dtenedor commented on code in PR #41750:
URL: https://github.com/apache/spark/pull/41750#discussion_r1244640645
##########
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/parser/AstBuilder.scala:
##########
@@ -1539,6 +1539,31 @@ class AstBuilder extends
SqlBaseParserBaseVisitor[AnyRef] with SQLConfHelper wit
RelationTimeTravel(plan, timestamp, version)
}
+ /**
+ * Create a relation argument for a table-valued function argument.
+ */
+ override def visitFunctionTableRelationArgument(
+ ctx: FunctionTableRelationArgumentContext): Expression = withOrigin(ctx)
{
+ val p = if (ctx.identifierReference != null) {
Review Comment:
optional: you could use `.map` and do:
```
val p = Option(ctx.identifierReference).map { r =>
createUnresolvedRelation(r)
}.getOrElse {
plan(ctx.query)
}
```
Same for L1560 below.
##########
sql/catalyst/src/main/antlr4/org/apache/spark/sql/catalyst/parser/SqlBaseParser.g4:
##########
@@ -788,8 +788,18 @@ inlineTable
: VALUES expression (COMMA expression)* tableAlias
;
+functionTableRelationArgument
Review Comment:
I would name these `functionTableSubqueryArgument` and
`functionTableReferenceArgument`, respectively, to help differentiate between
the two (since both are relations).
##########
python/pyspark/sql/tests/test_udtf.py:
##########
@@ -397,6 +397,138 @@ def test_udtf(a: int):
with self.assertRaisesRegex(TypeError, err_msg):
udtf(test_udtf, returnType="a: int")
+ def test_udtf_with_table_argument_query(self):
+ class TestUDTF:
+ def eval(self, row: Row):
+ if row["id"] > 5:
+ yield row["id"],
+
+ func = udtf(TestUDTF, returnType="a: int")
+ self.spark.udtf.register("test_udtf", func)
+ self.assertEqual(
+ self.spark.sql("SELECT * FROM test_udtf(TABLE (SELECT id FROM
range(0, 8)))").collect(),
+ [Row(a=6), Row(a=7)],
+ )
+
+ def test_udtf_with_table_argument_identifier(self):
+ class TestUDTF:
+ def eval(self, row: Row):
+ if row["id"] > 5:
+ yield row["id"],
+
+ func = udtf(TestUDTF, returnType="a: int")
+ self.spark.udtf.register("test_udtf", func)
+
+ with self.tempView("v"):
+ self.spark.sql("CREATE OR REPLACE TEMPORARY VIEW v as SELECT id
FROM range(0, 8)")
+ self.assertEqual(
+ self.spark.sql("SELECT * FROM test_udtf(TABLE v)").collect(),
+ [Row(a=6), Row(a=7)],
+ )
+
+ def test_udtf_with_table_argument_unknown_identifier(self):
+ class TestUDTF:
+ def eval(self, row: Row):
+ if row["id"] > 5:
+ yield row["id"],
+
+ func = udtf(TestUDTF, returnType="a: int")
+ self.spark.udtf.register("test_udtf", func)
+
+ with self.assertRaisesRegex(AnalysisException,
"TABLE_OR_VIEW_NOT_FOUND"):
+ self.spark.sql("SELECT * FROM test_udtf(TABLE v)").collect()
+
+ def test_udtf_with_table_argument_malformed_query(self):
+ class TestUDTF:
+ def eval(self, row: Row):
+ if row["id"] > 5:
+ yield row["id"],
+
+ func = udtf(TestUDTF, returnType="a: int")
+ self.spark.udtf.register("test_udtf", func)
+
+ with self.assertRaisesRegex(AnalysisException,
"TABLE_OR_VIEW_NOT_FOUND"):
+ self.spark.sql("SELECT * FROM test_udtf(TABLE (SELECT * FROM
v))").collect()
+
+ def test_udtf_with_table_argument_cte_inside(self):
+ class TestUDTF:
+ def eval(self, row: Row):
+ if row["id"] > 5:
+ yield row["id"],
+
+ func = udtf(TestUDTF, returnType="a: int")
+ self.spark.udtf.register("test_udtf", func)
+ self.assertEqual(
+ self.spark.sql(
+ """
+ SELECT * FROM test_udtf(TABLE (
+ WITH t AS (
+ SELECT id FROM range(0, 8)
+ )
+ SELECT * FROM t
+ ))
+ """
+ ).collect(),
+ [Row(a=6), Row(a=7)],
+ )
+
+ def test_udtf_with_table_argument_cte_outside(self):
+ class TestUDTF:
+ def eval(self, row: Row):
+ if row["id"] > 5:
+ yield row["id"],
+
+ func = udtf(TestUDTF, returnType="a: int")
+ self.spark.udtf.register("test_udtf", func)
+ self.assertEqual(
+ self.spark.sql(
+ """
+ WITH t AS (
+ SELECT id FROM range(0, 8)
+ )
+ SELECT * FROM test_udtf(TABLE (SELECT id FROM t))
+ """
+ ).collect(),
+ [Row(a=6), Row(a=7)],
+ )
+
+ self.assertEqual(
+ self.spark.sql(
+ """
+ WITH t AS (
+ SELECT id FROM range(0, 8)
+ )
+ SELECT * FROM test_udtf(TABLE t)
Review Comment:
This is awesome! Some more test ideas:
* Should we add one more test with a lateral join against a UDTF that takes
a relation argument like this, and the final query can SELECT from both the
table on the left side of the lateral join and the UDTF output?
