[GitHub] incubator-carbondata pull request #805: [WIP] Cast Push Down Optimization

[qiuchenjian-2]

Github user ravipesala commented on a diff in the pull request:

    https://github.com/apache/incubator-carbondata/pull/805#discussion_r112659241
 
    --- Diff: integration/spark2/src/main/scala/org/apache/spark/sql/execution/CarbonLateDecodeStrategy.scala ---
    @@ -396,43 +402,41 @@ private[sql] class CarbonLateDecodeStrategy extends SparkStrategy {
         (unrecognizedPredicates ++ unhandledPredicates, translatedFilters)
       }
     
    +
       /**
        * Tries to translate a Catalyst [[Expression]] into data source [[Filter]].
        * @return a `Some[Filter]` if the input [[Expression]] is convertible, otherwise a `None`.
        */
    -  protected[sql] def translateFilter(predicate: Expression, or: Boolean = false): Option[Filter] = {
    +  protected[sql] def translateFilter(predicate: Expression): Option[Filter] = {
         predicate match {
           case or@Or(left, right) =>
     
    -        val leftFilter = translateFilter(left, true)
    -        val rightFilter = translateFilter(right, true)
    +        val leftFilter = translateFilter(left)
    +        val rightFilter = translateFilter(right)
             if (leftFilter.isDefined && rightFilter.isDefined) {
               Some(sources.Or(leftFilter.get, rightFilter.get))
             } else {
               None
             }
     
           case And(left, right) =>
    -        val leftFilter = translateFilter(left, or)
    -        val rightFilter = translateFilter(right, or)
    -        if (or) {
    -          if (leftFilter.isDefined && rightFilter.isDefined) {
    -            (translateFilter(left) ++ translateFilter(right)).reduceOption(sources.And)
    -          } else {
    -            None
    -          }
    -        } else {
    -          (translateFilter(left) ++ translateFilter(right)).reduceOption(sources.And)
    -        }
    -
    +        (translateFilter(left) ++ translateFilter(right)).reduceOption(sources.And)
    --- End diff --
   
    Please check why this code is remmoved


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [hidden email] or file a JIRA ticket
with INFRA.
---

[qiuchenjian-2]

Github user ravipesala commented on a diff in the pull request:

