Review request for JDK-8035948 - Redesign property listeners for shared classes

Marcus Lagergren marcus.lagergren at oracle.com
Mon Mar 3 05:01:13 PST 2014


Yes looks good otherwise and fairly simple to merge with optimistic types.

/M

On 03 Mar 2014, at 10:36, Hannes Wallnoefer <hannes.wallnoefer at oracle.com> wrote:

> I didn't notice any overhead from synchronization.
> 
> I did notice a small reduction in performance from using soft references instead of hard references in PropertyMap history in some benchmarks (mostly deltablue I think). However, other solutions but they either had worse effects on performance or could possibly cause memory leaks.
> 
> Still I think the net effect on performance is a positive one (even if only slightly) caused by reduced listener creation and history lookup overhead.
> 
> Hannes
> 
> 
> Am 2014-03-02 10:39, schrieb Marcus Lagergren:
>> Have you done any benchmarking whether the synchronization in PropertyListeners gives us overhead? Any other numbers we should know about?
>> 
>> On 27 Feb 2014, at 18:22, Hannes Wallnoefer <hannes.wallnoefer at oracle.com> wrote:
>> 
>>> Please review JDK-8035948 - Redesign property listeners for shared classes:
>>> 
>>> http://cr.openjdk.java.net/~hannesw/8035948/
>>> 
>>> This prepares property maps and property listeners to work better with classes that are reused between JS environments. The central ideas are:
>>> 
>>> - Reuse property maps between globals in order to minimize polymorphism.
>>> - Use SoftReferences for property map history to avoid memory leaks
>>> - Redesign property listeners to be attatched to property maps only since ScriptObjects are specific to one JS environment.
>>> - Reduce number of property listeners by only creating them when needed (when a callsite accesses an inherited property).
>>> - Reduce number of prototype history lookups in property maps by using a unique combination of prototype/property maps from the start.
>>> 
>>> Actual class sharing functionality is coming in a later webrev. Let me know if you have any questions about this.
>>> 
>>> Hannes
> 



More information about the nashorn-dev mailing list