[Haskell-cafe] Re: Chameneos

Aaron Denney wnoise at ofb.net
Wed Jan 11 21:40:01 EST 2006

On 2006-01-11, Isaac Gouy <igouy at yahoo.com> wrote:
> Ah! Useful for finding an announcement - maybe not.
> otoh the forums do allow Q&A without subscription.

And requiring subscriptions is necessary to avoid spam.  Being able to
hash things out without checking yet another bulletin board regularly is
really nice.  (Though things did get out of hand, and rather heated on
the shootout list).

> In this case, there was no announcement


> http://shootout.alioth.debian.org/gp4/benchmark.php?test=chameneos&lang=ghc&id=2#about

Yeah, I found that, eventually.  The thing is, it'd be nice to have
clarifications to the benchmark spec caused by such creative attempts on
the benchmark page somewhere.  A seperate "benchmark clarifications"
forum might also work, but be less convenient.

Also, even though I brought up the concern, that the manager thread
might not be considered acceptable, I personally think the seperate
threads should be accepted.

I see the manager thread as just an implementation technique for
symmetrically synchronizing the other threads.  It's not a primitive,
like a built in "barrier n" function would be.  Of course, neither is the
"lock this and see what else is there" technique, which I see as being
at the same level.

I guess this goes to my dissatisfaction with the "same *exact* way"
evolution of the shootout, which is now a lost battle.  Is the task
"synchronize these four threads, two at a time", or "synchronize these
four threads, two at a time using the most primitive facilities
available in your language that most closely resemble the orthodox
parallelism paradigm".  I don't, for example, see a good way to do this
synchronization in Occam without using a manager thread.

In fact, the linked paper describes the meeting place as being "a
server" which has the connotation to me of being at least another
thread, though none of the example programs use a seperate thread.

Anyways, your shootout, your hard work, your rules, but having rulings
on what's acceptable be easier to find would be nice.

> And that was already commented on by folk on this mailing-list.

Right, but no pointer to it, and it wasn't immediately obvious to me.

Aaron Denney

More information about the Haskell-Cafe mailing list