[Openmcl-devel] kmrcl broken with openmcl-cvs

[Gary Byers]

On Fri, 20 May 2005, Valentino Volonghi aka Dialtone wrote:

> I'm aware that latest cvs is not supposed to be always working (or
> compiling) so if what I say is
> already known, just ignore it.
>
> I'm running  0.14.3-050518a compiled with latest cvs without any problems
> (also hemlock compiled without problems), and this is why I'm sending this,
> if it failed to compile anything I wouldn't have bothered.
>
> I've been working with cl-base64 lately and it needs kmrcl just for
> testing purposes.
>
> unfortunately there is an error while compiling sockets.lisp in kmrcl:
>> Error in process listener(1): Unknown vinsn: CCL::NATURAL-SHIFT-RIGHT
>> While executing: CCL::NEED-VINSN-TEMPLATE
>
> grepping ccl source shows that natural-shift-right is there, but
> probably not exported (or something like that).

The left hand doesn't know what the right hand is doing.  In 
"ccl:compiler;PPC;PPC32;ppc32-vinsns.lisp", the form

(define-ppc32-vinsn u32-shift-right (...)
   ...)

should be renamed

(define-ppc32-vinsn natural-shift-right (...)
   ...)

(and similarly for U32-SHIFT-LEFT, and possibly a couple of other things
that I missed ...).


>
> I hope I've been useful.

Thanks.  I might be able to make it easier to catch some things like
this earlier, but some of the bugs that I'm (probably) introducing
are only going to be detected if the code is exercised (and I usually
only exercise what happens during the build process.)

It's not surprising that bugs like this this are getting introduced
(it'd be more surprisng if they weren't), but it's very helpful if
they're reported.