[SPARK-16044][SQL] input_file_name() returns empty strings in data sources based on NewHadoopRDD

core/src/main/scala/org/apache/spark/rdd/InputFileNameHolder.scala

@@ -21,7 +21,7 @@ import org.apache.spark.unsafe.types.UTF8String
 
 /**
  * This holds file names of the current Spark task. This is used in HadoopRDD,
- * FileScanRDD and InputFileName function in Spark SQL.
+ * FileScanRDD, NewHadoopRDD and InputFileName function in Spark SQL.
  */
 private[spark] object InputFileNameHolder {
   /**

core/src/main/scala/org/apache/spark/rdd/NewHadoopRDD.scala

@@ -135,6 +135,12 @@ class NewHadoopRDD[K, V](
       val inputMetrics = context.taskMetrics().inputMetrics
       val existingBytesRead = inputMetrics.bytesRead
 
+      // Sets the thread local variable for the file's name
+      split.serializableHadoopSplit.value match {
+        case fs: FileSplit => InputFileNameHolder.setInputFileName(fs.getPath.toString)
+        case _ => InputFileNameHolder.unsetInputFileName()
+      }
+
       // Find a function that will return the FileSystem bytes read by this thread. Do this before
       // creating RecordReader, because RecordReader's constructor might read some bytes
       val getBytesReadCallback: Option[() => Long] = split.serializableHadoopSplit.value match {
@@ -201,6 +207,7 @@ class NewHadoopRDD[K, V](
 
       private def close() {
         if (reader != null) {
+          InputFileNameHolder.unsetInputFileName()
           // Close the reader and release it. Note: it's very important that we don't close the
           // reader more than once, since that exposes us to MAPREDUCE-5918 when running against
           // Hadoop 1.x and older Hadoop 2.x releases. That bug can lead to non-deterministic

sql/core/src/test/scala/org/apache/spark/sql/ColumnExpressionSuite.scala

@@ -17,6 +17,8 @@
 
 package org.apache.spark.sql
 
+import org.apache.hadoop.io.{LongWritable, Text}
+import org.apache.hadoop.mapreduce.lib.input.{TextInputFormat => NewTextInputFormat}
 import org.scalatest.Matchers._
 
 import org.apache.spark.sql.catalyst.expressions.NamedExpression
@@ -592,7 +594,7 @@ class ColumnExpressionSuite extends QueryTest with SharedSQLContext {
     )
   }
 
-  test("input_file_name") {
+  test("input_file_name - FileScanRDD") {
     withTempPath { dir =>
       val data = sparkContext.parallelize(0 to 10).toDF("id")
       data.write.parquet(dir.getCanonicalPath)
@@ -604,6 +606,35 @@ class ColumnExpressionSuite extends QueryTest with SharedSQLContext {
     }
   }
 
+  test("input_file_name - HadoopRDD") {
+    withTempPath { dir =>
+      val data = sparkContext.parallelize((0 to 10).map(_.toString)).toDF()
+      data.write.text(dir.getCanonicalPath)
+      val df = spark.sparkContext.textFile(dir.getCanonicalPath).toDF()
+      val answer = df.select(input_file_name()).head.getString(0)
+      assert(answer.contains(dir.getCanonicalPath))
+
+      checkAnswer(data.select(input_file_name()).limit(1), Row(""))
+    }
+  }
+
+  test("input_file_name - NewHadoopRDD") {
+    withTempPath { dir =>
+      val data = sparkContext.parallelize((0 to 10).map(_.toString)).toDF()
+      data.write.text(dir.getCanonicalPath)
+      val rdd = spark.sparkContext.newAPIHadoopFile(
+        dir.getCanonicalPath,
+        classOf[NewTextInputFormat],
+        classOf[LongWritable],
+        classOf[Text])
+      val df = rdd.map(pair => pair._2.toString).toDF()
+      val answer = df.select(input_file_name()).head.getString(0)
+      assert(answer.contains(dir.getCanonicalPath))
+
+      checkAnswer(data.select(input_file_name()).limit(1), Row(""))
+    }
+  }
+
   test("columns can be compared") {
     assert('key.desc == 'key.desc)
     assert('key.desc != 'key.asc)
@@ -707,5 +738,4 @@ class ColumnExpressionSuite extends QueryTest with SharedSQLContext {
       testData2.select($"a".bitwiseXOR($"b").bitwiseXOR(39)),
       testData2.collect().toSeq.map(r => Row(r.getInt(0) ^ r.getInt(1) ^ 39)))
   }
-
 }