Re: Pliant and ref counting

[Russ McManus <russell.mcmanus@gs.com>]

> I am coming to the conclusion that garbage collection won't work for
> large databases. At least, you can't garbage collect them often
> because there is so much data in the system that collection time
> slows everything down.

Would generational gc help?  This sounds like the canonical case for
it.

> I feel that maybe within the database I sould be reference counting
> and then copy the data out to a garbage collection environment on the
> outside of the database. I would prefer to avoid needless copying but
> copying is fast compared to scanning the whole database.

If you know when your object are supposed to die, you can do some
management yourself using dynamic-wind.  This has been discussed
recently over on comp.lang.lisp; you may want to check dejanews.

Just some marginally related ideas...

> The argument of ref counting against garbage collection is still largely
> unsolved, and depends on the application.

Personally, I think of refcounting as a simplistic gc strategy, so I
don't see the two concept in conflict.  There are tradeoffs, I agree
to different gc strategies.

> The difficult side of things is when you have a throw and catch
> system that can figure out which variables fall out of scope and
> thus which references to update. I have yet to see and elegant
> solution to that problem.

Yes.  Exceptions and cycles make refcounting more difficult, and it's
supposed simplicity benefits start to look suspect.

  "For every complex problem, there is a solution that is simple,
  neat, and wrong."  
               -- H. L. Mencken

-russ


--
If at first you don't succeed, don't try sky diving.