> |I wonder if this would work well with the 'multi processed' GC patch
> |discussed earlier :)
>
> Which one?  The one with forking before starting GC?  If so, it is
> crucial to utilize copy-on-write behavior of virtual memory.  But
> forking-before-GC method requires write barrier that may hinder the
> performance.

Yeah the one with forking before starting GC [which is...itself a
write barrier, but at least it's handled by the OS so I didn't have to
write any code for it and hopefully the origin process continues
quickly]. [ruby-core:19569]
And, as you said, it basically needs a copy on write friendly
collector to work :)

I did run into some weirdness running the multi process GC against the
tests at http://redmine.ruby-lang.org/issues/show/670 once, so there
may be a lurking bug in there, but it passed "make test" all right.
Thanks for helping with this.
-=R