Merge pull request #1526 from benjchristensen/ring-buffer-queue

Restore use of SpmcArrayQueue in RxRingBuffer

Author: benjchristensen

rxjava-core/src/main/java/rx/internal/util/RxRingBuffer.java

@@ -21,6 +21,8 @@
 import rx.Subscription;
 import rx.exceptions.MissingBackpressureException;
 import rx.internal.operators.NotificationLite;
+import rx.internal.util.unsafe.SpmcArrayQueue;
+import rx.internal.util.unsafe.SpscArrayQueue;
 import rx.internal.util.unsafe.UnsafeAccess;
 
 /**
@@ -31,28 +33,16 @@ public class RxRingBuffer implements Subscription {
 
     public static RxRingBuffer getSpscInstance() {
         if (UnsafeAccess.isUnsafeAvailable()) {
-            // using SynchronizedQueue until issues are solved with SpscArrayQueue offer rejection
-            //      RxRingBufferSpmcTest.testConcurrency occasionally fails with a 
-            //      BackpressureException when using SpscArrayQueue 
-            //            return new RxRingBuffer(SPSC_POOL, SIZE); // this is the one we were trying to use
-            //            return new RxRingBuffer(new SpscArrayQueue<Object>(SIZE), SIZE);
-            // the performance of this is sufficient (actually faster in some cases)
-            return new RxRingBuffer(new SynchronizedQueue<Object>(SIZE), SIZE);
+            // TODO the SpscArrayQueue isn't ready yet so using SpmcArrayQueue for now
+            return new RxRingBuffer(SPMC_POOL, SIZE);
         } else {
             return new RxRingBuffer();
         }
     }
 
     public static RxRingBuffer getSpmcInstance() {
         if (UnsafeAccess.isUnsafeAvailable()) {
-            // using SynchronizedQueue until issues are solved with SpmcArrayQueue offer rejection
-            //      RxRingBufferSpmcTest.testConcurrency occasionally fails with a 
-            //      BackpressureException when using SpmcArrayQueue/MpmcArrayQueue
-            //            return new RxRingBuffer(SPMC_POOL, SIZE); // this is the one we were trying to use
-            //            return new RxRingBuffer(new SpmcArrayQueue<Object>(SIZE), SIZE);
-            //            return new RxRingBuffer(new MpmcArrayQueue<Object>(SIZE), SIZE);
-            // the performance of this is sufficient (actually faster in some cases)
-            return new RxRingBuffer(new SynchronizedQueue<Object>(SIZE), SIZE);
+            return new RxRingBuffer(SPMC_POOL, SIZE);
         } else {
             return new RxRingBuffer();
         }
@@ -170,6 +160,24 @@ public static RxRingBuffer getSpmcInstance() {
 
     public static final int SIZE = 1024;
 
+    private static ObjectPool<Queue<Object>> SPSC_POOL = new ObjectPool<Queue<Object>>() {
+
+        @Override
+        protected SpscArrayQueue<Object> createObject() {
+            return new SpscArrayQueue<Object>(SIZE);
+        }
+
+    };
+
+    private static ObjectPool<Queue<Object>> SPMC_POOL = new ObjectPool<Queue<Object>>() {
+
+        @Override
+        protected SpmcArrayQueue<Object> createObject() {
+            return new SpmcArrayQueue<Object>(SIZE);
+        }
+
+    };
+    
     private RxRingBuffer(Queue<Object> queue, int size) {
         this.queue = queue;
         this.pool = null;

rxjava-core/src/main/java/rx/internal/util/unsafe/SpmcArrayQueue.java

@@ -135,7 +135,15 @@ public boolean offer(final E e) {
         final long currProducerIndex = lvProducerIndex();
         final long offset = calcElementOffset(currProducerIndex);
         if (null != lvElement(lb, offset)) {
-            return false;
+            // strict check as per https://github.com/JCTools/JCTools/issues/21#issuecomment-50204120
+            int size = (int) (currProducerIndex - lvConsumerIndex());
+            if (size == capacity) {
+                return false;
+            }
+            else {
+                // spin wait for slot to clear, buggers wait freedom
+                while (null != lvElement(lb, offset));
+            }
         }
         spElement(lb, offset, e);
         // single producer, so store ordered is valid. It is also required to correctly publish the element