Concurrency: threads share memory on purpose

The naive mental model is broken by caches

Modern CPUs have per-core caches and instruction reordering. Without synchronization, a thread on core A may never see a write that a thread on core B just made to an ordinary field — not out of malice, but because the hardware optimized for speed. The Java Memory Model specifies when one thread's writes become visible to another via happens-before edges.

Tools in order of preference

1. Immutable data — no synchronization needed, shareable by construction.
2. Higher-level concurrency utilities — ExecutorService, ConcurrentHashMap, CompletableFuture, structured concurrency (Loom).
3. volatile fields — establish a happens-before for a single variable.
4. synchronized blocks and ReentrantLock — protect critical sections.
5. Low-level primitives (VarHandle, Unsafe) — last resort, for library authors.