Mixing reproducible and non-reproducible code
As usual, trying to be specific about reproducibility raises various
tricky issues. I'm referring now to clause 1.8 item (8) in John's
20121130Clauses1-2forVote.pdf, currently under discussion.
In my earlier comments I wrote:
> 1.8 item (8): This needs to be qualified. It might be necessary to
> invoke a "reproducible transformation unit" though a wrapping
> interface ("glue module") to incorporate it in a non-reproducible
> environment, as the level 3 representations might differ.
There may be more at stake than representation issues. Reproducible mode
may affect global properties, such as forcing single-threaded execution.
So what do we expect to happen when modes are mixed?
(1) The interface module could force single-threaded execution while the
reproducible-mode component is running. Is that generally feasible?
(2) The reproducibility properties of the reproducible-mode component
could be voided, as basic assumptions of that mode don't pertain.
I understand the motivations behind requirement (8): one would like to
be able to debug components in reproducible mode, and then incorporate
the debugged code in a larger program. But given that reproducible mode
typically requires a number of programming rules to be obeyed, validation
in reproducible mode becomes meaningless unless those restrictions are
maintained. That may be ok if the effect is relatively small when not
actually running in reproducible mode, but if it involves re-partitioning
for single-threaded execution this may not be so.
Michel.
---Sent: 2012-12-02 00:46:30 UTC