Re: [Sbcl-devel] *repack-blocks*

[bryan o'c. <bry...@lu...>]

This indeed fixes the openmcl-as-build-host breakage.

can this be committed?

thanks,

     ..bryan


On Sep 13, 2005, at 3:34 AM, Rob MacLachlan wrote:

> Yeah, it really does add any and every key during maphash. The  
> simplest fix is to make a new *repack-blocks* on each trip around  
> the loop which is in effect during the iteration over the previous  
> one.  (all you really have to do is rebind *repack-blocks* to NIL,  
> then see if a new one gets created.  This also eliminates the  
> (zerop (hash-table-count test)).  This would also eliminate the  
> need to do the remhash at all. *repack-blocks* is basically a hash  
> set.
>
> Here's a patch of what I mean.  be aware that this is against an  
> old version of cmucl, not current sbcl sources.
>
> --- opack.lisp    2005-09-13 06:28:24.986953907 -0400
> +++ pack.lisp    2005-09-13 06:30:51.722579315 -0400
> @@ -1678,14 +1678,14 @@
>           (do-ir2-blocks (block component)
>             (emit-saves block)
>             (pack-load-tns block))))
> -      (when *repack-blocks*
> -    (loop
> -      (when (zerop (hash-table-count *repack-blocks*)) (return))
> +      (loop
> +        (unless *repack-blocks* (return))
> +    (let ((orpb *repack-blocks*))
> +      (setq *repack-blocks* nil)
>        (maphash #'(lambda (block v)
>                 (declare (ignore v))
> -               (remhash block *repack-blocks*)
>                 (event repack-block)
>                 (pack-load-tns block))
> -           *repack-blocks*)))))
> +           orpb)))))
>
>    (undefined-value))
>
>
>   Rob