Paul Graham: Lisp in Web-Based Applications (2001)

Libraries tend to be generic, so you can often do better for your specific problem if you are willing to pay the opportunity cost.

Sometimes it is quicker to write the exact thing you need rather than struggling with something that wants you to think it is the solution to your problem.

Another case in point was the Aha! moment someone described when realized that the Lisp Postgress interface was not going against the library but against the wire protocol itself. Just as easy.

I think this argument in some ways hinges on how far from the tree you're falling. If you're doing essentially a variation on a common theme, you'll be the fastest by picking a language where the bulk of the work is already done for you in a variety of reusable modules. It'd have to be a really poor language for its badness to outweigh having most of the work done for you up front.

If, however, you're starting from scratch on something relatively novel, e.g. where Viaweb was in 1996 (and not so much in 2001), then it's true, you should write in the language that fits the bill rather than following the course of momentum.

One proviso I would add there though is that Lisp is in the position of being fairly mature and non-traditional, unlike many flavor-of-the-weeks. There are many pains to be found in using tools which haven't stabilized yet, even if they do have a gee-whiz-bang factor.

As to your postscript, Google does also extensively use C++; I think that's still fairly common in high-performance stuff at the lower levels.

  (define (whitepage . content)
    `(html (body ,@content)))

  (define (para)
    `(p))

  (define (link prose url)
    `(a ((href ,url)) ,prose))

  (define (prbold text)
    `(strong ,text))

  (whitepage "Hello world!" (para) (link "Click here" "http://news.ycombinator.com")
             "for" (prbold "more stuff"))

  (html
   (body
    "Hello world!"
    (p)
    (a ((href "http://news.ycombinator.com")) "Click here")
    "for"
    (strong "more stuff")))

  (mac uvar (u k) `((profs* ,u) ',k))
  (def user-age (u) (hours-since (uvar u created)))

(Did you mean uvar instead of upvar?)

Not necessarily... even using functions, you can create a 'lazy node' in the page graph to delay completion of calculation until the final renderer actually starts rendering the node.

I guess the only advantage of using macros is that you can make page generation code without consing relatively easily. I'm not sure how necessary it is for web servers, but it may matter for extremely high-traffic sites.