Garbage Collector question?

[chase@xxxxxxxxxxxxx (David Chase)]

At 01:37 PM 11/30/99 +0200, David Soroko wrote:
>I vaguely remember that some (early) versions of hot spot were
>so aggressive in their GC algorithm that local variables were
>GCed while the were still in scope and not used beyond certain point.
>I am not sure that this approach was in any sense illegal, just strange.

As far as I know, this behavior is not illegal.  If you read the
spec, either language or VM, nowhere does it guarantee that you
can reliably observe the size of the heap, nor does it guarantee
that the total amount of memory "available" to you must remain
constant.  What this means is that it is very difficult to observe
exactly when GC takes place, or exactly how much memory is consumed
(or not) by a given allocation.

In particular (and we've implemented this, as a specific outcome
of aggressively applying a general-purpose dead code eliminator)
if you allocate an object that has no constructors (or whose
constructors are all inlined and found to be empty/useless), no
finalizers (*), and do not make subsequent use of the object, it
can be "collected" at compile-time.  Ask yourself, how would you
write a program that could detect this without externally observing
the heap size?  And, if you cannot detect it, then it has no
bearing on whether the implementation does or not does not
conform to the spec, and it surely does not break WORA.

(*) The finalizer could be inlined, but that is special-purpose
and probably a little silly.

David Chase
chase@xxxxxxxxxxxxxxxxx



---
To unsubscribe, mail advanced-java-unsubscribe@xxxxxxxxxxxxxxxx
To get help, mail advanced-java-help@xxxxxxxxxxxxxxxx