On Thursday, 23 August 2012 at 14:49:11 UTC, Alex Rønne Petersen
wrote:
On 23-08-2012 16:47, dsimcha wrote:
On Thursday, 23 August 2012 at 14:38:19 UTC, Alex Rønne
Petersen wrote:
Yes, but parallelization of the mark phase is fairly trivial,
and
something we should probably look into.
Ironically, Antti-ville's original proposal involved
parallelization.
This was scrapped because after rtinfo was added, we agreed
that precise
heap scanning was more important and looked newly feasible.
Oh, I agree it's more important. The GC is reasonably fast
as-is, but has severe issues with false pointers. Parallel
marking is just in the "nice to have" category.
Funny, I've just read an interview from Narihiro Nakamura who is
working on parallel marking in Ruby:
http://rubysource.com/narihiro-nakamura-rubys-gc-innovator/
BR,
renoX