* Also, can we have a test where we pass both a constant scalar argument
before the TABLE argument, e.g. `SELECT * FROM test_udtf(42, TABLE (SELECT id
FROM range(0, 8))` where the UDTF consumes two integers?
##########
python/pyspark/sql/tests/test_udtf.py:
##########
@@ -397,6 +397,138 @@ def test_udtf(a: int):
with self.assertRaisesRegex(TypeError, err_msg):
udtf(test_udtf, returnType="a: int")
+ def test_udtf_with_table_argument_query(self):
+ class TestUDTF:
+ def eval(self, row: Row):
+ if row["id"] > 5:
+ yield row["id"],
+
+ func = udtf(TestUDTF, returnType="a: int")
+ self.spark.udtf.register("test_udtf", func)
+ self.assertEqual(
+ self.spark.sql("SELECT * FROM test_udtf(TABLE (SELECT id FROM
range(0, 8)))").collect(),
+ [Row(a=6), Row(a=7)],
+ )
+
+ def test_udtf_with_table_argument_identifier(self):
+ class TestUDTF:
+ def eval(self, row: Row):
+ if row["id"] > 5:
+ yield row["id"],
+
+ func = udtf(TestUDTF, returnType="a: int")
+ self.spark.udtf.register("test_udtf", func)
+
+ with self.tempView("v"):
+ self.spark.sql("CREATE OR REPLACE TEMPORARY VIEW v as SELECT id
FROM range(0, 8)")
+ self.assertEqual(
+ self.spark.sql("SELECT * FROM test_udtf(TABLE v)").collect(),
+ [Row(a=6), Row(a=7)],
+ )
+
+ def test_udtf_with_table_argument_unknown_identifier(self):
+ class TestUDTF:
+ def eval(self, row: Row):
+ if row["id"] > 5:
+ yield row["id"],
+
+ func = udtf(TestUDTF, returnType="a: int")
+ self.spark.udtf.register("test_udtf", func)
+
+ with self.assertRaisesRegex(AnalysisException,
"TABLE_OR_VIEW_NOT_FOUND"):
+ self.spark.sql("SELECT * FROM test_udtf(TABLE v)").collect()
+
+ def test_udtf_with_table_argument_malformed_query(self):
+ class TestUDTF:
+ def eval(self, row: Row):
+ if row["id"] > 5:
+ yield row["id"],
+
+ func = udtf(TestUDTF, returnType="a: int")
+ self.spark.udtf.register("test_udtf", func)
+
+ with self.assertRaisesRegex(AnalysisException,
"TABLE_OR_VIEW_NOT_FOUND"):
+ self.spark.sql("SELECT * FROM test_udtf(TABLE (SELECT * FROM
v))").collect()
+
+ def test_udtf_with_table_argument_cte_inside(self):
+ class TestUDTF:
+ def eval(self, row: Row):
+ if row["id"] > 5:
+ yield row["id"],
+
+ func = udtf(TestUDTF, returnType="a: int")
+ self.spark.udtf.register("test_udtf", func)
+ self.assertEqual(
+ self.spark.sql(
+ """
+ SELECT * FROM test_udtf(TABLE (
+ WITH t AS (
+ SELECT id FROM range(0, 8)
+ )
+ SELECT * FROM t
+ ))
+ """
+ ).collect(),
+ [Row(a=6), Row(a=7)],
+ )
+
+ def test_udtf_with_table_argument_cte_outside(self):
+ class TestUDTF:
+ def eval(self, row: Row):
+ if row["id"] > 5:
+ yield row["id"],
+
+ func = udtf(TestUDTF, returnType="a: int")
+ self.spark.udtf.register("test_udtf", func)
+ self.assertEqual(
+ self.spark.sql(
+ """
+ WITH t AS (
+ SELECT id FROM range(0, 8)
+ )
+ SELECT * FROM test_udtf(TABLE (SELECT id FROM t))
+ """
+ ).collect(),
+ [Row(a=6), Row(a=7)],
+ )
+
+ self.assertEqual(
+ self.spark.sql(
+ """
+ WITH t AS (
+ SELECT id FROM range(0, 8)
+ )
+ SELECT * FROM test_udtf(TABLE t)
+ """
+ ).collect(),
+ [Row(a=6), Row(a=7)],
+ )
+
+ def test_udtf_with_table_argument_multiple(self):
+ class TestUDTF:
+ def eval(self, a: Row, b: Row):
+ yield a[0], b[0]
+
+ func = udtf(TestUDTF, returnType="a: int, b: int")
+ self.spark.udtf.register("test_udtf", func)
+ self.assertEqual(
+ self.spark.sql(
+ """
+ SELECT * FROM test_udtf(
+ TABLE (SELECT id FROM range(0, 2)),
+ TABLE (SELECT id FROM range(0, 3)))
Review Comment:
This is a bit weird. I guess we haven't defined any restriction that the
UDTF can't take two or more TABLE arguments, and receive the cartesian product
of all the rows of these tables. Can we create a SQL conf to control this
behavior, and test the conf in these tests? I would imagine the conf would be
false by default to restrict the number of TABLE arguments to at most one, and
then the behavior would match the SQL standard. But then Spark users may decide
which behavior they prefer as they wish.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
