Re: Bug in http module of guile-www

[Thien-Thi Nguyen]

   From: Robert Marlow <address@hidden>
   Date: Wed, 14 Jul 2004 15:38:17 +0800

   I happened accross a bug in the HTTP module of guile-www which seemed to
   trigger when I visited a page which gave no headers and just whitespace
   in the body. In such a case the variable "second" of parse-status-line
   gets bound to #f which messes up the later make-shared-substrings which
   use it.

in the status line (which has the form: VERSION CODE TEXT), i believe
TEXT is optional, but i don't have the http spec (RFC 2616) handy to
confirm this.  if TEXT is actually optional, that indicates a bug in
http.scm.  if TEXT is required, that indicates a bug in the server's
http implementation.

note that the status line cannot be omitted altogether; the most minimal
valid HTTP response a server can send is STATUS-LINE, CRLF, CRLF.  you
may wish to check the server response for conformance using a command
like "w3m -dump_head" or similar.

   [patch]

unfortunately the patch is incorrect because it allows the possibility
to derive #f for CODE, which is illegal (CODE must be a 3-digit sequence
like "404" or "200").  assuming TEXT is indeed optional, i have changed
`parse-status-line' (in cvs) to handle a missing TEXT component by
returning the empty string.  the proc now reads as follows:

(define (parse-status-line statline)
  ;; Handle:     VERSION CODE
  ;; as well as: VERSION CODE TEXT
  ;; For the former, use the null string for TEXT.
  (let* ((first (string-index statline #\space))
         (second (string-index statline #\space (1+ first))))
    (list (make-shared-substring statline 0 first)
          (make-shared-substring statline (1+ first)
                                 (or second (string-length statline)))
          (if second
              (make-shared-substring statline (1+ second))
              ""))))

and will appear in guile-www 2.6 (to be released shortly).

thi