Re: Gwydion & threads

[Kim Barrett <address@hidden>]

At 4:30 PM -0500 3/15/02, Bruce Hoult wrote:
> > Could we live with a threads implementation that doesn't guarantee
> > atomicity?
>
>I don't see why not.  It's not as if we have to implemnent mutex
>primitives in Dylan itself -- we can use those of the OS or a library.

Since Eric Kidd hasn't piped up, I'll remind folks that he and I had a
discussion about this aspect of the Harlequin threads proposal some
time ago.  As I recall, there was some discussion of a new modifier for
variable and slot definitions that basically said "this location is
referenced by multiple threads - access must be atomic".  I don't remember
if it was in the last version of the Harlequin proposal or if it was an
idea being discussed for addition.  Eric was strongly in favor of it,
for reasons that I didn't find compelling, but in thinking through how
to rebut his arguments I came up with some additional ones that did
convince me of the need for something like that.  His arguments were
framed around the need to support the Gwydion 2-word descriptor_t
object format.

I'll see if I can dredge up that discussion.  I don't right now even
remember what forum it occurred in - info-dylan, gd-hackers, private
email...