Yes, that comment is much better.

By "extra arguments" I meant the second example you give in the
comment, where there are two arbitrary arguments instead of one lambda
argument.  The alternative approach I was suggesting is to restrict
the optimization to the case where there's a single lambda argument.
In that case the argument ordering is not an issue and neither are
side effects.

On Mon, Sep 21, 2009 at 9:25 AM, Taylor R Campbell <[email protected]> wrote:
>   Date: Sun, 20 Sep 2009 12:57:43 -0700
>   From: Chris Hanson <[email protected]>
>
>   I think it would be useful to amend the comments in the patch to show
>   the additional arguments.  Without those arguments, there's no
>   justification for the conservative treatment.
>
> What are the additional arguments here?  I've attached an amended
> patch with more comments; does this explain it better?
>
>   Alternatively, you could restrict the optimization to one without
>   addtional arguments, and be less conservative.
>
> I think there are few programs that more aggressive transformations
> would improve beyond what the conservative transformation improves,
> and exceedingly few that any middle ground between the two
> transformations I implemented would improve beyond the the
> conservative transformation.
>
> By the way, I don't know anything about how LIAR exploits programs'
> ambivalence about order of evaluation.  Maybe the more aggressive
> transformation improves the code better than LIAR would anyway, in
> which case I'd be happy to commit that instead.
>


_______________________________________________
MIT-Scheme-devel mailing list
[email protected]
http://lists.gnu.org/mailman/listinfo/mit-scheme-devel

Reply via email to