[8u] Regression after JDK-8237117: Better ForkJoinPool behavior

[Anton Kozlov]

Hi,

backport of JDK-8237117 [1] apparently introduces a regression.

It was found during runs of running lucene-solr 8.2.0 [2].
The problem reproduces on a simpler example (in attach) on openjdk8u262-b10[3], but doesn't reproduce on openjdk8u252-b09[4].

When I comment out line [5], the lucene-solr test and the provided reproducer passes.

The change original version is [6].
Newer JDKs have two package private constructors ForkJoinWorkerThread, that call base protected constructor and then initialize inherited ACC to the INNOCUOUS_ACC.
When users subclass ForkJoinWorkerThread, they are free to provide any ACC, which is by default is the current ACC.
Same was in 8u until backport of JDK-8237117, which changes default ACC to INNOCUOUS_ACC instead of the current one.

The backport in 8u is also inconsistent with one in jdk11u[7], which does not change ACC in the protected constructor.
The reproducer also passes on openjdk11.0.8 [8], which do have own backport of the original patch.

Please discuss. It doesn't look the 8u backport is correct.

I've opened JDK-8249677 to track the issue

Thanks,
Anton

[1] http://hg.openjdk.java.net/jdk8u/jdk8u/jdk/rev/f2610c95d0fe
[2] https://github.com/apache/lucene-solr/commit/31d7ec7bbfdcd2c4cc61d9d35e962165410b65fe
[3] https://github.com/AdoptOpenJDK/openjdk8-upstream-binaries/releases/tag/jdk8u262-b10
[4] https://github.com/AdoptOpenJDK/openjdk8-upstream-binaries/releases/tag/jdk8u252-b09
[5] http://hg.openjdk.java.net/jdk8u/jdk8u/jdk/rev/f2610c95d0fe#l1.19
[6] https://hg.openjdk.java.net/jdk/jdk/rev/50d5266f275f
[7] http://hg.openjdk.java.net/jdk-updates/jdk11u/file/0e9eb367df7a/src/java.base/share/classes/java/util/concurrent/ForkJoinWorkerThread.java#l92
[8] https://github.com/AdoptOpenJDK/openjdk11-upstream-binaries/releases/tag/jdk-11.0.8%2B10