Handle character elements mixed text/node elements (#416)

* Exploratory change to handle _first_ character child as _VALUE in mixed text/node children

* Different take: ignore character elements

src/main/scala/com/databricks/spark/xml/parsers/StaxXmlParser.scala

@@ -155,10 +155,23 @@ private[xml] object StaxXmlParser extends Serializable {
         // For `ArrayType`, it needs to return the type of element. The values are merged later.
         convertTo(c.getData, st, options)
       case (c: Characters, st: StructType) =>
-        // This case can be happen when current data type is inferred as `StructType`
-        // due to `valueTag` for elements having attributes but no child.
-        val dt = st.filter(_.name == options.valueTag).head.dataType
-        convertTo(c.getData, dt, options)
+        // If a value tag is present, this can be an attribute-only element whose values is in that
+        // value tag field. Or, it can be a mixed-type element with both some character elements
+        // and other complex structure. Character elements are ignored.
+        val attributesOnly = st.fields.forall { f =>
+          f.name == options.valueTag || f.name.startsWith(options.attributePrefix)
+        }
+        if (attributesOnly) {
+          // If everything else is an attribute column, there's no complex structure.
+          // Just return the value of the character element
+          val dt = st.find(_.name == options.valueTag).get.dataType
+          convertTo(c.getData, dt, options)
+        } else {
+          // Otherwise, ignore this character element, and continue parsing the following complex
+          // structure
+          parser.next
+          convertObject(parser, st, options)
+        }
       case (c: Characters, dt: DataType) =>
         convertTo(c.getData, dt, options)
       case (e: XMLEvent, dt: DataType) =>

src/main/scala/com/databricks/spark/xml/util/InferSchema.scala

@@ -163,8 +163,19 @@ private[xml] object InferSchema {
           case _ => inferField(parser, options)
         }
       case c: Characters if !c.isWhiteSpace =>
-        // This means data exists
-        inferFrom(c.getData, options)
+        // This could be the characters of a character-only element, or could have mixed
+        // characters and other complex structure
+        val characterType = inferFrom(c.getData, options)
+        parser.nextEvent()
+        parser.peek match {
+          case _: StartElement =>
+            // Some more elements follow; so ignore the characters.
+            // Use the schema of the rest
+            inferObject(parser, options).asInstanceOf[StructType]
+          case _ =>
+            // That's all, just the character-only body; use that as the type
+            characterType
+        }
       case e: XMLEvent =>
         throw new IllegalArgumentException(s"Failed to parse data with unexpected event $e")
     }

src/test/resources/mixed_children.xml

@@ -0,0 +1,5 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<root>
+  <foo> issue <bar> lorem </bar> text ignored </foo>
+  <missing> ipsum </missing>
+</root>

src/test/resources/mixed_children_2.xml

@@ -0,0 +1,5 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<root>
+  <foo> 3.0 <bar> lorem </bar> text ignored <baz><bing>2</bing></baz> text ignored </foo>
+  <missing> ipsum </missing>
+</root>

src/test/scala/com/databricks/spark/xml/XmlSuite.scala

@@ -75,6 +75,8 @@ final class XmlSuite extends FunSuite with BeforeAndAfterAll {
   private val selfClosingTag = resDir + "self-closing-tag.xml"
   private val textColumn = resDir + "textColumn.xml"
   private val processing = resDir + "processing.xml"
+  private val mixedChildren = resDir + "mixed_children.xml"
+  private val mixedChildren2 = resDir + "mixed_children_2.xml"
 
   private val booksTag = "book"
   private val booksRootTag = "books"
@@ -1055,4 +1057,24 @@ final class XmlSuite extends FunSuite with BeforeAndAfterAll {
     assert(processingDF.count() === 1)
   }
 
+  test("test mixed text and element children") {
+    val mixedDF = spark.read
+      .option("rowTag", "root")
+      .option("inferSchema", true)
+      .xml(mixedChildren)
+    val mixedRow = mixedDF.head()
+    assert(mixedRow.getAs[Row](0).toSeq === Seq(" lorem "))
+    assert(mixedRow.getString(1) === " ipsum ")
+  }
+
+  test("test mixed text and complex element children") {
+    val mixedDF = spark.read
+      .option("rowTag", "root")
+      .option("inferSchema", true)
+      .xml(mixedChildren2)
+    assert(mixedDF.select("foo.bar").head().getString(0) === " lorem ")
+    assert(mixedDF.select("foo.baz.bing").head().getLong(0) === 2)
+    assert(mixedDF.select("missing").head().getString(0) === " ipsum ")
+  }
+
 }