Re: Blocking I/O and Ravenscar

[Surfer <surfer@no.spam.net>]

On 24 Sep 2007 11:03:42 +0200, Ole-Hjalmar Kristensen
<ole-hjalmar.kristensen@substitute_employer_here.com> wrote:

>Hmmm. Why not use the rendezvous for implementing CSP communication?
>It would seem to be a much closer match than protected objects.
>

Its because the authors are considering the Ravenscar profile which
requires zero entries per task. That means that no Ada rendezvous are
possible.

However, implementation of CSP communication within the Ravenscar
profile, allows CSP rendezvous to be used in place of Ada rendezvous.

According to the paper[1]:

"Ada has proven successful in the area of high integrity sys-
tems development, but its tasking model is hard to reason about. Thus,
Ravenscar has been defined as a restricted subset of the Ada tasking
model, which meets the requirements of producing analysable and de-
terministic code."

The authors also write:

"This allows us to transform the data-oriented asynchronous tasking
model of Ravenscar into the cleaner, and better understood,
message-passing synchronous model of CSP."

So they seem to consider the CSP tasking model to be easier to reason
about than those of either Ada or Ravenscar.


[1]  Communicating Ada Tasks (2003)  
http://citeseer.ist.psu.edu/577581.html

[2] Extending Ravenscar with CSP Channels
http://www.springerlink.com/index/j7h8rr665r0x20n9.pdf