Re: [Sbcl-devel] Regression in stack allocation of dynamic-extent vectors

[Nikodemus S. <nik...@ra...>]

> On Tue, May 6, 2008 at 1:47 PM, Alexey Dejneka <ade...@to...> wrote:
>
>  >  > +(defun lvar-real-dx-lvar (lvar)
>  >  > +  ;; FIXME/KLUDGE: TRULY-THE introduces a cast which inhibits
>  >  > +  ;; DX allocation -- look through casts. Punt on hairier cases
>  >  > +  ;; involving multiple uses, etc, for now.
>  >  > +  (let ((uses (lvar-uses lvar)))
>  >  > +    (if (cast-p uses)
>  >  > +        (lvar-real-dx-lvar (cast-value uses))
>  >  > +        lvar)))
>  >
>  >  It seems to be the same as PRINCIPAL-LVAR-USE.

Almost. Unless I misunderstand something, it seems to me that If
LVAR-REAL-DX-LVAR was instead

(defun principal-lvar (lvar)
  (labels ((pl (lvar)
             (let ((use (lvar-uses lvar)))
               (if (cast-p use)
                   (pl (cast-value use))
                   lvar))))
    (pl lvar)))

then PRINCIPAL-LVAR-USE would be (LVAR-USES (PRINCIPAL-LVAR X)).

Cheers,

  -- Nikodemus