    https://github.com/apache/incubator-carbondata/pull/805#discussion_r112659542
 
    --- Diff: integration/spark2/src/main/scala/org/apache/spark/sql/optimizer/CarbonFilters.scala ---
    @@ -0,0 +1,380 @@
    +/*
    + * 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.carbondata.spark
    +
    +import java.text.SimpleDateFormat
    +import java.util.Date
    +
    +import scala.collection.mutable.ArrayBuffer
    +
    +import org.apache.spark.sql.catalyst.expressions._
    +import org.apache.spark.sql.catalyst.expressions.aggregate.AggregateExpression
    +import org.apache.spark.sql.execution.CastExpressionOptimization
    +import org.apache.spark.sql.optimizer.AttributeReferenceWrapper
    +import org.apache.spark.sql.sources
    +import org.apache.spark.sql.sources.Filter
    +import org.apache.spark.sql.types._
    +
    +import org.apache.carbondata.core.constants.CarbonCommonConstants
    +import org.apache.carbondata.core.keygenerator.directdictionary.timestamp.TimeStampDirectDictionaryGenerator
    +import org.apache.carbondata.core.metadata.datatype.DataType
    +import org.apache.carbondata.core.metadata.schema.table.CarbonTable
    +import org.apache.carbondata.core.metadata.schema.table.column.CarbonColumn
    +import org.apache.carbondata.core.scan.expression.{ColumnExpression => CarbonColumnExpression, Expression => CarbonExpression, LiteralExpression => CarbonLiteralExpression}
    +import org.apache.carbondata.core.scan.expression.conditional._
    +import org.apache.carbondata.core.scan.expression.logical.{AndExpression, FalseExpression, OrExpression}
    +import org.apache.carbondata.spark.util.CarbonScalaUtil
    +
    +
    +/**
    + * All filter conversions are done here.
    + */
    +object CarbonFilters {
    +
    +
    +  /**
    +   * Converts data sources filters to carbon filter predicates.
    +   */
    +  def createCarbonFilter(schema: StructType,
    +      predicate: sources.Filter): Option[CarbonExpression] = {
    +    val dataTypeOf = schema.map(f => f.name -> f.dataType).toMap
    +
    +    def createFilter(predicate: sources.Filter): Option[CarbonExpression] = {
    +      predicate match {
    +
    +        case sources.EqualTo(name, value) =>
    +          Some(new EqualToExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, value)))
    +        case sources.Not(sources.EqualTo(name, value)) =>
    +          Some(new NotEqualsExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, value)))
    +        case sources.EqualNullSafe(name, value) =>
    +          Some(new EqualToExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, value)))
    +        case sources.Not(sources.EqualNullSafe(name, value)) =>
    +          Some(new NotEqualsExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, value)))
    +        case sources.GreaterThan(name, value) =>
    +          Some(new GreaterThanExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, value)))
    +        case sources.LessThan(name, value) =>
    +          Some(new LessThanExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, value)))
    +        case sources.GreaterThanOrEqual(name, value) =>
    +          Some(new GreaterThanEqualToExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, value)))
    +        case sources.LessThanOrEqual(name, value) =>
    +          Some(new LessThanEqualToExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, value)))
    +        case sources.In(name, values) =>
    +          Some(new InExpression(getCarbonExpression(name),
    +            new ListExpression(
    +              convertToJavaList(values.map(f => getCarbonLiteralExpression(name, f)).toList))))
    +        case sources.Not(sources.In(name, values)) =>
    +          Some(new NotInExpression(getCarbonExpression(name),
    +            new ListExpression(
    +              convertToJavaList(values.map(f => getCarbonLiteralExpression(name, f)).toList))))
    +        case sources.IsNull(name) =>
    +          Some(new EqualToExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, null), true))
    +        case sources.IsNotNull(name) =>
    +          Some(new NotEqualsExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, null), true))
    +        case sources.And(lhs, rhs) =>
    +          (createFilter(lhs) ++ createFilter(rhs)).reduceOption(new AndExpression(_, _))
    +        case sources.Or(lhs, rhs) =>
    +          for {
    +            lhsFilter <- createFilter(lhs)
    +            rhsFilter <- createFilter(rhs)
    +          } yield {
    +            new OrExpression(lhsFilter, rhsFilter)
    +          }
    +        case CastExpr(expr: Expression) =>
    +          Some(transformExpression(expr))
    +        case _ => None
    +      }
    +    }
    +
    +    def getCarbonExpression(name: String) = {
    +      new CarbonColumnExpression(name,
    +        CarbonScalaUtil.convertSparkToCarbonDataType(dataTypeOf(name)))
    +    }
    +
    +    def getCarbonLiteralExpression(name: String, value: Any): CarbonExpression = {
    +      val dataTypeOfAttribute = CarbonScalaUtil.convertSparkToCarbonDataType(dataTypeOf(name))
    +      val dataType = if (Option(value).isDefined
    +                         && dataTypeOfAttribute == DataType.STRING
    +                         && value.isInstanceOf[Double]) {
    +        DataType.DOUBLE
    +      } else {
    +        dataTypeOfAttribute
    +      }
    +      new CarbonLiteralExpression(value, dataType)
