[coyotos-dev] GC as a helper process

[Jonathan Adams]

On Tue, 2007-01-09 at 23:37 +0100, Pierre THIERRY wrote:
> Scribit Jonathan S. Shapiro dies 09/01/2007 hora 15:49:
> > I'm not an expert on GC. My suspicion is that GC is best done as an
> > in-process activity. The compiler needs to help to get efficient write
> > barriers, and that has a pretty tight coupling to the runtime system.
> 
> In fact, I was not really thinking about a generic GC, but about an
> implementation-specific one. That is, Mono would have it's own GC, SBCL
> it's own, and so on.
> 
> A generic GC would porbably not be worth the effort (and the
> complication, and the bugs coming with it).

I'm not sure having GC be in a separate process would be all that
helpful, unless it was effectively just a "GC thread", sharing its
address space with the process it GCs.  Otherwise, you would just end up
complicating the GC algorithms with base-offset calculations, and
possibly having to do "windows" into parts of the process's address
space.

Cheers,
- jonathan

> > Within the EROS-native subsystems, we simply use a different design
> > discipline to begin with.
> 
> Do you mean you split up the design in small objects directly?
> 
> Curiously,
> Pierre
> _______________________________________________
> coyotos-dev mailing list
> coyotos-dev at smtp.coyotos.org
> http://www.coyotos.org/mailman/listinfo/coyotos-dev
-- 
- jonathan