HADOOP-8419. Fixed GzipCode NPE reset for IBM JDK. (Yu Li via eyang)

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/trunk@1431739 13f79535-47bb-0310-9956-ffa450edef68

hadoop-common-project/hadoop-common/CHANGES.txt

@@ -148,6 +148,8 @@ Trunk (Unreleased)
 
   BUG FIXES
 
+    HADOOP-8419. Fixed GzipCode NPE reset for IBM JDK. (Yu Li via eyang)
+
     HADOOP-9041. FsUrlStreamHandlerFactory could cause an infinite loop in
     FileSystem initialization. (Yanbo Liang and Radim Kolar via llu)
 

hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/io/compress/GzipCodec.java

@@ -40,14 +40,74 @@ public class GzipCodec extends DefaultCodec {
   protected static class GzipOutputStream extends CompressorStream {
 
     private static class ResetableGZIPOutputStream extends GZIPOutputStream {
-      
+      private static final int TRAILER_SIZE = 8;
+      public static final String JVMVendor= System.getProperty("java.vendor");
+      public static final String JVMVersion= System.getProperty("java.version");
+      private static final boolean HAS_BROKEN_FINISH =
+          (JVMVendor.contains("IBM") && JVMVersion.contains("1.6.0"));
+
       public ResetableGZIPOutputStream(OutputStream out) throws IOException {
         super(out);
       }
-      
+
       public void resetState() throws IOException {
         def.reset();
       }
+
+      /**
+       * Override this method for HADOOP-8419.
+       * Override because IBM implementation calls def.end() which
+       * causes problem when reseting the stream for reuse.
+       *
+       */
+      @Override
+      public void finish() throws IOException {
+        if (HAS_BROKEN_FINISH) {
+          if (!def.finished()) {
+            def.finish();
+            while (!def.finished()) {
+              int i = def.deflate(this.buf, 0, this.buf.length);
+              if ((def.finished()) && (i <= this.buf.length - TRAILER_SIZE)) {
+                writeTrailer(this.buf, i);
+                i += TRAILER_SIZE;
+                out.write(this.buf, 0, i);
+
+                return;
+              }
+              if (i > 0) {
+                out.write(this.buf, 0, i);
+              }
+            }
+
+            byte[] arrayOfByte = new byte[TRAILER_SIZE];
+            writeTrailer(arrayOfByte, 0);
+            out.write(arrayOfByte);
+          }
+        } else {
+          super.finish();
+        }
+      }
+
+      /** re-implement for HADOOP-8419 because the relative method in jdk is invisible */
+      private void writeTrailer(byte[] paramArrayOfByte, int paramInt)
+        throws IOException {
+        writeInt((int)this.crc.getValue(), paramArrayOfByte, paramInt);
+        writeInt(this.def.getTotalIn(), paramArrayOfByte, paramInt + 4);
+      }
+
+      /** re-implement for HADOOP-8419 because the relative method in jdk is invisible */
+      private void writeInt(int paramInt1, byte[] paramArrayOfByte, int paramInt2)
+        throws IOException {
+        writeShort(paramInt1 & 0xFFFF, paramArrayOfByte, paramInt2);
+        writeShort(paramInt1 >> 16 & 0xFFFF, paramArrayOfByte, paramInt2 + 2);
+      }
+
+      /** re-implement for HADOOP-8419 because the relative method in jdk is invisible */
+      private void writeShort(int paramInt1, byte[] paramArrayOfByte, int paramInt2)
+        throws IOException {
+        paramArrayOfByte[paramInt2] = (byte)(paramInt1 & 0xFF);
+        paramArrayOfByte[(paramInt2 + 1)] = (byte)(paramInt1 >> 8 & 0xFF);
+      }
     }
 
     public GzipOutputStream(OutputStream out) throws IOException {