Linux and GC

Alaric B. Williams
Tue, 12 May 1998 08:52:55 +0100

Scott L. Burson wrote:

> The technique Kragen suggests may have been studied, although I don't
> think I had heard of it, but it's not what's called a "write barrier".
>  A write barrier is a piece of code or possibly hardware that detects
> writes of pointers into memory having certain properties: roughly,
> where the cell into which the pointer is written will not be scanned
> (again) during the current collection, but the object to which the
> pointer points has not been marked as live.  A write barrier is needed
> in both fully incremental collectors, such as the Lisp machine's, as
> well as semi-incremental collectors (a term I just made up to cover
> collectors, such as Allegro's, that run uninterruptibly during a
> collection but scan and collect only a small amount of memory
> each time).

I'd still call it a write barrier, though, since it's extra checking
done on every write for the GC's benefit. It's just a funny kind of
write barrier.

> -- Scott


Al Williams  ICQ#5129948  Pager +44 4325 325477
PGP: D8F3 FA8E 32A3 2F3B 13A9  B3DF DB50 1FE9 9913 1E3E

  And you see a girl's brown body dancing through the turquoise,
  And her footprints make you follow where the sky loves the sea.
  And when your fingers find her, she drowns you in her body,
  Carving deep blue ripples in the tissues of your mind.