[slime-devel] Re: custom tweaks for object system

[David O'Toole]

Hello Helmut, 

On Wed, 2008-08-20 at 14:43 +0200, Helmut Eller wrote:

> I would recommend to use short operator names instead of reader macros.

Thanks for your comments and suggestions everyone! I will try things out
and see if I can get a basic lambda list / docstring thing going.

I debated the syntax issue for a long time, with a few online friends,
back before this object system was ported to Common Lisp. (It was
originally in elisp with identical syntax.) And originally the project
did use @ and so on. But I moved to the [method-name object foo bar
blaz] for several reasons. As a lisp programmer my eye is naturally
drawn at once to the leftmost token in a list so that I can start
getting hold of the sentence. Seeing ! and @ all the time at that moment
would seem to dull the habit and I found it somewhat annoying. Instead I
put the operation in the leftmost spot, just like function calls. This
is consistent with my reading and writing style. 

(@ :slot) is still too verbose and uses a sublist. CLOS with-slots is
better than this. using < and > may be jarring at first but I find that
when slot references are visually distinct from ordinary variable
references, one thinks about optimizing slot access, and this object
system is designed for games and other performant interactive programs.

Besides. I've already written about 3000 lines of OO-style code using
this custom object system (entitled "CLON"). So it would be pretty hard
to change now anyway. But I'm pretty comfortable with it... matching
things like (([([(([ is trivial with Emacs' various delimiter
highlighting and navigational modes, show-paren for example, and with
these configs in Emacs the syntax will work properly in Lisp mode: 

;;; Font-locking

;; Put this in your emacs initialization file to get the highlighting:
;; (add-hook 'emacs-lisp-mode-hook #'clon-do-font-lock)

(defvar clon-font-lock-keywords
  `((,(rx (sequence "(" (group "define-method")
		   (one-or-more space)
		   (group (one-or-more (not (any space))))
		   (one-or-more space)
		   (group (one-or-more (not (any space))))))
     (1 font-lock-keyword-face)
     (2 font-lock-function-name-face) ;; this still doesn't work
				      ;; properly.
     (3 font-lock-type-face))
    (,(rx (sequence "(" (group "define-prototype")
		   (one-or-more space)
		   (group (one-or-more (not (any space))))))
      (1 font-lock-keyword-face)
      (2 font-lock-type-face))
;    ("\\<\\(\<[^<>]*\>\\)\\>" (1 font-lock-preprocessor-face))
    ("(.*\\(\>\>\\>\\)" (1 font-lock-type-face))))

(defun clon-do-font-lock ()
  (interactive)
  "Highlight the keywords used in prototype-oriented programming."
  (font-lock-add-keywords nil clon-font-lock-keywords))

;;; Bracket matching in Common Lisp mode

;; Matching square brackets are turned off by default in Common Lisp
;; mode. This will turn them back on, which makes it work with
;; paredit, show-paren, and highlight-parentheses modes.

(modify-syntax-entry ?\[ "(]" lisp-mode-syntax-table)
(modify-syntax-entry ?\] ")[" lisp-mode-syntax-table)


> There is some code in contrib/swank-arglists.lisp which understands
> things like (make-instance 'class ...).  That could probably be reused
> by adding methods for @ and !.  
> 
> However, in the absence of static type information, it's almost
> impossible to deduce the arglist of those methods.  It may work for
> methods with a unique name, but if different classes have methods with
> the same names (ala "add") it's hard to know which class should be used.
> Java IDEs can do this easily due to the static nature of the language,
> but it requires a lot of work for dynamic languages, like Smalltalk or
> Python.
> 
> Helmut.
> 
> _______________________________________________
> slime-devel site list
> slime-devel at common-lisp.net
> http://common-lisp.net/mailman/listinfo/slime-devel