    +    }
    +
    +    createFilter(predicate)
    +  }
    +
    +
    +  // Check out which filters can be pushed down to carbon, remaining can be handled in spark layer.
    +  // Mostly dimension filters are only pushed down since it is faster in carbon.
    +  def selectFilters(filters: Seq[Expression],
    +      attrList: java.util.HashSet[AttributeReferenceWrapper],
    +      aliasMap: CarbonAliasDecoderRelation): Unit = {
    +    def translate(expr: Expression, or: Boolean = false): Option[sources.Filter] = {
    +      expr match {
    +        case or@Or(left, right) =>
    +
    +          val leftFilter = translate(left, or = true)
    +          val rightFilter = translate(right, or = true)
    +          if (leftFilter.isDefined && rightFilter.isDefined) {
    +            Some(sources.Or(leftFilter.get, rightFilter.get))
    +          } else {
    +            or.collect {
    +              case attr: AttributeReference =>
    +                attrList.add(AttributeReferenceWrapper(aliasMap.getOrElse(attr, attr)))
    +            }
    +            None
    +          }
    +        case And(left, right) =>
    +          (translate(left) ++ translate(right)).reduceOption(sources.And)
    +        case EqualTo(a: Attribute, Literal(v, t)) =>
    +          Some(sources.EqualTo(a.name, v))
    +        case EqualTo(l@Literal(v, t), a: Attribute) =>
    +          Some(sources.EqualTo(a.name, v))
    +        case c@EqualTo(Cast(a: Attribute, _), Literal(v, t)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case c@EqualTo(Literal(v, t), Cast(a: Attribute, _)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case Not(EqualTo(a: Attribute, Literal(v, t))) =>
    +          Some(sources.Not(sources.EqualTo(a.name, v)))
    +        case Not(EqualTo(Literal(v, t), a: Attribute)) =>
    +          Some(sources.Not(sources.EqualTo(a.name, v)))
    +        case c@Not(EqualTo(Cast(a: Attribute, _), Literal(v, t))) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case c@Not(EqualTo(Literal(v, t), Cast(a: Attribute, _))) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case IsNotNull(a: Attribute) =>
    +          Some(sources.IsNotNull(a.name))
    +        case IsNull(a: Attribute) =>
    +          Some(sources.IsNull(a.name))
    +        case Not(In(a: Attribute, list)) if !list.exists(!_.isInstanceOf[Literal]) =>
    +          val hSet = list.map(e => e.eval(EmptyRow))
    +          Some(sources.Not(sources.In(a.name, hSet.toArray)))
    +        case In(a: Attribute, list) if !list.exists(!_.isInstanceOf[Literal]) =>
    +          val hSet = list.map(e => e.eval(EmptyRow))
    +          Some(sources.In(a.name, hSet.toArray))
    +        case c@Not(In(Cast(a: Attribute, _), list)) if !list.exists(!_.isInstanceOf[Literal]) =>
    +          Some(CastExpr(c))
    +        case c@In(Cast(a: Attribute, _), list) if !list.exists(!_.isInstanceOf[Literal]) =>
    +            Some(CastExpr(c))
    +        case GreaterThan(a: Attribute, Literal(v, t)) =>
    +          Some(sources.GreaterThan(a.name, v))
    +        case GreaterThan(Literal(v, t), a: Attribute) =>
    +          Some(sources.LessThan(a.name, v))
    +        case c@GreaterThan(Cast(a: Attribute, _), Literal(v, t)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case c@GreaterThan(Literal(v, t), Cast(a: Attribute, _)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case LessThan(a: Attribute, Literal(v, t)) =>
    +          Some(sources.LessThan(a.name, v))
    +        case LessThan(Literal(v, t), a: Attribute) =>
    +          Some(sources.GreaterThan(a.name, v))
    +        case c@LessThan(Cast(a: Attribute, _), Literal(v, t)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case c@LessThan(Literal(v, t), Cast(a: Attribute, _)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case GreaterThanOrEqual(a: Attribute, Literal(v, t)) =>
    +          Some(sources.GreaterThanOrEqual(a.name, v))
    +        case GreaterThanOrEqual(Literal(v, t), a: Attribute) =>
    +          Some(sources.LessThanOrEqual(a.name, v))
    +        case c@GreaterThanOrEqual(Cast(a: Attribute, _), Literal(v, t)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case c@GreaterThanOrEqual(Literal(v, t), Cast(a: Attribute, _)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case LessThanOrEqual(a: Attribute, Literal(v, t)) =>
    +          Some(sources.LessThanOrEqual(a.name, v))
    +        case LessThanOrEqual(Literal(v, t), a: Attribute) =>
    +          Some(sources.GreaterThanOrEqual(a.name, v))
    +        case c@LessThanOrEqual(Cast(a: Attribute, _), Literal(v, t)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case c@LessThanOrEqual(Literal(v, t), Cast(a: Attribute, _)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case c@Cast(a: Attribute, _) =>
    +          Some(CastExpr(c))
    +        case others =>
    +          if (!or) {
    +            others.collect {
    +              case attr: AttributeReference =>
    +                attrList.add(AttributeReferenceWrapper(aliasMap.getOrElse(attr, attr)))
    +            }
    +          }
    +          None
    +      }
    +    }
    +
    +    filters.flatMap(translate(_, false)).toArray
    +  }
    +
    +  def transformExpression(expr: Expression): CarbonExpression = {
    +    expr match {
    +      case Or(left, right)
    +        if (isCarbonSupportedDataTypes(left) && isCarbonSupportedDataTypes(right)) => new
    --- End diff --
   
