[SPARK-11836][SQL] udf/cast should not create new SQLContext

They should use the existing SQLContext.

Author: Davies Liu <davies@databricks.com>

Closes #9914 from davies/create_udf.

python/pyspark/sql/column.py

@@ -346,9 +346,10 @@ class Column(object):
         if isinstance(dataType, basestring):
             jc = self._jc.cast(dataType)
         elif isinstance(dataType, DataType):
-            sc = SparkContext._active_spark_context
-            ssql_ctx = sc._jvm.SQLContext(sc._jsc.sc())
-            jdt = ssql_ctx.parseDataType(dataType.json())
+            from pyspark.sql import SQLContext
+            sc = SparkContext.getOrCreate()
+            ctx = SQLContext.getOrCreate(sc)
+            jdt = ctx._ssql_ctx.parseDataType(dataType.json())
             jc = self._jc.cast(jdt)
         else:
             raise TypeError("unexpected type: %s" % type(dataType))

python/pyspark/sql/functions.py

@@ -1457,14 +1457,15 @@ class UserDefinedFunction(object):
         self._judf = self._create_judf(name)
 
     def _create_judf(self, name):
+        from pyspark.sql import SQLContext
         f, returnType = self.func, self.returnType  # put them in closure `func`
         func = lambda _, it: map(lambda x: returnType.toInternal(f(*x)), it)
         ser = AutoBatchedSerializer(PickleSerializer())
         command = (func, None, ser, ser)
-        sc = SparkContext._active_spark_context
+        sc = SparkContext.getOrCreate()
         pickled_command, broadcast_vars, env, includes = _prepare_for_python_RDD(sc, command, self)
-        ssql_ctx = sc._jvm.SQLContext(sc._jsc.sc())
-        jdt = ssql_ctx.parseDataType(self.returnType.json())
+        ctx = SQLContext.getOrCreate(sc)
+        jdt = ctx._ssql_ctx.parseDataType(self.returnType.json())
         if name is None:
             name = f.__name__ if hasattr(f, '__name__') else f.__class__.__name__
         judf = sc._jvm.UserDefinedPythonFunction(name, bytearray(pickled_command), env, includes,