[SPARK-9460] Fix prefix generation for UTF8String.

Previously we could be getting garbage data if the number of bytes is 0, or on JVMs that are 4 byte aligned, or when compressedoops is on.

Author: Reynold Xin <rxin@databricks.com>

Closes #7789 from rxin/utf8string and squashes the following commits:

86ffa3e [Reynold Xin] Mask out data outside of valid range.
4d647ed [Reynold Xin] Mask out data.
c6e8794 [Reynold Xin] [SPARK-9460] Fix prefix generation for UTF8String.

1 files changed, 8 insertions, 0 deletions

diff --git a/unsafe/src/test/java/org/apache/spark/unsafe/types/UTF8StringSuite.java b/unsafe/src/test/java/org/apache/spark/unsafe/types/UTF8StringSuite.java
index 42e09e435a..f2cc19ca6b 100644
--- a/unsafe/src/test/java/org/apache/spark/unsafe/types/UTF8StringSuite.java
+++ b/unsafe/src/test/java/org/apache/spark/unsafe/types/UTF8StringSuite.java
@@ -71,6 +71,14 @@ public class UTF8StringSuite {
       fromString("abbbbbbbbbbbasdf").getPrefix() - fromString("bbbbbbbbbbbbasdf").getPrefix() < 0);
     assertTrue(fromString("").getPrefix() - fromString("a").getPrefix() < 0);
     assertTrue(fromString("你好").getPrefix() - fromString("世界").getPrefix() > 0);
+
+    byte[] buf1 = {1, 2, 3, 4, 5, 6, 7, 8, 9};
+    byte[] buf2 = {1, 2, 3};
+    UTF8String str1 = UTF8String.fromBytes(buf1, 0, 3);
+    UTF8String str2 = UTF8String.fromBytes(buf1, 0, 8);
+    UTF8String str3 = UTF8String.fromBytes(buf2);
+    assertTrue(str1.getPrefix() - str2.getPrefix() < 0);
+    assertEquals(str1.getPrefix(), str3.getPrefix());
   }
 
   @Test