    remove unnecessary extra parenthesis in if conditions for all cases.


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [hidden email] or file a JIRA ticket
with INFRA.
---

[qiuchenjian-2]

Github user ravipesala commented on a diff in the pull request:

    https://github.com/apache/incubator-carbondata/pull/805#discussion_r112659832
 
    --- Diff: integration/spark2/src/main/scala/org/apache/spark/sql/execution/CastExpressionOptimization.scala ---
    @@ -0,0 +1,387 @@
    +/*
    + * 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.execution
    +
    +import java.text.{ParseException, SimpleDateFormat}
    +import java.util
    +import java.util.{Date, Locale, TimeZone}
    +
    +import scala.collection.JavaConverters._
    +
    +import org.apache.spark.sql.catalyst.expressions.{Attribute, Cast, EmptyRow, EqualTo, Expression, GreaterThan, GreaterThanOrEqual, In, LessThan, LessThanOrEqual, Literal, Not}
    +import org.apache.spark.sql.sources
    +import org.apache.spark.sql.types.{DoubleType, IntegerType, StringType, TimestampType}
    +
    +import org.apache.carbondata.core.constants.CarbonCommonConstants
    +import org.apache.carbondata.core.util.CarbonProperties
    +import org.apache.carbondata.spark.CastExpr
    +
    +object CastExpressionOptimization {
    +
    +
    +  def typeCastStringToLong(v: Any): Any = {
    +    val parser: SimpleDateFormat = new SimpleDateFormat(CarbonProperties.getInstance
    +      .getProperty(CarbonCommonConstants.CARBON_TIMESTAMP_FORMAT,
    +        CarbonCommonConstants.CARBON_TIMESTAMP_DEFAULT_FORMAT))
    +    try {
    +      val value = parser.parse(v.toString).getTime() * 1000L
    +      value
    +    } catch {
    +      case e: ParseException =>
    +        try {
    +          val parsenew: SimpleDateFormat = new SimpleDateFormat("yyyy-MM-dd'T'HH:mm:ss.SSSz")
    +          parsenew.parse(v.toString).getTime() * 1000L
    +        } catch {
    +          case e: ParseException =>
    +            val gmtDay = new SimpleDateFormat("yyyy-MM-dd", Locale.US)
    +            gmtDay.setTimeZone(TimeZone.getTimeZone("GMT"))
    +            try {
    +              gmtDay.parse(v.toString).getTime()
    +            } catch {
    +              case e: ParseException =>
    +                v
    +              case e: Exception =>
    +                v
    +            }
    +          case e: Exception =>
    +            v
    +        }
    +      case e: Exception =>
    +        v
    +    }
    +  }
    +
    +  def typeCastStringToLongList(list: Seq[Expression]): Seq[Expression] = {
    +    val tempList = new util.ArrayList[Expression]()
    +    list.foreach { value =>
    +      val output = typeCastStringToLong(value)
    +      if (!output.equals(value)) {
    +        tempList.add(output.asInstanceOf[Expression])
    +      }
    +    }
    +    if (tempList.size() != list.size) {
    +      list
    +    } else {
    +      tempList.asScala
    +    }
    +  }
    +
    +  def typeCastDoubleToIntList(list: Seq[Expression]): Seq[Expression] = {
    +    val tempList = new util.ArrayList[Expression]()
    +    list.foreach { value =>
    +      val output = value.asInstanceOf[Double].toInt
    +      if (value.asInstanceOf[Double].toInt.equals(output)) {
    +        tempList.add(output.asInstanceOf[Expression])
    +      }
    +    }
    +    if (tempList.size() != list.size) {
    +      list
    +    } else {
    +      tempList.asScala
    +    }
    +  }
    +
    +  /**
    +   * This routines tries to apply rules on Cast Filter Predicates and if the rules applied and the
    +   * values can be toss back to native datatypes the cast is removed. Current two rules are applied
    +   * a) Left : timestamp column      Right : String Value
    +   * Input from Spark : cast (col as string) <> 'String Literal'
    +   * Change to        : Column <> 'Long value of Timestamp String'
    +   *
    +   * b) Left : Integer Column        Right : String Value
    +   * Input from Spark : cast (col as double) <> 'Double Literal'
    +   * Change to        : Column <> 'Int value'
    +   *
    +   * @param expr
    +   * @return
    +   */
    +  def checkIfCastCanBeRemove(expr: Expression): Option[sources.Filter] = {
    +    expr match {
    +      case c@EqualTo(Cast(a: Attribute, _), Literal(v, t)) =>
    +        if ((a.dataType.isInstanceOf[TimestampType]) && (t.sameType(StringType))) {
    --- End diff --
   
    remove extra parenthesis for all if conditions


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [hidden email] or file a JIRA ticket
with INFRA.
---

