[plt-scheme] Feature: Expired-continuation response for send/forward, send/finish

This is a good idea and worthy of consideration.

Why don't you alter the web server (it's straightforward), play with 
the code for a while, and when you're happy mail it to Greg Pettyjohn 
and me as a change request.

-- Matthias


On Nov 21, 2004, at 2:06 PM, Jordan Johnson wrote:

>  For list-related administrative tasks:
>  http://list.cs.brown.edu/mailman/listinfo/plt-scheme
>
> [Warning: half-baked idea follows.  May need further baking.]
>
> One thing I find I keep wanting to do is install a 
> failure-continuation for when users accidentally reload a form or use 
> the back button; it's needlessly frustrating for them to see "The 
> transaction referred to...has expired", and I've mostly abandoned use 
> of send/forward for that reason, though I like the concept of it.
>
> So, I was thinking, perhaps send/forward could be modified somehow, 
> such that when called with one argument it maintains its current 
> behavior, but if called with a second procedure, it points all the 
> session's defunct k-urls to that procedure as a continuation, rather 
> than purging them?
>
> It seems to me that such a continuation might expect a request; or 
> perhaps, simply, that it could receive the same k-url generated by 
> send/forward for the last valid page, since the only sensible thing to 
> do with a request for an expired k-url would be to ignore its 
> bindings.
>
> Following is an example of one way it might be used, but first, the 
> caveat:  I'm not pleased with the way that this example would lead to 
> semi-CPS'd code.  I know there's got to be a better way to do it.  
> Still, I feel a need for a construct that addresses this, and I hope 
> my description and example get the point across.
>
> ;;;;;
> ;; not defined here:
> ;; make-question-widget: question-struct -> xexpr
> ;; correct?: question-struct answer -> boolean
> ;; request->answer unpacks an answer from the bindings in a request.
>
> ;; question-struct (req -> void) -> answer
> (define (get-user-answer question bad-user-k)
>   (request->answer
>     (send/forward
>       (lambda (k-url)
>        `(html
>           (body
>             (form ((action ,k-url) (method "POST"))
>               ,(make-question-widget question)
>               (input ((type "submit") (name "submit") (value 
> "submit")))))))
>       bad-user-k)))
>
> ;; (listof question-struct) -> void
> (define (quiz-servlet loq)
>   (send-start-page)
>   (let loop ((questions-left loq)
>              (number-answered 0)
>              (incorrect '()))
>     (if (null? loq)
>       (send-end-page answered incorrect)
>       (let* ((question (car loq))
>              (answer (get-user-answer
>                        (car loq)
>                        (lambda (req)    ;; here's the new user-error 
> continuation
>                          (send/suspend  ;; (which would have to 
> escape, yes)
>                            (lambda (k-url)
>                              `(html
>                                 (body
>                                   (p "Sorry, you may not go back, only 
> "
>                                      (a ((href ,k-url)) "forward") 
> ".")))))
>                          (loop loq number-answered incorrect)))))
>        (if (correct? question answer)
>          (loop (cdr loq) (add1 number-answered) incorrect)
>          (loop (cdr loq)
>                (add1 number-answered)
>                (cons question incorrect)))))))
>
> ;;;;;
>
> Is there some merit in this?  Has it been discussed already?  Or is 
> there an existing mechanism I could be using instead?
>
> jmj
>
> : Jordan Johnson - jorjohns @ cs . indiana . edu
> : Thrive defensively.
>