[PPL-devel] New deterministic timeout facilities of the PPL and PWL

[Roberto Bagnara]

Sebastian Pop wrote:
> We will play with PWL before integrating it in GCC.

OK, but please do tell us well in advance if you need an official PPL
release with that feature in.

> I was thinking about starting the counter at the beginning of Graphite
> when we initialize PPL and reseting it after code generation is done.

This cannot be done.  In fact, the timeout can leave the object
upon which the PPL was working in an inconsistent state.
The only safe thing that you can do with a PPL object that was
subject to an operation that timed out is to delete it.
This means that you must check the result of every PPL operation
that may result in a timeout;  in case of timeout, you must
delete the affected PPL object(s) and implement your "plan B".
Cheers,

    Roberto

P.S. We have uploaded a new snapshot, which offers a better implementation
      of deterministic timeouts.

P.S. The documentation of the PPL currently does not say which operations
      can timeout and which of the arguments is affected in case this
      happens.  We will remedy this limitation as soon as possible.

-- 
Prof. Roberto Bagnara
Computer Science Group
Department of Mathematics, University of Parma, Italy
http://www.cs.unipr.it/~bagnara/
mailto:bagnara at cs.unipr.it