- * This cache maps a ({@link PackFile},position) tuple to an Object.
- *
- * This cache is suitable for objects that are "relative expensive" to compute - * from the underlying PackFile, given some known position in that file. - *
- * Whenever a cache miss occurs, {@link #load(PackFile, long)} is invoked by
- * exactly one thread for the given (PackFile,position) key tuple.
- * This is ensured by an array of locks, with the tuple hashed to a lock
- * instance.
- *
- * During a miss, older entries are evicted from the cache so long as - * {@link #isFull()} returns true. - *
- * Its too expensive during object access to be 100% accurate with a least - * recently used (LRU) algorithm. Strictly ordering every read is a lot of - * overhead that typically doesn't yield a corresponding benefit to the - * application. - *
- * This cache implements a loose LRU policy by randomly picking a window - * comprised of roughly 10% of the cache, and evicting the oldest accessed entry - * within that window. - *
- * Entities created by the cache are held under SoftReferences, permitting the - * Java runtime's garbage collector to evict entries when heap memory gets low. - * Most JREs implement a loose least recently used algorithm for this eviction. - *
- * The internal hash table does not expand at runtime, instead it is fixed in - * size at cache creation time. The internal lock table used to gate load - * invocations is also fixed in size. - *
- * The key tuple is passed through to methods as a pair of parameters rather - * than as a single Object, thus reducing the transient memory allocations of - * callers. It is more efficient to avoid the allocation, as we can't be 100% - * sure that a JIT would be able to stack-allocate a key tuple. - *
- * This cache has an implementation rule such that: - *
(PackFile,position) tuple.load() invocation there is exactly one
- * {@link #createRef(PackFile, long, Object)} invocation to wrap a SoftReference
- * around the cached entity.createRef() there will be
- * exactly one call to {@link #clear(Ref)} to cleanup any resources associated
- * with the (now expired) cached entity.- * Therefore, it is safe to perform resource accounting increments during the - * {@link #load(PackFile, long)} or {@link #createRef(PackFile, long, Object)} - * methods, and matching decrements during {@link #clear(Ref)}. Implementors may - * need to override {@link #createRef(PackFile, long, Object)} in order to embed - * additional accounting information into an implementation specific - * {@link OffsetCache.Ref} subclass, as the cached entity may have already been - * evicted by the JRE's garbage collector. - *
- * To maintain higher concurrency workloads, during eviction only one thread
- * performs the eviction work, while other threads can continue to insert new
- * objects in parallel. This means that the cache can be temporarily over limit,
- * especially if the nominated eviction thread is being starved relative to the
- * other threads.
- *
- * @param
- * This is a last-ditch effort to clear out the cache, such as before it
- * gets replaced by another cache that is configured differently. This
- * method tries to force every cached entry through {@link #clear(Ref)} to
- * ensure that resources are correctly accounted for and cleaned up by the
- * subclass. A concurrent reader loading entries while this method is
- * running may cause resource accounting failures.
- */
- void removeAll() {
- for (int s = 0; s < tableSize; s++) {
- Entry
- * Typically this method is invoked during {@link PackFile#close()}, when we
- * know the pack is never going to be useful to us again (for example, it no
- * longer exists on disk). A concurrent reader loading an entry from this
- * same pack may cause the pack to become stuck in the cache anyway.
- *
- * @param pack
- * the file to purge all entries of.
- */
- void removeAll(final PackFile pack) {
- for (int s = 0; s < tableSize; s++) {
- final Entry
- * This method is invoked by {@link #getOrLoad(PackFile, long)} when the
- * specified entity does not yet exist in the cache. Internal locking
- * ensures that at most one thread can call this method for each unique
- *
- * Implementing this is only necessary if the subclass is performing
- * resource accounting during {@link #load(PackFile, long)} and
- * {@link #clear(Ref)} requires some information to update the accounting.
- *
- * Implementors MUST ensure that the returned reference uses the
- * {@link #queue} ReferenceQueue, otherwise {@link #clear(Ref)} will not be
- * invoked at the proper time.
- *
- * @param pack
- * the file to materialize the entry from.
- * @param position
- * offset within the file of the entry.
- * @param v
- * the object returned by {@link #load(PackFile, long)}.
- * @return a soft reference subclass wrapped around
- * This method is invoked exactly once for the combined
- * {@link #load(PackFile, long)} and
- * {@link #createRef(PackFile, long, Object)} invocation pair that was used
- * to construct and insert an object into the cache.
- *
- * @param ref
- * the reference wrapped around the object. Implementations must
- * be prepared for
- * By default this method returns false. Implementors may override to
- * consult with the accounting updated by {@link #load(PackFile, long)},
- * {@link #createRef(PackFile, long, Object)} and {@link #clear(Ref)}.
- *
- * @return true if the cache is still over-limit and requires eviction of
- * more entries.