[qiuchenjian-2]

Github user sounakr commented on a diff in the pull request:

    https://github.com/apache/incubator-carbondata/pull/805#discussion_r112674727
 
    --- Diff: integration/spark2/src/main/scala/org/apache/spark/sql/execution/CarbonLateDecodeStrategy.scala ---
    @@ -396,43 +402,41 @@ private[sql] class CarbonLateDecodeStrategy extends SparkStrategy {
         (unrecognizedPredicates ++ unhandledPredicates, translatedFilters)
       }
     
    +
       /**
        * Tries to translate a Catalyst [[Expression]] into data source [[Filter]].
        * @return a `Some[Filter]` if the input [[Expression]] is convertible, otherwise a `None`.
        */
    -  protected[sql] def translateFilter(predicate: Expression, or: Boolean = false): Option[Filter] = {
    +  protected[sql] def translateFilter(predicate: Expression): Option[Filter] = {
         predicate match {
           case or@Or(left, right) =>
     
    -        val leftFilter = translateFilter(left, true)
    -        val rightFilter = translateFilter(right, true)
    +        val leftFilter = translateFilter(left)
    +        val rightFilter = translateFilter(right)
             if (leftFilter.isDefined && rightFilter.isDefined) {
               Some(sources.Or(leftFilter.get, rightFilter.get))
             } else {
               None
             }
     
           case And(left, right) =>
    -        val leftFilter = translateFilter(left, or)
    -        val rightFilter = translateFilter(right, or)
    -        if (or) {
    -          if (leftFilter.isDefined && rightFilter.isDefined) {
    -            (translateFilter(left) ++ translateFilter(right)).reduceOption(sources.And)
    -          } else {
    -            None
    -          }
    -        } else {
    -          (translateFilter(left) ++ translateFilter(right)).reduceOption(sources.And)
    -        }
    -
    +        (translateFilter(left) ++ translateFilter(right)).reduceOption(sources.And)
    --- End diff --
   
    Replaced Back.


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [hidden email] or file a JIRA ticket
with INFRA.
---

[qiuchenjian-2]

Github user sounakr commented on a diff in the pull request:

