On Mon, 5 Feb 2001, Mark Fowler wrote:
> As for the real time nature of garbage collectors, Shevek and I (and some
> others) wrote a paper together on this, so I could say a lot here.
> Especially about how Perl is much more real time as it can do better at
> allocating and deallocating the same bit of memory (as it instantly
> knows when memory is free, unlike mark and sweep) and is a lot less likely
> to cause page faults. Of course, reference counting suffers from the
> dreaded deep free problem so if you don't program carefully you can very
I later prove that it is possible to do a hard real time reference
counting garbage collector, without the deep free problem, and without a
fallback mark and sweep. However, the other problems, including reference
cycles, still exist. Allocation time could be time proportional to the
requested region size, but that's fine as you can control allocation.
> very easily shoot yourself in the time constraint foot too.
Didn't we also manage mark and sweep in something approximating real time
with a high enough hard constant? I've been looking at a lot more systems
since then, region inference is interesting. You can free data which isn't
actually unreferenced. :-) Scary.
S.
--
Shevek
I am the Borg.
sub AUTOLOAD { ($s=$AUTOLOAD)=~s/.*:://; eval qq{ *$AUTOLOAD=$s
?sub {$s*&{$s-1}} :sub {1}; }; goto &$AUTOLOAD; } print &{'4'};
