One three-round worker pipeline implemented with CountDownLatch (one latch
per round), CyclicBarrier (reusable, with a barrier action), and Phaser
(reusable, plus dynamic mid-run registration), alongside Semaphore solving
the genuinely different problem of bounding concurrent access. Covers
timeout behavior (CyclicBarrier permanently breaks on one timeout, Phaser
doesn't) and virtual-thread compatibility, including the corrected finding
that Object.wait() releases its monitor and never pinned, unlike
Thread.sleep() inside synchronized.
Co-Authored-By: Claude Sonnet 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01FhzLY5p6okFva3qsnsRyvM
Reproduces the pinning JEP 491 did not fix (native JNI frames) with a real
native library, alongside the monitor case it did fix. Covers the
jdk.VirtualThreadPinned JFR event, reading pinned vs unmounted virtual
threads from a jcmd Thread.dump_to_file, the measured throughput cost on a
capped carrier pool, and a bounded-executor pattern to contain the blast
radius.
Co-Authored-By: Claude Sonnet 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01FhzLY5p6okFva3qsnsRyvM
AtomicLong/LongAdder/VarHandle counters benchmarked against the synchronized
and ReentrantLock baselines from the locks module, a deterministic ABA race
against a hand-rolled Treiber stack plus the AtomicStampedReference fix, and
a VarHandle access-modes demo (plain/opaque/acquire-release/volatile).
Co-Authored-By: Claude Sonnet 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01FhzLY5p6okFva3qsnsRyvM
JMH throughput sweep (1-64 threads), 9:1 read-heavy @Group benchmark,
tryLock(timeout) deadlock-avoidance demo, and a JDK21-vs-JDK25 JEP 491
virtual-thread-pinning comparison. Also promotes LICENSE to repo root
now that a second module exists.
asmhatreandClaude Sonnet 5