On Thu, 8 Aug 2024 11:43:07 GMT, Shaojin Wen <d...@openjdk.org> wrote:
>> This PR implements the same algorithm as the current generateMHInlineCopy >> based on bytecode to improve startup performance. > > Shaojin Wen has updated the pull request incrementally with one additional > commit since the last revision: > > fix comments The unbound `CallSite`s should put up the concat classes for GC as soon as they go out of scope, so this doesn't even lean on the instigating classloader being GC'd. Excellent! I think we should try to create a jtreg test from this. @DanHeidinga suggested offline that we could consider using `j.l.instrument.Instrumentation::getAllLoadedClasses()` - which should enumerate all classes, including hidden ones. Perhaps that's easy to mix in with your test above. But having verified that classes *can* be unloaded manually then perhaps we can file an RFE to have such a test worked out later and move ahead with this PR without such in-depth testing. I think such a test might have some challenges to get exactly right since you need to set it up and tune it so that unloading actually happens reliably on all platforms. ------------- PR Comment: https://git.openjdk.org/jdk/pull/20273#issuecomment-2284174629
