Re: Bug on HPs

[Roger Burkhart]

In response to Rick Riolo's question on the HP probe bug:

> Hi Manor (&others),
> Any progress on this bug?  
>   - r

We're very concerned about this, not only because it's serious in
its own right ...

> It is a serious bug for teaching-Swarm
> (amoung other things).   I was hoping to at least
> demo swarm in the week of 5-10 August to a summer
> school class, so if we could get this fixed 
> soon that would be great.  if not, I understand,
> I've had to track down elusive bugs myself now
> and again (not bugs I created, of course ;-)

... but because Manor has isolated the bug down to gcc's _builtin_apply
function as used by tclobjc, so it's out of our code altogether.

We've had enough difficulties with gcc's _builtin_apply extension to C
(which provides an essential capability for both us and the tclobjc
library to dynamically construct function calls) that we've decided
we'll have to attack it full-on and build our own generic interface to
_builtin_apply.  We'll use this interface to exercise _builtin_apply as
thoroughly as possible with different argument and return types on every
machine type.  Then we'll be able to reproduce any such bugs outside of
Swarm, and then either find our work-arounds or get them fixed directly
in gcc or both.  Our concern is not only with HP where things still mostly
work, but with IBM RS/6000 where none of this has ever worked (which is
why we don't currently run there).

I'm starting to work on this myself but with other travel by both
Manor and me we aren't likely to get a fix implemented in time for
your class.  But it is a serious bug right at the heart of our generic
probe machinery.  We'll be working hard to get it fixed whereever
we can, and we hope that'll eventually be directly in gcc where we
currently think the fix is needed.

Roger Burkhart