On Thu, 2002-09-12 at 17:43, Miguel de Icaza wrote:
> One thing I wanted to explore in the past (but there has been no need to
> do so far) is to use implicit reference counting with cycle detection
> for the GC engine. It uses some extra 32 bits of memory per object
> allocated, but it gives you the predictability you are looking for.
It's an amusing situation where for a real time machine it doesn't so
much matter how slow the mechanism is as long as it's deterministic :-)
As one totally ignorant of the issues - it seems that if people want
real time sections, could there not be an IFreeable interface - since I
assume each object is separately heap allocated; such that it would be
actually freed, and leave dangling invalid pointers ;-) Unsafe in the
extreme of course, but certainly a quick solution. Of course, supporting
that interface in the future with a different GC would perhaps be more
difficult, so it's a fast ugly fix.
Then again, I imagine people want GC and also determinism, which means
cycle detection, hmm ...
Regards,
Michael.
--
[EMAIL PROTECTED] <><, Pseudo Engineer, itinerant idiot
_______________________________________________
Mono-list maillist - [EMAIL PROTECTED]
http://lists.ximian.com/mailman/listinfo/mono-list
