP even if you did have a Scheme implementation and the (sxml simple) module.  But it is true that just about any web application written in PHP could also be written in Scheme or another Lisp dialect.   And yes you can get it hosted, more easily than you might think; I’ve done it (see this example, described at the end) with just the basic services at my hosting provider.  There are even complete Lisp web frameworks available.

So web developers do use Scheme, even if it is far from the most popular web language.  But web development in Lisp is not just possible: it is also a good idea, especially as your project evolves past “a page with some dynamic content” to “a full application that runs over the web”.  HTML can be understood in two ways:

  1. As plain text with a few formatting tags thrown in
  2. As a tree of nodes, where nodes have names and optionally, attributes and values

These two views lead to two different approaches to dynamic web programming:

  1. Treat the web page as a plain text document, with little embedded snippets of code that expand like macros when the page is displayed
  2. Represent the entire web page as a tree using a native data structure, which is then rendered as HTML at page display time

Both HTML and many web programming languages were designed with view (1) in mind because it can be very simple.  You don’t need to be a programmer, or to understand what the phrase “tree of nodes” means, in order to understand what this is supposed to do:

Today is: <?php echo date('Y-m-d'); ?> and it is a <i>good</i> day

But the problem with this approach is that it doesn’t scale well to more complicated uses.   Take, for instance, this production PHP snippet from the WordPress code base (pasted here as an image because of the difficulty of embedding actual literal PHP here):

That, despite being an example of well written PHP, is not terribly intuitive even to a programmer.   I really dislike the  intermingling of two distinct syntaxes (that of HTML and PHP).   It forces you to switch back and forth between the two languages when mentally parsing such code.  It leads to odd orderings of delimiters –  <?php { ?> etc.   It’s what leads to PHP being called things like “bloated, messy, and counter-intuitive“.

Such contortions do not appear in web applications that treat the whole page as a native data structure which the system then renders as HTML.   Since all of the actual HTML is generated by the renderer, the application developer need not even know any HTML syntax in order to program this way!  Scheme is well-suited to representing such a tree of nodes as a native data structure, and there is a standard convention for doing so known as SXML.  One of the things I like best about Scheme for web development is the clear equivalence of a live Scheme object, its serialized representation as SXML, and its serialized representation as HTML – all of which can be interconverted with a single command.

Interconversion of HTML and its equivalent Scheme representation

This means we can take the result of any Scheme computation, manipulate it into SXML, and render it as HTML.  A working implementation of the following example can be found here.  Consider a function top-word-usage which splits a string into words and returns the number of times each word appears, sorted with the most common words first:

> (top-word-usage "Hello, world.  Yes- hello, world, hello!" 5)
$1 = (("hello" . 3) ("world" . 2) ("yes" . 1))

Now define a function to format each entry as a string, and another function to wrap any content as an SXML “list element”:

(define (format-usage usage)
(format #f "~a -- ~a occurrences" (car usage)(cdr usage)))

(define (as-li content)
`(li ,content))

> (map as-li (map format-usage $1))

$2 = ((li "hello -- 3 occurrences")
      (li "world -- 2 occurrences")
      (li "yes -- 1 occurrences"))

Around this core functionality, we can use SXML to build a template for a complete web application:

(display "Content-type: text/html\n\n")
   (head (title "wc demo"))
    (h1 (a (@ (href ,(car (command-line)))) "Word count demo"))
    ,(if input
	 `(p "Top words of at least " ,min-word-length " characters:"
	      ,(map as-li
		    (map format-usage (top-word-usage input 5)))))
	    '(p "Enter some text below, to count word usage:")))
    (form (@ (method "post")
	     (name "input-form")
	     (action ,(car (command-line))))
	  (textarea (@ (name "input-content")
		       (rows 20)
		       (cols 60))
		    ,(or input ""))
	  (p "Minimum word length: "
	     (input (@ (type "text")
		       (name "min-length"))))
	  (input (@ (type "submit")
		    (name "input-submit")
		    (value "Determine word usage")))))))

The full application is less than 100 lines long and can be found here. Try pasting in a favorite piece of writing and seeing what happens – it’s actually kind of fun!