- */
- protected boolean isFull() {
- return false;
- }
-
- @SuppressWarnings("unchecked")
- private void gc() {
- R r;
- while ((r = (R) queue.poll()) != null) {
- // Sun's Java 5 and 6 implementation have a bug where a Reference
- // can be enqueued and dequeued twice on the same reference queue
- // due to a race condition within ReferenceQueue.enqueue(Reference).
- //
- // http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6837858
- //
- // We CANNOT permit a Reference to come through us twice, as it will
- // skew the resource counters we maintain. Our canClear() check here
- // provides a way to skip the redundant dequeues, if any.
- //
- if (r.canClear()) {
- clear(r);
-
- boolean found = false;
- final int s = slot(r.pack, r.position);
- final Entry
- * For example,
- * A true here indicates that the ref is no longer accessible, and that
- * we therefore need to eventually purge this Entry object out of the
- * bucket's chain.
- */
- volatile boolean dead;
-
- Entry(final Entry
+ * Whenever a cache miss occurs, {@link #load(PackFile, long)} is invoked by
+ * exactly one thread for the given
+ * During a miss, older entries are evicted from the cache so long as
+ * {@link #isFull()} returns true.
+ *
+ * Its too expensive during object access to be 100% accurate with a least
+ * recently used (LRU) algorithm. Strictly ordering every read is a lot of
+ * overhead that typically doesn't yield a corresponding benefit to the
+ * application.
+ *
+ * This cache implements a loose LRU policy by randomly picking a window
+ * comprised of roughly 10% of the cache, and evicting the oldest accessed entry
+ * within that window.
+ *
+ * Entities created by the cache are held under SoftReferences, permitting the
+ * Java runtime's garbage collector to evict entries when heap memory gets low.
+ * Most JREs implement a loose least recently used algorithm for this eviction.
+ *
+ * The internal hash table does not expand at runtime, instead it is fixed in
+ * size at cache creation time. The internal lock table used to gate load
+ * invocations is also fixed in size.
+ *
+ * The key tuple is passed through to methods as a pair of parameters rather
+ * than as a single Object, thus reducing the transient memory allocations of
+ * callers. It is more efficient to avoid the allocation, as we can't be 100%
+ * sure that a JIT would be able to stack-allocate a key tuple.
+ *
+ * This cache has an implementation rule such that:
+ *
+ * Therefore, it is safe to perform resource accounting increments during the
+ * {@link #load(PackFile, long)} or
+ * {@link #createRef(PackFile, long, ByteWindow)} methods, and matching
+ * decrements during {@link #clear(Ref)}. Implementors may need to override
+ * {@link #createRef(PackFile, long, ByteWindow)} in order to embed additional
+ * accounting information into an implementation specific {@link Ref} subclass,
+ * as the cached entity may have already been evicted by the JRE's garbage
+ * collector.
+ *
+ * To maintain higher concurrency workloads, during eviction only one thread
+ * performs the eviction work, while other threads can continue to insert new
+ * objects in parallel. This means that the cache can be temporarily over limit,
+ * especially if the nominated eviction thread is being starved relative to the
+ * other threads.
*/
-public class WindowCache extends OffsetCache
+ * This is a last-ditch effort to clear out the cache, such as before it
+ * gets replaced by another cache that is configured differently. This
+ * method tries to force every cached entry through {@link #clear(Ref)} to
+ * ensure that resources are correctly accounted for and cleaned up by the
+ * subclass. A concurrent reader loading entries while this method is
+ * running may cause resource accounting failures.
+ */
+ private void removeAll() {
+ for (int s = 0; s < tableSize; s++) {
+ Entry e1;
+ do {
+ e1 = table.get(s);
+ for (Entry e = e1; e != null; e = e.next)
+ e.kill();
+ } while (!table.compareAndSet(s, e1, null));
+ }
+ gc();
+ }
+
+ /**
+ * Clear all entries related to a single file.
+ *
+ * Typically this method is invoked during {@link PackFile#close()}, when we
+ * know the pack is never going to be useful to us again (for example, it no
+ * longer exists on disk). A concurrent reader loading an entry from this
+ * same pack may cause the pack to become stuck in the cache anyway.
+ *
+ * @param pack
+ * the file to purge all entries of.
+ */
+ private void removeAll(final PackFile pack) {
+ for (int s = 0; s < tableSize; s++) {
+ final Entry e1 = table.get(s);
+ boolean hasDead = false;
+ for (Entry e = e1; e != null; e = e.next) {
+ if (e.ref.pack == pack) {
+ e.kill();
+ hasDead = true;
+ } else if (e.dead)
+ hasDead = true;
+ }
+ if (hasDead)
+ table.compareAndSet(s, e1, clean(e1));
+ }
+ gc();
+ }
+
+ @SuppressWarnings("unchecked")
+ private void gc() {
+ Ref r;
+ while ((r = (Ref) queue.poll()) != null) {
+ // Sun's Java 5 and 6 implementation have a bug where a Reference
+ // can be enqueued and dequeued twice on the same reference queue
+ // due to a race condition within ReferenceQueue.enqueue(Reference).
