Michael L Maraist wrote:
> No barriers for us?
Generational collectors require a write barrier because
old objects must never point to younger ones. ('Course Dan
said he's starting with a simple copying collector, so we
don't need a barrier. Hmm. I guess Dan's not *reject*ing
a barrier, just rejecting a barrier *now*. ;)
Your point about XS code is a good one. I'm not sure the
write barrier is a problem though because all XS code
currently goes through the inc/dec/store API. The main
problem IMHO is that XS code caches PMC internals -- if
a GC moves the internals the XS code will crash. All
copying collectors have this problem.
Good article. I haven't gotten through the vmem article
yet and now here's another one. ;)
- Ken
