Re: [cp-patches] RFC: Use WeakHashMap in java.util.ResourceBundle

[David Daney]

Robert Schuster wrote:
> Hi,
>
>
> > Unfortunately I don't think this is the correct solution.  In practice
> > I think this patch would not cause caching for very long -- the next
> > GC cycle would notice that there are no other references to the keys
> > in the map, and it would be cleared.
>
>
> > One idea would be a map where the entries are soft references, so that
> > (in theory) they are only cleared when memory is needed
>
> OK, I thought WeakHashMap would work this way.
>
>
> > (though at
> > least libgcj treats soft references like weak ones atm).  Or, we could
> > have a fixed size cache that removes the least-recently-used entry.
>
> We have so many VM with so many different runtime parameters. I am in favor of
> implementing a SoftHashMap to let the GC decide when to clean out unused 
> references.

The problem (as Tom pointed out) with this is that for many VMs (like 
libgcj) there is no difference between Soft and Weak references.  So the 
end effect is that your approach would do very little caching.  A 
limited size cache using LRU purging and hard references on the other 
hand would work well.

David Daney.