Minimize object allocation when calling AbstractByteBuf.toString(..., Charset)

Motivation:

Calling AbstractByteBuf.toString(..., Charset) is used quite frequently by users but produce a lot of GC.

Modification:

- Use a FastThreadLocal to store the CharBuffer that are needed for decoding.
- Use internalNioBuffer(...) when possible

Result:

Less object creation / Less GC

buffer/src/main/java/io/netty/buffer/AbstractByteBuf.java

@@ -974,20 +974,7 @@ public abstract class AbstractByteBuf extends ByteBuf {
 
     @Override
     public String toString(int index, int length, Charset charset) {
-        if (length == 0) {
+        return ByteBufUtil.decodeString(this, index, length, charset);
-            return "";
-        }
-
-        ByteBuffer nioBuffer;
-        if (nioBufferCount() == 1) {
-            nioBuffer = nioBuffer(index, length);
-        } else {
-            nioBuffer = ByteBuffer.allocate(length);
-            getBytes(index, nioBuffer);
-            nioBuffer.flip();
-        }
-
-        return ByteBufUtil.decodeString(nioBuffer, charset);
     }
 
     @Override

buffer/src/main/java/io/netty/buffer/ByteBufUtil.java

@@ -18,6 +18,7 @@ package io.netty.buffer;
 import io.netty.util.CharsetUtil;
 import io.netty.util.Recycler;
 import io.netty.util.Recycler.Handle;
+import io.netty.util.concurrent.FastThreadLocal;
 import io.netty.util.internal.ObjectUtil;
 import io.netty.util.internal.PlatformDependent;
 import io.netty.util.internal.StringUtil;
@@ -42,7 +43,14 @@ import java.util.Locale;
 public final class ByteBufUtil {
 
     private static final InternalLogger logger = InternalLoggerFactory.getInstance(ByteBufUtil.class);
+    private static final FastThreadLocal<CharBuffer> CHAR_BUFFERS = new FastThreadLocal<CharBuffer>() {
+        @Override
+        protected CharBuffer initialValue() throws Exception {
+            return CharBuffer.allocate(1024);
+        }
+    };
 
+    private static final int MAX_CHAR_BUFFER_SIZE;
     private static final char[] HEXDUMP_TABLE = new char[256 * 4];
     private static final String NEWLINE = StringUtil.NEWLINE;
     private static final String[] BYTE2HEX = new String[256];
@@ -125,6 +133,9 @@ public final class ByteBufUtil {
 
         THREAD_LOCAL_BUFFER_SIZE = SystemPropertyUtil.getInt("io.netty.threadLocalDirectBufferSize", 64 * 1024);
         logger.debug("-Dio.netty.threadLocalDirectBufferSize: {}", THREAD_LOCAL_BUFFER_SIZE);
+
+        MAX_CHAR_BUFFER_SIZE = SystemPropertyUtil.getInt("io.netty.maxThreadLocalCharBufferSize", 16 * 1024);
+        logger.debug("-Dio.netty.maxThreadLocalCharBufferSize: {}", MAX_CHAR_BUFFER_SIZE);
     }
 
