Proxy.isProxyClass scalability (original) (raw)

Peter Levart peter.levart at gmail.com
Thu Apr 25 15:40:26 UTC 2013

Hi Mandy,

Here's another update that changes the sub-key back to WeakReference based:

http://dl.dropboxusercontent.com/u/101777488/jdk8-tl/proxy-wc/webrev.05/index.html

On 04/25/2013 03:38 AM, Mandy Chung wrote:

I like this one too. The mapping is straight forward and clean that avoids the fallback and post validation path. Let's proceed with this one. It's good to optimize for the common case (1-interface). For 2 or more interfaces, we can improve the memory usage in the future when it turns out to be an issue. I'm fine with the zero-interface proxy which is the current implementation too.

I made 3 straight-forward implementations of sub-keys:

The cache-structure size increment for each new cached proxy-class is (32 bit OOPS):

#of intfcs original patched

      1       152  128(Key1)
      2       152  168(Key2), 208(KeyX)
      3       160  248(KeyX)
      4       160  280(KeyX)

...so you can decide if Key2 is worth having or not.

WeakCache.java The javadoc for the get(K,P) method - @throws RuntimeException and @throws Error are not needed here since any method being called in the implementation may throw unchecked exceptions. There are a couple places that checks if (reverseMap != null) .... that check is not needed since it's always non-null. But I'm fine if you keep it as it is - just wanted to mention it in case it was just leftover from the previous version.

Removed the unneeded @throws and reverseMap != null checks (the later was already removed in latest string-based webrev and I used that version here).

I think we're very close of getting this pushed.

Do you think this could also get backported to JDK7? The WeakCache uses two interfaces from java.util.function. Should we make private equivalents in this patch or do we leave that for the possible back-porting effort. I should note that JDK7's ConcurrentHashMap is not that space-efficient as proposed JDK8's, so space-usage would be different on JDK7...

Regards, Peter

thanks Mandy

More information about the core-libs-dev mailing list