+ //
+ // http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6837858
+ //
+ // We CANNOT permit a Reference to come through us twice, as it will
+ // skew the resource counters we maintain. Our canClear() check here
+ // provides a way to skip the redundant dequeues, if any.
+ //
+ if (r.canClear()) {
+ clear(r);
+
+ boolean found = false;
+ final int s = slot(r.pack, r.position);
+ final Entry e1 = table.get(s);
+ for (Entry n = e1; n != null; n = n.next) {
+ if (n.ref == r) {
+ n.dead = true;
+ found = true;
+ break;
+ }
+ }
+ if (found)
+ table.compareAndSet(s, e1, clean(e1));
+ }
+ }
+ }
+
+ private int slot(final PackFile pack, final long position) {
+ return (hash(pack.hash, position) >>> 1) % tableSize;
+ }
+
+ private Lock lock(final PackFile pack, final long position) {
+ return locks[(hash(pack.hash, position) >>> 1) % locks.length];
+ }
+
+ private static Entry clean(Entry top) {
+ while (top != null && top.dead) {
+ top.ref.enqueue();
+ top = top.next;
+ }
+ if (top == null)
+ return null;
+ final Entry n = clean(top.next);
+ return n == top.next ? top : new Entry(n, top.ref);
+ }
+
+ private static class Entry {
+ /** Next entry in the hash table's chain list. */
+ final Entry next;
+
+ /** The referenced object. */
+ final Ref ref;
+
+ /**
+ * Marked true when ref.get() returns null and the ref is dead.
+ *
+ * A true here indicates that the ref is no longer accessible, and that
+ * we therefore need to eventually purge this Entry object out of the
+ * bucket's chain.
+ */
+ volatile boolean dead;
+
+ Entry(final Entry n, final Ref r) {
+ next = n;
+ ref = r;
+ }
+
+ final void kill() {
+ dead = true;
+ ref.enqueue();
+ }
+ }
+
+ /** A soft reference wrapped around a cached object. */
+ private static class Ref extends SoftReferencepack of the object.
- * @return the object reference.
- * @throws IOException
- * the object reference was not in the cache and could not be
- * obtained by {@link #load(PackFile, long)}.
- */
- V getOrLoad(final PackFile pack, final long position) throws IOException {
- final int slot = slot(pack, position);
- final Entry(pack,position), but multiple threads can call this method
- * concurrently for different (pack,position) tuples.
- *
- * @param pack
- * the file to materialize the entry from.
- * @param position
- * offset within the file of the entry.
- * @return the materialized object. Must never be null.
- * @throws IOException
- * the method was unable to materialize the object for this
- * input pair. The usual reasons would be file corruption, file
- * not found, out of file descriptors, etc.
- */
- protected abstract V load(PackFile pack, long position) throws IOException;
-
- /**
- * Construct a Ref (SoftReference) around a cached entity.
- * v.
- */
- @SuppressWarnings("unchecked")
- protected R createRef(final PackFile pack, final long position, final V v) {
- return (R) new Refref.get() to return null.
- */
- protected void clear(final R ref) {
- // Do nothing by default.
- }
-
- /**
- * Determine if the cache is full and requires eviction of entries.
- * (PackFile,position) tuple.
- * return packHash + (int) (position >>> 4).
- * Implementors must override with a suitable hash (for example, a different
- * right shift on the position).
- *
- * @param packHash
- * hash code for the file being accessed.
- * @param position
- * position within the file being accessed.
- * @return a reasonable hash code mixing the two values.
- */
- protected abstract int hash(int packHash, long position);
-
- private int slot(final PackFile pack, final long position) {
- return (hash(pack.hash, position) >>> 1) % tableSize;
- }
-
- private Lock lock(final PackFile pack, final long position) {
- return locks[(hash(pack.hash, position) >>> 1) % locks.length];
- }
-
- private static (PackFile,position) key tuple.
+ * This is ensured by an array of locks, with the tuple hashed to a lock
+ * instance.
+ *
+ *
+ * (PackFile,position) tuple.load() invocation there is exactly one
+ * {@link #createRef(PackFile, long, ByteWindow)} invocation to wrap a
+ * SoftReference around the cached entity.createRef() there will be
+ * exactly one call to {@link #clear(Ref)} to cleanup any resources associated
+ * with the (now expired) cached entity.pack of the object.
+ * @return the object reference.
+ * @throws IOException
+ * the object reference was not in the cache and could not be
+ * obtained by {@link #load(PackFile, long)}.
+ */
+ private ByteWindow getOrLoad(final PackFile pack, final long position)
+ throws IOException {
+ final int slot = slot(pack, position);
+ final Entry e1 = table.get(slot);
+ ByteWindow v = scan(e1, pack, position);
+ if (v != null)
+ return v;
- WindowRef(final PackFile pack, final long position, final ByteWindow v,
- final ReferenceQueue