stataru8 commented on PR #564: URL: https://github.com/apache/felix-dev/pull/564#issuecomment-5955464874
Hello, thanks for taking the time to check this PR. > Are the changes made measurable when ran in isolation in a red/green fashion? I would assume Object.hashCode would be highly optimised already Yes, it's measurable, but only in a specific JVM state. All the details are in [FELIX-6863](https://issues.apache.org/jira/browse/FELIX-6863). For the red/green comparison, the ticket has a reproducer with before/after numbers in the same scenario: |Kars installed|feature:uninstall before restart|after restart|after restart, with the fix| | -------- | -------- |-------- | -------- | |1|~1.3s|~2.3s|~1.4s| |300|~4s|~22s|~3.5s| (https://issues.apache.org/jira/secure/attachment/13084874/feature-uninstall-measurements_300_kars.md) You're right that `Object.hashCode` is normally very cheap. What I observed is that after a JVM restart, it can end up on a much slower path when called from `ResourceImpl.hashCode`: in the JFR recordings, almost all samples of the features thread are inside the native `Object.hashCode`, called from `ResourceImpl.hashCode`. My understanding is that this depends on how the JIT compiled that call site at startup. I first ran into this issue on **Karaf 4.4.6**, and I was able to reproduce it on **4.4.11**. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
