[SPARK-21015] Check field name is not null and empty in GenericRowWit…

[darionyaphet]

What changes were proposed in this pull request?

When we get field index from row with schema , we shoule make sure the field name is not null and empty .

[AmplabJenkins]

Can one of the admins verify this patch?

[rxin]

Why do we want this check? If the user passes in null value, it is ok if it is not found, isn't it?

[HyukjinKwon]

It looks we already throw the similar exception with what this PR proposes (not NPE) as below:

importorg.apache.spark.sql.catalyst.expressions.GenericRowWithSchemaimportorg.apache.spark.sql.types._valschema=StructType(
StructField("col1", StringType) ::StructField("col2", StringType) ::StructField("col3", IntegerType) ::Nil)
valvalues=Array("value1", "value2", 1)
valsampleRow=newGenericRowWithSchema(values, schema)

sampleRow.fieldIndex("")

java.lang.IllegalArgumentException: Field "" does not exist.
at org.apache.spark.sql.types.StructType$$anonfun$fieldIndex$1.apply(StructType.scala:292)
at org.apache.spark.sql.types.StructType$$anonfun$fieldIndex$1.apply(StructType.scala:292)
at scala.collection.MapLike$class.getOrElse(MapLike.scala:128)
at scala.collection.AbstractMap.getOrElse(Map.scala:59)
...

sampleRow.fieldIndex(null)

java.lang.IllegalArgumentException: Field "null" does not exist.
at org.apache.spark.sql.types.StructType$$anonfun$fieldIndex$1.apply(StructType.scala:292)
at org.apache.spark.sql.types.StructType$$anonfun$fieldIndex$1.apply(StructType.scala:292)
at scala.collection.MapLike$class.getOrElse(MapLike.scala:128)
at scala.collection.AbstractMap.getOrElse(Map.scala:59)
...

I think we should close this.

[darionyaphet]

@rxin@HyukjinKwon Thank you looking into this PR :D

Maybe I don't make it clearly : GenericRowWithSchema fetch field with field name and the field name shouldn't null or empty . So I add a check before schema.fieldIndex.

[gatorsmile]

How about just updating the comment of this function to explain the behavior we have now?