maropu commented on a change in pull request #31349: URL: https://github.com/apache/spark/pull/31349#discussion_r565757032
########## File path: sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/AnsiTypeCoercion.scala ########## @@ -0,0 +1,305 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql.catalyst.analysis + +import scala.annotation.tailrec + +import org.apache.spark.sql.catalyst.analysis.TypeCoercion.numericPrecedence +import org.apache.spark.sql.catalyst.expressions._ +import org.apache.spark.sql.catalyst.expressions.aggregate._ +import org.apache.spark.sql.catalyst.plans.logical.LogicalPlan +import org.apache.spark.sql.catalyst.rules.Rule +import org.apache.spark.sql.types._ + +/** + * In Spark ANSI mode, the type coercion rules are based on the type precedence lists of the input + * data types. + * As per the section "Type precedence list determination" of "ISO/IEC 9075-2:2011 + * Information technology - Database languages - SQL - Part 2: Foundation (SQL/Foundation)", + * the type precedence lists of primitive data types are as following: + * * Byte: Byte, Short, Int, Long, Decimal, Float, Double + * * Short: Short, Int, Long, Decimal, Float, Double + * * Int: Int, Long, Decimal, Float, Double + * * Long: Long, Decimal, Float, Double + * * Decimal: Any wider Numeric type + * * Float: Float, Double + * * Double: Double + * * String: String + * * Date: Date, Timestamp + * * Timestamp: Timestamp + * * Binary: Binary + * * Boolean: Boolean + * * Interval: Interval + * As for complex data types, Spark will determine the precedent list recursively based on their + * sub-types. + * + * With the definition of type precedent list, the general type coercion rules are as following: + * * Data type S is allowed to be implicitly cast as type T iff T is in the precedence list of S + * * Comparison is allowed iff the data type precedence list of both sides has at least one common + * element. When evaluating the comparison, Spark casts both sides as the tightest common data + * type of their precedent lists. + * * There should be at least one common data type among all the children's precedence lists for + * the following operators. The data type of the operator is the tightest common precedent + * data type. + * * In + * * Except(odd) + * * Intersect + * * Greatest + * * Least + * * Union + * * If + * * CaseWhen + * * CreateArray + * * Array Concat + * * Sequence + * * MapConcat + * * CreateMap + * * For complex types (struct, array, map), Spark recursively looks into the element type and + * applies the rules above. If the element nullability is converted from true to false, add + * runtime null check to the elements. + * Note: this new type coercion system will allow implicit converting String type literals as other + * primitive types, in case of breaking too many existing Spark SQL queries. + */ +object AnsiTypeCoercion extends TypeCoercionBase { + override def typeCoercionRules: List[Rule[LogicalPlan]] = + InConversion :: + WidenSetOperationTypes :: + PromoteStringLiterals :: + DecimalPrecision :: + FunctionArgumentConversion :: + ConcatCoercion :: + MapZipWithCoercion :: + EltCoercion :: + CaseWhenCoercion :: + IfCoercion :: + StackCoercion :: + Division :: + IntegralDivision :: + ImplicitTypeCasts :: + DateTimeOperations :: + WindowFrameCoercion :: + StringLiteralCoercion :: + Nil + + /** + * Find the tightest common type of two types that might be used in a binary expression. + */ + override def findTightestCommonType(t1: DataType, t2: DataType): Option[DataType] = { + (t1, t2) match { + case (t1, t2) if t1 == t2 => Some(t1) + case (NullType, t1) => Some(t1) + case (t1, NullType) => Some(t1) + + case (t1: NumericType, t2: NumericType) => + findTightestCommonNumericType(t1, t2) + + case (_: TimestampType, _: DateType) | (_: DateType, _: TimestampType) => + Some(TimestampType) + + case (t1, t2) => findTypeForComplex(t1, t2, findTightestCommonType) + } + } + + @tailrec Review comment: Just out of curiosity; the compiler can correctly handle this tail recursions? (not sure about if this annotation is really needed here) ########## File path: sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala ########## @@ -219,6 +219,12 @@ class Analyzer(override val catalogManager: CatalogManager) */ val postHocResolutionRules: Seq[Rule[LogicalPlan]] = Nil + def typeCoercionRules(): List[Rule[LogicalPlan]] = if (conf.ansiEnabled) { Review comment: `private` ########## File path: sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/AnsiTypeCoercion.scala ########## @@ -0,0 +1,305 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql.catalyst.analysis + +import scala.annotation.tailrec + +import org.apache.spark.sql.catalyst.analysis.TypeCoercion.numericPrecedence +import org.apache.spark.sql.catalyst.expressions._ +import org.apache.spark.sql.catalyst.expressions.aggregate._ +import org.apache.spark.sql.catalyst.plans.logical.LogicalPlan +import org.apache.spark.sql.catalyst.rules.Rule +import org.apache.spark.sql.types._ + +/** + * In Spark ANSI mode, the type coercion rules are based on the type precedence lists of the