     /**
@@ -519,10 +530,41 @@ public final class ByteBufUtil {
         }
     }
 
-    static String decodeString(ByteBuffer src, Charset charset) {
+    static String decodeString(ByteBuf src, int readerIndex, int len, Charset charset) {
+        if (len == 0) {
+            return StringUtil.EMPTY_STRING;
+        }
         final CharsetDecoder decoder = CharsetUtil.getDecoder(charset);
-        final CharBuffer dst = CharBuffer.allocate(
+        final int maxLength = (int) ((double) len * decoder.maxCharsPerByte());
-                (int) ((double) src.remaining() * decoder.maxCharsPerByte()));
+        CharBuffer dst = CHAR_BUFFERS.get();
+        if (dst.length() < maxLength) {
+            dst = CharBuffer.allocate(maxLength);
+            if (maxLength <= MAX_CHAR_BUFFER_SIZE) {
+                CHAR_BUFFERS.set(dst);
+            }
+        } else {
+            dst.clear();
+        }
+        if (src.nioBufferCount() == 1) {
+            // Use internalNioBuffer(...) to reduce object creation.
+            decodeString(decoder, src.internalNioBuffer(readerIndex, len), dst);
+        } else {
+            // We use a heap buffer as CharsetDecoder is most likely able to use a fast-path if src and dst buffers
+            // are both backed by a byte array.
+            ByteBuf buffer = src.alloc().heapBuffer(len);
+            try {
+                buffer.writeBytes(src, readerIndex, len);
+                // Use internalNioBuffer(...) to reduce object creation.
+                decodeString(decoder, buffer.internalNioBuffer(readerIndex, len), dst);
+            } finally {
+                // Release the temporary buffer again.
+                buffer.release();
+            }
+        }
+        return dst.flip().toString();
+    }
+
+    private static void decodeString(CharsetDecoder decoder, ByteBuffer src, CharBuffer dst) {
         try {
             CoderResult cr = decoder.decode(src, dst, true);
             if (!cr.isUnderflow()) {
@@ -535,7 +577,6 @@ public final class ByteBufUtil {
         } catch (CharacterCodingException x) {
             throw new IllegalStateException(x);
         }
-        return dst.flip().toString();
     }
 
     /**

buffer/src/test/java/io/netty/buffer/ByteBufUtilTest.java

@@ -20,6 +20,8 @@ import io.netty.util.ReferenceCountUtil;
 import org.junit.Assert;
 import org.junit.Test;
 
+import java.nio.charset.Charset;
+
 public class ByteBufUtilTest {
 
     @Test
@@ -73,4 +75,20 @@ public class ByteBufUtilTest {
     private static void assertWrapped(ByteBuf buf) {
         Assert.assertTrue(buf instanceof WrappedByteBuf);
     }
+
+    @Test
+    public void testDecodeUsAscii() {
+        testDecodeString("This is a test", CharsetUtil.US_ASCII);
+    }
+
+    @Test
+    public void testDecodeUtf8() {
+        testDecodeString("Some UTF-8 like äÄ∏ŒŒ", CharsetUtil.UTF_8);
+    }
+
+    private static void testDecodeString(String text, Charset charset) {
+        ByteBuf buffer = Unpooled.copiedBuffer(text, charset);
+        Assert.assertEquals(text, ByteBufUtil.decodeString(buffer, 0, buffer.readableBytes(), charset));
+        buffer.release();
+    }
 }

microbench/src/test/java/io/netty/microbench/buffer/ByteBufUtilBenchmark.java

@@ -30,12 +30,14 @@ import org.openjdk.jmh.annotations.Warmup;
 
 
 @State(Scope.Benchmark)
-@Warmup(iterations = 10)
+@Warmup(iterations = 5)
-@Measurement(iterations = 25)
+@Measurement(iterations = 10)
 public class
     ByteBufUtilBenchmark extends AbstractMicrobenchmark {
     private ByteBuf buffer;
     private ByteBuf wrapped;
+    private ByteBuf asciiBuffer;
+    private ByteBuf utf8Buffer;
 
     private StringBuilder asciiSequence;
     private String ascii;
@@ -62,12 +64,17 @@ public class
         }
         utf8 = utf8Sequence.toString();
         asciiSequence = utf8Sequence;
+
+        asciiBuffer = Unpooled.copiedBuffer(ascii, CharsetUtil.US_ASCII);
+        utf8Buffer = Unpooled.copiedBuffer(utf8, CharsetUtil.UTF_8);
     }
 
     @TearDown
     public void tearDown() {
         buffer.release();
         wrapped.release();
+        asciiBuffer.release();
+        utf8Buffer.release();
     }
 
     @Benchmark
@@ -165,4 +172,14 @@ public class
         wrapped.resetWriterIndex();
         ByteBufUtil.writeUtf8(wrapped, utf8Sequence);
     }
+
+    @Benchmark
+    public String decodeStringAscii() {
+        return asciiBuffer.toString(CharsetUtil.US_ASCII);
+    }
+
+    @Benchmark
+    public String decodeStringUtf8() {
+        return utf8Buffer.toString(CharsetUtil.UTF_8);
+    }
 }