commons-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jack, Paul" <pj...@sfaf.org>
Subject RE: [collections] SoftRefHashMap (was RE: Avalon excalibur collec tions migration status)
Date Thu, 18 Jul 2002 02:14:44 GMT
> Ummm..  I'd go so far as to say that in *all* cases, WeakReferences 
> would behave differently than SoftReferences.  WeakReferences are 
> cleared once an object is only weakly reachable.  SoftReferences are 
> *not* cleared when they are only softly reachable -- they are cleared 
> when memory constraints demand.  That's two very different 
> behaviors and it doesn't matter whether the GC runs incrementally or 
> in a stop-the-world manner. 

No, Joshua has a point here.  WeakReferences aren't automatically
cleared the instant the object becomes weakly reachable; the garbage
collector first has to notice that there aren't any hard references
to the object.  If the gc is "stop-the-world" and only
runs when there's demand for memory, then it won't notice weakly
reachable objects, and won't clear any WeakReferences, until there's
memory demand.  Essentially they'd be acting like SoftReferences,
without the bias you mention below.

But even in that presumably rare case, the behavior would still be
different enough to warrant the inclusion of both weak and soft 
references in ReferenceMap, so I guess the point is moot.


> > We could use an internal SequencedHashMap to store strong 
> > references, actually, since it only keeps the x most recent 
> > mappings.
> 
> I think it's unnecessary.  If the JVM needs the memory, it needs the
> memory.  The VM should not be clearing SoftReferences if it 
> doesn't need the memory.  It is true, though, that there are no 
> constraints on *which* soft references are cleared, but the API spec 
> says that "Virtual machine implementations are, however, encouraged 
> to bias against clearing recently-created or recently-used soft 
> references."  Thus, you already have a SequencedHashMap type thing 
> for those recently created/used references.  Yeah, ok, recently 
> created isn't as useful as recently used.  But I think it's better 
> to leave it up to the application of SoftRefHashMap to hold onto 
> strong references rather than build it into SoftRefhashMap itself.

Yeah, the more I think about it, there are many different ways people
might want to keep hard references.  For instance, they might want to
keep hard references to objects that were created/used within the past
five minutes, but soft references to older objects.  

Point being, I think people could use ReferenceMap as a building block
for more complicated memory-sensitive caches if they needed to, but 
building those complications into ReferenceMap doesn't make sense.  So,
I won't add any freaky MRU stuff.

I'll probably submit ReferenceMap minus phantoms in about an hour...

-Paul


--
To unsubscribe, e-mail:   <mailto:commons-dev-unsubscribe@jakarta.apache.org>
For additional commands, e-mail: <mailto:commons-dev-help@jakarta.apache.org>


Mime
View raw message