Over-engineered code is over-engineered

Lars is one of the few mononymous people in the Emacs universe. His greatest contribution to Emacs would be Gnus, a 100k SLOC heavy NNTP reader people commonly customize to read Email. I found the following after taking a quick look at gnus.el:

(defun gnus-interactive (string &optional params)
  "Return a list that can be fed to `interactive'.
See `interactive' for full documentation.

Adds the following specs:

y -- The current symbolic prefix.
Y -- A list of the current symbolic prefix(es).
A -- Article number.
H -- Article header.
g -- Group name."
  (let ((i 0)
        out c prompt)
    (while (< i (length string))
      (string-match ".\\([^\n]*\\)\n?" string i)
      (setq c (aref string i))
      (when (match-end 1)
        (setq prompt (match-string 1 string)))
      (setq i (match-end 0))
      ;; We basically emulate just about everything that
      ;; `interactive' does, but add the specs listed above.
      (push
       (cond
        ((= c ?a)
         (completing-read prompt obarray 'fboundp t))
        ((= c ?b)
         (read-buffer prompt (current-buffer) t))
        ((= c ?B)
         (read-buffer prompt (other-buffer (current-buffer))))
        ((= c ?c)
         (read-char))
        ((= c ?C)
         (completing-read prompt obarray 'commandp t))
        ((= c ?d)
         (point))
        ((= c ?D)
         (read-directory-name prompt nil default-directory 'lambda))
        ((= c ?f)
         (read-file-name prompt nil nil 'lambda))
        ((= c ?F)
         (read-file-name prompt))
        ((= c ?k)
         (read-key-sequence prompt))
        ((= c ?K)
         (error "Not implemented spec"))
        ((= c ?e)
         (error "Not implemented spec"))
        ((= c ?m)
         (mark))
        ((= c ?N)
         (error "Not implemented spec"))
        ((= c ?n)
         (string-to-number (read-from-minibuffer prompt)))
        ((= c ?p)
         (prefix-numeric-value current-prefix-arg))
        ((= c ?P)
         current-prefix-arg)
        ((= c ?r)
         'gnus-prefix-nil)
        ((= c ?s)
         (read-string prompt))
        ((= c ?S)
         (intern (read-string prompt)))
        ((= c ?v)
         (read-variable prompt))
        ((= c ?x)
         (read-minibuffer prompt))
        ((= c ?x)
         (eval-minibuffer prompt))
        ;; And here the new specs come.
        ((= c ?y)
         gnus-current-prefix-symbol)
        ((= c ?Y)
         gnus-current-prefix-symbols)
        ((= c ?g)
         (gnus-group-group-name))
        ((= c ?A)
         (gnus-summary-skip-intangible)
         (or (get-text-property (point) 'gnus-number)
             (gnus-summary-last-subject)))
        ((= c ?H)
         (gnus-data-header (gnus-data-find (gnus-summary-article-number))))
        (t
         (error "Non-implemented spec")))
       out)
      (cond
       ((= c ?r)
        (push (if (< (point) (mark)) (point) (mark)) out)
        (push (if (> (point) (mark)) (point) (mark)) out))))
    (setq out (delq 'gnus-prefix-nil out))
    (nreverse out)))

First of all, a bit of background. Emacs Lisp knows functions, like many other programming languages. However, commands in Emacs often allow arguments to be supplied, be it by a prefix argument or reading in extra data. This is achieved by adding an interactive specification after its docstring; command-execute or call-interactively will then fetch its interactive specification via interactive-form and execute the respective C code to fill in the unspecified function arguments. To allow for non-standard ways of providing arguments, a form returning a list of the arguments can be used instead of a string.

So, what’s Gnus doing? Instead of using said alternative form, it reinvents the majority of call-interactively, leaves out a few supported codes, adds a few of its own, alters the behaviour of the r code, then spits out a list.

Is this code actually used? Well, if ag is to be trusted, there’s exactly one variant of gnus-interactively sprinkled all over its code, (interactive (gnus-interactive "P\ny")). This could be replaced by a function returning a list containing current-prefix-arg and whatever the relevant code of gnus-symbolic-argument for its symbolic prefixes feature is, but would that have been as fun as this nonsensical exercise in NIH? Surely not!