Promptness of collection

 

I've just been referred to a recent paper by Blackburn et al. called "Down
for the Count?" that repeats a claim that grates with me. Specifically, it
says that reference counting "can reclaim objects as soon as they are no
longer referenced" and describes "promptness of recovery" as an advantage of
reference counting. While I don't doubt that reference counting at the level
of registers can reclaim aggressively, that is practically unheard of
because the overheads are crippling. Real reference counting implementations
seem to at least defer to scope-based and sometimes defer much further, at
which point I think there is a big question mark over the amount of floating
garbage that lingers. My impression is that even scope-based reference
counting (e.g. C++ smart pointers) produces more floating garbage than a
simple generational mark-sweep collector.

 

Has anyone tried to quantify the promptness of recovery of real reference
counting garbage collectors and compared them to other garbage collections
(e.g. generational mark-sweep)?

 

-- 

Dr Jon Harrop, Flying Frog Consultancy Ltd.

 <http://www.ffconsultancy.com/> http://www.ffconsultancy.com