On Fri, 14 Jan 2022 19:27:18 GMT, Aleksey Shipilev <sh...@openjdk.org> wrote:
> JDK-8277072 introduced java.io.ClassCache, but there seem to be at least two > issues with it: > - The cache cannot disambiguate between cleared SoftReference and the > accidental passing of `null` value; in that case, the retry loop would spin > indefinitely; > - If retry loop would spin several times, every time discovering a cleared > SoftReference, it would create and register new SoftReference on the > ReferenceQueue. However, it would not *drain* the RQ in the loop; in that > case, we might have the unbounded garbage accumulating in RQ; > > Attention @rkennke, @plevart. > > Additional testing: > - [x] Linux x86_64 fastdebug `java/io/ObjectStreamClass` > - [x] Linux x86_64 fastdebug `tier1` > - [ ] Linux x86_64 fastdebug `tier2` > - [ ] Linux x86_64 fastdebug `tier3` Right, good catches! It seems a bit weird that the constructors of Reference would accept null referents anyway, but it is like it is. *shrugs* One additional improvement would be to change r.get() == null to r.refersTo(null) to check for cleared reference, that avoids reviving the referent, e.g. in SATB GCs. I leave that up to you. ------------- Marked as reviewed by rkennke (Reviewer). PR: https://git.openjdk.java.net/jdk/pull/7092
