Re: Priority revisited: a new primitive

[A E Lawrence <adrian.lawrence@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx>]

M_Boosten wrote:
> 
> > > > The fact that one talks about (1) PRI PAR, (2) PAR, (3) PRI
> > > ALT, (4) ALT,
> > > > and (5) PAR PRI tells me: all this is TOO COMPLEX!  Something is
> > > > FUNDAMENTALLY WRONG!
> > >
> >
> > I also think that the learning cycle is not too complex. Students pick it up
> > easily because this concept is relatively simple and clean. For some is
> > understanding and learning the look-and-feel of the PAR, ALT, PRI PAR, PRI
> > ALT, SEQ, and PAR PRI all together confusing and therefore complex . perhaps
> > unnecessarily complex. On one hand, these keywords capture most fundamental
> > artefacts of the real-time and concurrent world . nevertheless; this is a
> > confusing world. On the other hand, we need to leave out what we don't need.
> 
> You can build complex CSP processes, that depend on the
> _priorities_ for correctness.  I (feel) that this should
> not be encouraged.
> 
> I think programs should be correct, independent of the
> priorities.  Priorities should only affect the
> performance, the order, but not the correctness of the
> order.

Hey! I have been very very careful about maintaining the compositional
semantics of all of the PRI versions of the constructors :-)

Until Barry or someone can tell us how to specify and implement high
level requirements on the correctness of deadline driven programs in the
presence of limited resources, I do not know how you can have a correct
program that does not depend on priority. I realize that you said
"correctness of order" above, but we need the real time stuff as well.
So does Philips?

And even if we do get some higher level description, a compiler will
presumably have to compile this into prioritized processes, so we still
need to reason precisely about priority.

We all agree that it is hard. We all wish that it wasn't. As far as we
know, there is no better concept. We are trying to find the simplest,
most powerful and elegant tools to help us. So far the best we have are
{ALT, PAR} PRI.
Yes?

Adrian
-- 
Dr A E Lawrence (from home)