Re: Garbage Collection

[sommar@enea.se (Erland Sommarskog)]

This is where this discussion ends for my part. It has been going
on far too long. It seems quite obvious that no one here, at least
not me, will succeed in converting Bill Wolfe to believe that garbage 
collection is good. And it does not seem like he will convert anyone 
either.

I'm glad to hear that Reference.all will survive the block exit.
I am still a little puzzled, though. Assume that instead of a "simple 
pointer" we actually have a linked list, declared as (limited) 
private in some package. Then, it's Destroy routine will be called
at block exit. Now, I don't know how you folks write your Assign
procedures for linked lists, but I do simple pointer copying. (And
I assume that Bill Wolfe does the same, since he is so keen on that
his ADTs will be useful in a real-time environment. Copying value
and allocating memory for the new lists would be as harmful as
garbage collection, if not worse.) This means of course that my
implicit Destroy procedure must be just NULL(*). Then I have an another
for the user to call when is tired of his list.
  Excuse me, but is call-destroy-on-block-exit really a useful device? 
Seems like adding something to the langauge which you almost should never 
use. 

And same goes for the user to supply a Destroy procedure to the 
generic package. Since its body often will be NULL, its appearance
in the instantiating package is just white noise.

(*) This isn't really true. If the reference assignment also includes
updating of reference counters, we can check them in the Destroy
procedure. We have now drifted in to the land of poor man's garbage
collection, where to have to very careful everytime so we are not
doing something stupid.

Finally, Bill Wolfe has claimed that not having garbage collection
helps the programmer to understands all aspects of his code, both
time and space. Eh, there are some more aspects on the code than
so. For example, that it really does what it supposed to. It may very
elegantly sweep away all used memory and that, but tell us that
2 + 2 = 5. No good program. Fact is, in many applications, space is
seldom an important problem, if at all. All you want to be sure of
is that you really got rid all of you allocated, so that your interactive
user donesn't get a dump after five hours work.

And, oh, the world is not divided into ADT implementors and application
programmers. It just doesn't look that way.
-- 
Erland Sommarskog
ENEA Data, Stockholm              This signature is not to be quoted.
sommar@enea.se