Please join me...

[dsimcha]

== Quote from div0 (div0 at users.sourceforge.net)'s article
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> Andrei Alexandrescu wrote:
> > ...in welcoming David Simcha to the Phobos developers list.
> >
> > David has done great work and we think he will add great value to
> > Phobos. I suggest Sean to add him to the druntime developers list to
> > make it easier to integrate David's garbage collector.
> >
> >
> > Andrei
> I'm not joining you. You are ugly and your mother smells of elderberries.
> Does this mean we get a precise GC?! Sweet!

Precise heap scanning.  The stack and static data segment will still be
conservative.  Also, there are still some plumbing issues to resolve as far as
getting the pointer offset info to the GC when an object is new'd as opposed to
malloc'd.

IMHO, though, precise heap scanning should solve most false pointer problems.  In
any program that uses enough memory for false pointers to be problematic, the heap
is orders of magnitude larger than the stack and static data segment.  However, I
haven't had a chance to test this in real-world programs b/c of the aforementioned
problems with new.

As far as precise stack and static data segment scanning, they're an order of
magnitude harder because they would require major changes to the compiler.  I was
able to do precise heap scanning with only a few templates and changes to the
runtime.  The only compiler changes that will be necessary are the minor plumbing
issues with new that I mentioned already.