On Wed, 15 Jan 2025 14:14:37 GMT, Richard Reingruber <rr...@openjdk.org> wrote:
> This PR reverts the fix from > [JDK-8339166](https://bugs.openjdk.org/browse/JDK-8339166) because it > increases the runtime of the test a lot. > Instead a full gc is requested via the whitebox api. This solves the issues > (see bug description) and it is also clearer and more reliable then > allocating objects and classes expecting a gc because of the small heap. > The heap setting is also removed since it is not needed anymore. > > #### Testing: Duration Running HiddenClassUnloading with fastdebug Builds > > I'm giving the minimum of a few runs because durations vary by +/- 10s. > > > Linux / x86_64: > After JDK-8339166: 3m15s > JDK-8339166 reverted: 50s > PR: 50s > > macOS / M1: > After JDK-8339166: 2m30s > JDK-8339166 reverted: 30s > PR: 30s > > Linux / ppc64le > After JDK-8339166: 12m30s > JDK-8339166 reverted: 2m30s (failure: no classes unloaded) > PR: 2m30s (success) LGTM; probably you do not need 2000 iterations now with the full GC, a smaller number would do as well, correct ? But it is okay for me as it is. ------------- Marked as reviewed by mbaesken (Reviewer). PR Review: https://git.openjdk.org/jdk/pull/23137#pullrequestreview-2563767103
