improve performance of projection instantiation

[stsypanov]

I've run into poor performance of projection creation with Spring Data JPA. Using async profiler I've measured 10 hottest methods for my particular case:

          ns  percent  samples  top
  ----------  -------  -------  ---
  1719680599   17.75%     1719  org.springframework.util.ConcurrentReferenceHashMap.calculateShift
  1075919136   11.11%     1076  org.springframework.util.ConcurrentReferenceHashMap.<init>
   995216304   10.27%      995  org.springframework.aop.framework.JdkDynamicAopProxy.findDefinedEqualsAndHashCodeMethods
   549197362    5.67%      549  java.util.LinkedList.toArray
   505954398    5.22%      506  java.lang.String.regionMatches
   441133709    4.55%      441  java.util.concurrent.ConcurrentHashMap.get
   187450280    1.93%      187  /usr/lib/jvm/java-8-openjdk-amd64/jre/lib/amd64/server/libjvm.so
   166974974    1.72%      167  java.util.ArrayList$Itr.<init>
   160991064    1.66%      161  org.springframework.aop.framework.JdkDynamicAopProxy.getProxy
   155019312    1.60%      155  org.apache.logging.slf4j.SLF4JLogger.isEnabledFor

The simple way to slightly improve this is to get rid of LinkedList in org.springframework.aop.framework.AdvisedSupport and use plain constructor call in org.springframework.util.ConcurrentReferenceHashMap.<init>.

LinkedList::toArray is obviously slower than ArrayList::toArray.

In ConcurrentReferenceHashMap method createReferenceArray can use plain constructor to instantiate array which yields equal performance when compiled with C2 but 3 times better performance in interpreter mode.

[jhoeller]

Thanks for pointing those out! I'll backport them to 5.0.8 in my next round of polishing there as well.

[stsypanov]

@jhoeller I've spotted a lot of usages of LinkedList in code of Spring Framework. Is it acceptable to create a separate pull request to replace them all with ArrayList/ArrayDeque?

[jhoeller]

Instead of a straight find-and-replace PR, I'd prefer reviewing those LinkedList usage spots my side, having a look at the context around them and potentially revisiting the overall state arrangement in the affected classes. Could you create a JIRA ticket for this please, I'm happy to pick it up for 5.1 RC1 still: https://jira.spring.io/browse/SPR

[stsypanov]

Ticket created: https://jira.spring.io/browse/SPR-17037