Fixed a rather bad bug in HDFS files that has been in for a while:

caching was not working because Split objects did not have a
consistent toString value

src/scala/spark/HdfsFile.scala

@@ -14,6 +14,7 @@ import org.apache.hadoop.mapred.Reporter
 @serializable class HdfsSplit(@transient s: InputSplit)
 extends Split { 
   val inputSplit = new SerializableWritable[InputSplit](s)
+  override def toString = inputSplit.toString
 }
 
 class HdfsTextFile(sc: SparkContext, path: String)

src/scala/spark/RDD.scala

@@ -198,6 +198,7 @@ extends RDD[T](prev.sparkContext) with Logging {
   
   override def iterator(split: Split): Iterator[T] = {
     val key = id + "::" + split.toString
+    logInfo("CachedRDD split key is " + key)
     val cache = CachedRDD.cache
     val loading = CachedRDD.loading
     val cachedVal = cache.get(key)