Re: Make register easier to hook

[Daniel Colascione <dan.colascione@gmail.com>]

On 3/28/2011 5:52 PM, Leo wrote:
 > On 2011-03-29 01:37 +0800, Daniel Colascione wrote:
 >> Thanks for doing this work.  An extensible register system is also on my
 >> wishlist.  Why did you decide to use the list for the register structure
 >> instead of the default vector representation?
 >
 > Probably because it looks nicer inside register-alist and there is
 > little performance to lose.
 >
 >> Also, have you considered eliminating the `info' slot? Users can
 >> construct closures over any necessary variables and assign these
 >> closures to the register structure's remaining function slots. After
 >> all, we're getting lexbind, and this feature will make constructing
 >> closures both easy and safe.
 >
 > Now I do. But I fail to see how that makes things more convenient. The
 > closures must be created inside the command/function that creates the
 > register, right? So one won't be able to define functions with defun for
 > the slots. Did I miss something?

Well, when we get lexbind, we'll be able to write something like this
safely, cleanly, and efficiency:

(defun foo-register-jump (some-datum)
   "Implement the jump operation for registers with foo values."
   ...
)

(defun foo-register-assign (register some-datum)
   "Assign a foo register value to some register"

   (register-set register
     (register-make
       :jump-func (lambda () (foo-register-jump some-datum))))


Today, we'd have to write this instead for the same effect:

(defun foo-register-assign (register some-datum)
   "Assign a foo register value to some register"
   (lexical-let ((some-datum some-datum)) ; <----- HACK -------<
     (register-set register
       (register-make
         :jump-func (lambda () (foo-register-jump some-datum)))))

I still think the version with lexical-let is cleaner than using a 
separate data value, especially because you can easily close over 
multiple variables without having to build some aggregate data structure 
to hold them.

By the way: I'd also suggest calling the structure register-value or 
somesuch. It's confusing to have the same word refer to the *identifier* 
for the register (a, b, c, ...) and for the thing *inside* a register (a 
window configuration, a marker, a string, etc.)