This turns out to be a JVM bug (really; no shit:
http://mail.openjdk.java.net/pipermail/hotspot-runtime-dev/2009-January/000386.h
tml
)
To trigger it, have lots of threads attempt to do a
IBuffer.getByteBuffer(...) simultaneously (which is what happened in that
exhaustive test).
They would trigger a deadlock in Sun's JNI NewDirectByteBuffer call.
Workaround coming real soon -- I'll make sure the first
JNI_NewDirectByteBuffer call occurs in a thread-safe way.
Original issue reported on code.google.com by art.cla...@gmail.com on 7 May 2009 at 6:18
Original issue reported on code.google.com by
art.cla...@gmail.com
on 7 May 2009 at 6:18