[gclist] Terminology question, "deferred reference counting"

David Chase chase@centerline.com
Tue, 30 Apr 96 10:42:54 -0400


This came up once before, but I managed to delete/forget the exact 
conclusions.  There's two modifications to reference counting that
I (recall, probably in error) seeing called "deferred reference counting".

1. Don't adjust the reference count for updates arising from local-variable 
operations, as long as it can be inferred that the RC is positive 
when it should be positive.

2. When freeing objects, instead of recursively decrementing and 
freeing an entire data structure, queue/stack those objects with 
reference count of zero and process the queue incrementally.

I suppose I could go dig up the Deutsch and Bobrow paper (I'm
pretty sure that I have a copy) but I think that both techniques
are mentioned in the same paper.  Which of these is called "deferred
reference counting", and what's the other one called?

Please bound your effort to answer this question to either activating 
stale neurons, or opening a drawer in your office, because my copy 
of the paper is almost certainly filed in a banker's box in my basement. 
(Snide remarks about garbage collecting my basement should be suppressed 
unless they are exceptionally clever.)

Thanks to Maeda Atusi for reminding me that I needed to fix this.

thanks much,

David Chase