input + * data types. + * As per the section "Type precedence list determination" of "ISO/IEC 9075-2:2011 + * Information technology - Database languages - SQL - Part 2: Foundation (SQL/Foundation)", + * the type precedence lists of primitive data types are as following: + * * Byte: Byte, Short, Int, Long, Decimal, Float, Double + * * Short: Short, Int, Long, Decimal, Float, Double + * * Int: Int, Long, Decimal, Float, Double + * * Long: Long, Decimal, Float, Double + * * Decimal: Any wider Numeric type + * * Float: Float, Double + * * Double: Double + * * String: String + * * Date: Date, Timestamp + * * Timestamp: Timestamp + * * Binary: Binary + * * Boolean: Boolean + * * Interval: Interval + * As for complex data types, Spark will determine the precedent list recursively based on their + * sub-types. + * + * With the definition of type precedent list, the general type coercion rules are as following: + * * Data type S is allowed to be implicitly cast as type T iff T is in the precedence list of S + * * Comparison is allowed iff the data type precedence list of both sides has at least one common + * element. When evaluating the comparison, Spark casts both sides as the tightest common data + * type of their precedent lists. + * * There should be at least one common data type among all the children's precedence lists for + * the following operators. The data type of the operator is the tightest common precedent + * data type. + * * In + * * Except(odd) + * * Intersect + * * Greatest + * * Least + * * Union + * * If + * * CaseWhen + * * CreateArray + * * Array Concat + * * Sequence + * * MapConcat + * * CreateMap + * * For complex types (struct, array, map), Spark recursively looks into the element type and + * applies the rules above. If the element nullability is converted from true to false, add + * runtime null check to the elements. + * Note: this new type coercion system will allow implicit converting String type literals as other + * primitive types, in case of breaking too many existing Spark SQL queries. + */ +object AnsiTypeCoercion extends TypeCoercionBase { + override def typeCoercionRules: List[Rule[LogicalPlan]] = Review comment: How about defining the common set of the rules in `TypeCoercionBase`? What's differences of the ansi/non-ansi are `BooleanEquality` and `EltCoercion`? ########## File path: sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/AnsiTypeCoercion.scala ########## @@ -0,0 +1,305 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql.catalyst.analysis + +import scala.annotation.tailrec + +import org.apache.spark.sql.catalyst.analysis.TypeCoercion.numericPrecedence +import org.apache.spark.sql.catalyst.expressions._ +import org.apache.spark.sql.catalyst.expressions.aggregate._ +import org.apache.spark.sql.catalyst.plans.logical.LogicalPlan +import org.apache.spark.sql.catalyst.rules.Rule +import org.apache.spark.sql.types._ + +/** + * In Spark ANSI mode, the type coercion rules are based on the type precedence lists of the input + * data types. + * As per the section "Type precedence list determination" of "ISO/IEC 9075-2:2011 + * Information technology - Database languages - SQL - Part 2: Foundation (SQL/Foundation)", + * the type precedence lists of primitive data types are as following: + * * Byte: Byte, Short, Int, Long, Decimal, Float, Double + * * Short: Short, Int, Long, Decimal, Float, Double + * * Int: Int, Long, Decimal, Float, Double + * * Long: Long, Decimal, Float, Double + * * Decimal: Any wider Numeric type + * * Float: Float, Double + * * Double: Double + * * String: String + * * Date: Date, Timestamp + * * Timestamp: Timestamp + * * Binary: Binary + * * Boolean: Boolean + * * Interval: Interval + * As for complex data types, Spark will determine the precedent list recursively based on their + * sub-types. + * + * With the definition of type precedent list, the general type coercion rules are as following: + * * Data type S is allowed to be implicitly cast as type T iff T is in the precedence list of S + * * Comparison is allowed iff the data type precedence list of both sides has at least one common + * element. When evaluating the comparison, Spark casts both sides as the tightest common data + * type of their precedent lists. + * * There should be at least one common data type among all the children's precedence lists for + * the following operators. The data type of the operator is the tightest common precedent + * data type. + * * In + * * Except(odd) + * * Intersect + * * Greatest + * * Least + * * Union + * * If + * * CaseWhen + * * CreateArray + * * Array Concat + * * Sequence + * * MapConcat + * * CreateMap + * * For complex types (struct, array, map), Spark recursively looks into the element type and + * applies the rules above. If the element nullability is converted from true to false, add + * runtime null check to the elements. + * Note: this new type coercion system will allow implicit converting String type literals as other + * primitive types, in case of breaking too many existing Spark SQL queries. + */ +object AnsiTypeCoercion extends TypeCoercionBase { + override def typeCoercionRules: List[Rule[LogicalPlan]] = + InConversion :: + WidenSetOperationTypes :: + PromoteStringLiterals :: + DecimalPrecision :: + FunctionArgumentConversion :: + ConcatCoercion :: + MapZipWithCoercion :: + EltCoercion :: + CaseWhenCoercion :: + IfCoercion :: + StackCoercion :: + Division :: + IntegralDivision :: + ImplicitTypeCasts :: + DateTimeOperations :: + WindowFrameCoercion :: + StringLiteralCoercion :: + Nil + + /** + * Find the tightest common type of two types that might be used in a binary expression. + */ + override def findTightestCommonType(t1: DataType, t2: DataType): Option[DataType] = { + (t1, t2) match { + case (t1, t2) if t1 == t2 => Some(t1) + case (NullType, t1) => Some(t1) + case (t1, NullType) => Some(t1) + + case (t1: NumericType, t2: NumericType) => + findTightestCommonNumericType(t1, t2) + + case (_: TimestampType, _: DateType) | (_: DateType, _: TimestampType) => + Some(TimestampType) + + case (t1, t2) => findTypeForComplex(t1, t2, findTightestCommonType) + } + } + + @tailrec + private def findTightestCommonNumericType(t1: DataType, t2: DataType): Option[DataType] = { + (t1, t2) match { + case (i: IntegralType, d: DecimalType) => + if (d.isWiderThan(i)) { + Some(t2) + } else { + findTightestCommonNumericType(DecimalType.forType(i), d) + } + + case (t1: DecimalType, t2: IntegralType) => + findTightestCommonNumericType(t2, t1) + + case (t1: DecimalType, t2: DecimalType) => + Some(DecimalPrecision.widerDecimalType(t1, t2)) + + case (_: FractionalType, _: DecimalType) | (_: DecimalType, _: FractionalType) => + Some(DoubleType) + + // Promote numeric types to the highest of the two + case (t1: NumericType, t2: NumericType) + if !t1.isInstanceOf[DecimalType] && !t2.isInstanceOf[DecimalType] => + val index = numericPrecedence.lastIndexWhere(t => t == t1 || t == t2) + val widerType = numericPrecedence(index) + if (widerType == FloatType) { + // If the input type is an Integral type and a Float type, simply return Double type as + // the tightest common type to avoid potential precision loss on converting the Integral + // type as Float type. + Some(DoubleType) + } else { + Some(widerType) + } Review comment: Looks we need a default case (`case (t1, t2) =>`) for a newer scala compiler? ########## File path: sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/AnsiTypeCoercion.scala ########## @@ -0,0 +1,305 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql.catalyst.analysis + +import scala.annotation.tailrec + +import org.apache.spark.sql.catalyst.analysis.TypeCoercion.numericPrecedence +import org.apache.spark.sql.catalyst.expressions._ +import org.apache.spark.sql.catalyst.expressions.aggregate._ +import org.apache.spark.sql.catalyst.plans.logical.LogicalPlan +import org.apache.spark.sql.catalyst.rules.Rule +import org.apache.spark.sql.types._ + +/** + * In Spark ANSI mode, the type coercion rules are based on the type precedence lists of the input + * data types. + * As per the section "Type precedence list determination" of "ISO/IEC 9075-2:2011 + * Information technology - Database languages - SQL - Part 2: Foundation (SQL/Foundation)", + * the type precedence lists of primitive data types are as following: + * * Byte: Byte, Short, Int, Long, Decimal, Float, Double + * * Short: Short, Int, Long, Decimal, Float, Double + * * Int: Int, Long, Decimal, Float, Double + * * Long: Long, Decimal, Float, Double + * * Decimal: Any wider Numeric type + * * Float: Float, Double + * * Double: Double + * * String: String + * * Date: Date, Timestamp + * * Timestamp: Timestamp + * * Binary: Binary + * * Boolean: Boolean + * * Interval: Interval + * As for complex data types, Spark will determine the precedent list recursively based on their + * sub-types. + * + * With the definition of type precedent list, the general type coercion rules are as following: + * * Data type S is allowed to be implicitly cast as type T iff T is in the precedence list of S + * * Comparison is allowed iff the data type precedence list of both sides has at least one common + * element. When evaluating the comparison, Spark casts both sides as the tightest common data + * type of their precedent lists. + * * There should be at least one common data type among all the children's precedence lists for + * the following operators. The data type of the operator is the tightest common precedent + * data type. + * * In + * * Except(odd) + * * Intersect + * * Greatest + * * Least + * * Union + * * If + * * CaseWhen + * * CreateArray + * * Array Concat + * * Sequence + * * MapConcat + * * CreateMap + * * For complex types (struct, array, map), Spark recursively looks into the element type and + * applies the rules above. If the element nullability is converted from true to false, add + * runtime null check to the elements. + * Note: this new type coercion system will allow implicit converting String type literals as other + * primitive types, in case of breaking too many existing Spark SQL queries. Review comment: The standard says something about this behaviour, too? ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
