On Fri, 17 Feb 2023 00:26:30 GMT, Sergey Bylokhov <s...@openjdk.org> wrote:
>> For this reason, GC logs are enabled. If, for whatever reason, the call to >> `System.gc()` is ignored, we'll see it in the test log. Then a specific GC >> could be selected, for example, or another fix implemented. >> >> OOME does not guarantee a full GC cycle either. >> >> Having the above in mind, I'd rather keep it simple. > >>For this reason, GC logs are enabled. If, for whatever reason, the call to >>System.gc() is ignored, we'll see it in the test log. Then a specific GC >>could be selected, for example, or another fix implemented. > > As of now this test can be executed with different GC, and some of them can > skip System.gc(). > >>OOME does not guarantee a full GC cycle either. > > But it guarantee that at least some GC will be always run, unlike System.gc(). I followed the piece of advice in the [JBS comment](https://bugs.openjdk.org/browse/JDK-8300727?focusedCommentId=14560139&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-14560139). If using `System.gc()` is good enough for testing references: https://github.com/openjdk/jdk/blob/f612dcfebea7ffd4390f833646ad45d6f0ebd04f/test/jdk/java/lang/ref/ReferenceEnqueue.java#L54-L60 https://github.com/openjdk/jdk/blob/f612dcfebea7ffd4390f833646ad45d6f0ebd04f/test/jdk/java/lang/ref/PhantomReferentClearing.java#L85-L92 Then it should be good enough for this test too. Perhaps, the same effect could be achieved by causing OOME in a loop. However, using `System.gc()` makes the intention clearer: *run GC*, wait for the phantom reference to be cleared and enqueued. I can explicitly select a GC: `-XX:+UseG1GC` (default) or `-XX:+UseSerialGC`. ------------- PR: https://git.openjdk.org/jdk/pull/12594
