bug#748: Elisp: lexical-let and cyclic structures

bug#748: Elisp: lexical-let and cyclic structures

[Michael Heerdegen]

When using the reader constructs `#N=' and `#N#' for cyclic
structures, lexical-let sometimes produces errors which don't occur
with let.

Example: Eval the following:

(defun f (start)
  (lexical-let (start start)         ;; return a closure
    #1=(lambda (x) (if (= x start) x
		     (+ x (#1# (1- x)))))))

Then evaluating f ends with an error:

   (f 3)

    "Variable binding depth exceeds max-specpdl-size"


The problem also occurs without lambdas. Examples:

   (let         () '#1=(#1#))               ;; ==> (#0)
   (lexical-let () '#1=(#1#))               ;;  Error
   (let ((x '#1=(#1#))) (lexical-let () x)) ;; ==> (#0)
   `#1=(#1#)                                ;;  Error

bug#748: Elisp: lexical-let and cyclic structures

[Stefan Monnier]

tag +748 wontfix
thanks

> When using the reader constructs `#N=' and `#N#' for cyclic
> structures, lexical-let sometimes produces errors which don't occur
> with let.

Most programming languages do not accept infinite programs.  Elisp is
no exception.  The fact that you can build cyclic abstract syntax trees
and that they sometimes get evaluated correctly is just an accident.

I.e. if it hurts, don't do it.

> (defun f (start)
>   (lexical-let (start start)         ;; return a closure
>     #1=(lambda (x) (if (= x start) x
> 		     (+ x (#1# (1- x)))))))

For this case, I'd recommend to make the recursion explicit, e.g.:

   (defun f (start)
     (lexical-let ((start start))         ;; return a closure
       (labels ((loop (x) (if (= x start) x
   		           (+ x (loop (1- x))))))
         loop)))


-- Stefan

bug#748: Elisp: lexical-let and cyclic structures

[Glenn Morris]

I don't see a need to keep open this particular report, which was marked
"wontfix" some time ago.

Stefan Monnier wrote:

> Most programming languages do not accept infinite programs.  Elisp is
> no exception.  The fact that you can build cyclic abstract syntax trees
> and that they sometimes get evaluated correctly is just an accident.
>
> I.e. if it hurts, don't do it.