On Fri, 5 Sep 2003, Peter Haworth wrote:
> On Thu, 4 Sep 2003 09:02:18 -0400 (EDT), Dan Sugalski wrote:
> > On Wed, 3 Sep 2003, Gordon Henriksen wrote:
> > > A seen hash is also threadsafe, can be interrupted by DoD, and is safe
> > > for recursion. (By "threadsafe," I mean that unchanged data structures
> > > can be safely serialized from multiple threads without ill effect or
> > > possibility of deadlock.)
> >
> > While a seen hash is DOD-interruptable (which, admittedly, the scheme I'm
> > preferring isn't, or is with a *lot* of care) it's also slower and
> > requires a lot more resources when running.
>
> If you're freezing a large graph of PMCs, that's going to take up a lot of
> memory, however you do it (other than streaming to disk). What happens when
> you run out, and need to garbage collect? With your scheme, this is a big
> problem, since as you say, it's either not possible or needs a lot of care.
Note that using the pointers in the PMCs means *DOD* must be disabled, not
*GC*. Doing a GC run to reclaim memory is perfectly fine, though there
still is the potential to have large amounts of memory tied up in dead but
not reaped Pobj things.
> With the seen hash approach, I wouldn't expect the hash itself to take
> nearly as much space as the frozen structure;
My experience tracing variable graphs in Perl 5 indicates otherwise.
Dan
