Bug Report: BinaryLinkClient::pushFrame, in fringe cases, is reentrant

[Aleksey Shipilev]

On 10/11/19 10:41 PM, Berry, Matt wrote:
> The result is that stderr is emitted right in the middle of the frame that runBenchmark was
> trying to emit. When the BinaryLinkServer reads this data, the deserialization process blows up
> violently. The exact behavior is going to depend on the data written to stderr. In my case, the
> message was [2] from (ObjectStreamClass.java:782) but, depending on what bits get set, some
> invariant in that class will not be happy.

That certainly never happened before. Odd corner case, that! Recorded as:
  https://bugs.openjdk.java.net/browse/CODETOOLS-7902528

The easiest patch I can see is:
  http://cr.openjdk.java.net/~shade/7902528/7902528.patch

> A demonstration of the issue is available in this git repository [3].

Note that reproducer uses JMH 1.0. It does reproduce on latest JMH 1.21 too.

-- 
Thanks,
-Aleksey