    https://github.com/apache/incubator-carbondata/pull/805#discussion_r112674740
 
    --- Diff: integration/spark2/src/main/scala/org/apache/spark/sql/optimizer/CarbonFilters.scala ---
    @@ -0,0 +1,380 @@
    +/*
    + * 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.carbondata.spark
    +
    +import java.text.SimpleDateFormat
    +import java.util.Date
    +
    +import scala.collection.mutable.ArrayBuffer
    +
    +import org.apache.spark.sql.catalyst.expressions._
    +import org.apache.spark.sql.catalyst.expressions.aggregate.AggregateExpression
    +import org.apache.spark.sql.execution.CastExpressionOptimization
    +import org.apache.spark.sql.optimizer.AttributeReferenceWrapper
    +import org.apache.spark.sql.sources
    +import org.apache.spark.sql.sources.Filter
    +import org.apache.spark.sql.types._
    +
    +import org.apache.carbondata.core.constants.CarbonCommonConstants
    +import org.apache.carbondata.core.keygenerator.directdictionary.timestamp.TimeStampDirectDictionaryGenerator
    +import org.apache.carbondata.core.metadata.datatype.DataType
    +import org.apache.carbondata.core.metadata.schema.table.CarbonTable
    +import org.apache.carbondata.core.metadata.schema.table.column.CarbonColumn
    +import org.apache.carbondata.core.scan.expression.{ColumnExpression => CarbonColumnExpression, Expression => CarbonExpression, LiteralExpression => CarbonLiteralExpression}
    +import org.apache.carbondata.core.scan.expression.conditional._
    +import org.apache.carbondata.core.scan.expression.logical.{AndExpression, FalseExpression, OrExpression}
    +import org.apache.carbondata.spark.util.CarbonScalaUtil
    +
    +
    +/**
    + * All filter conversions are done here.
    + */
    +object CarbonFilters {
    +
    +
    +  /**
    +   * Converts data sources filters to carbon filter predicates.
    +   */
    +  def createCarbonFilter(schema: StructType,
    +      predicate: sources.Filter): Option[CarbonExpression] = {
    +    val dataTypeOf = schema.map(f => f.name -> f.dataType).toMap
    +
    +    def createFilter(predicate: sources.Filter): Option[CarbonExpression] = {
    +      predicate match {
    +
    +        case sources.EqualTo(name, value) =>
    +          Some(new EqualToExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, value)))
    +        case sources.Not(sources.EqualTo(name, value)) =>
    +          Some(new NotEqualsExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, value)))
    +        case sources.EqualNullSafe(name, value) =>
    +          Some(new EqualToExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, value)))
    +        case sources.Not(sources.EqualNullSafe(name, value)) =>
    +          Some(new NotEqualsExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, value)))
    +        case sources.GreaterThan(name, value) =>
    +          Some(new GreaterThanExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, value)))
    +        case sources.LessThan(name, value) =>
    +          Some(new LessThanExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, value)))
    +        case sources.GreaterThanOrEqual(name, value) =>
    +          Some(new GreaterThanEqualToExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, value)))
    +        case sources.LessThanOrEqual(name, value) =>
    +          Some(new LessThanEqualToExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, value)))
    +        case sources.In(name, values) =>
    +          Some(new InExpression(getCarbonExpression(name),
    +            new ListExpression(
    +              convertToJavaList(values.map(f => getCarbonLiteralExpression(name, f)).toList))))
    +        case sources.Not(sources.In(name, values)) =>
    +          Some(new NotInExpression(getCarbonExpression(name),
    +            new ListExpression(
    +              convertToJavaList(values.map(f => getCarbonLiteralExpression(name, f)).toList))))
    +        case sources.IsNull(name) =>
    +          Some(new EqualToExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, null), true))
    +        case sources.IsNotNull(name) =>
    +          Some(new NotEqualsExpression(getCarbonExpression(name),
    +            getCarbonLiteralExpression(name, null), true))
    +        case sources.And(lhs, rhs) =>
    +          (createFilter(lhs) ++ createFilter(rhs)).reduceOption(new AndExpression(_, _))
    +        case sources.Or(lhs, rhs) =>
    +          for {
    +            lhsFilter <- createFilter(lhs)
    +            rhsFilter <- createFilter(rhs)
    +          } yield {
    +            new OrExpression(lhsFilter, rhsFilter)
    +          }
    +        case CastExpr(expr: Expression) =>
    +          Some(transformExpression(expr))
    +        case _ => None
    +      }
    +    }
    +
    +    def getCarbonExpression(name: String) = {
    +      new CarbonColumnExpression(name,
    +        CarbonScalaUtil.convertSparkToCarbonDataType(dataTypeOf(name)))
    +    }
    +
    +    def getCarbonLiteralExpression(name: String, value: Any): CarbonExpression = {
    +      val dataTypeOfAttribute = CarbonScalaUtil.convertSparkToCarbonDataType(dataTypeOf(name))
    +      val dataType = if (Option(value).isDefined
    +                         && dataTypeOfAttribute == DataType.STRING
    +                         && value.isInstanceOf[Double]) {
    +        DataType.DOUBLE
    +      } else {
    +        dataTypeOfAttribute
    +      }
    +      new CarbonLiteralExpression(value, dataType)
    +    }
    +
    +    createFilter(predicate)
    +  }
    +
    +
    +  // Check out which filters can be pushed down to carbon, remaining can be handled in spark layer.
    +  // Mostly dimension filters are only pushed down since it is faster in carbon.
    +  def selectFilters(filters: Seq[Expression],
    +      attrList: java.util.HashSet[AttributeReferenceWrapper],
    +      aliasMap: CarbonAliasDecoderRelation): Unit = {
    +    def translate(expr: Expression, or: Boolean = false): Option[sources.Filter] = {
    +      expr match {
    +        case or@Or(left, right) =>
    +
    +          val leftFilter = translate(left, or = true)
    +          val rightFilter = translate(right, or = true)
    +          if (leftFilter.isDefined && rightFilter.isDefined) {
    +            Some(sources.Or(leftFilter.get, rightFilter.get))
    +          } else {
    +            or.collect {
    +              case attr: AttributeReference =>
    +                attrList.add(AttributeReferenceWrapper(aliasMap.getOrElse(attr, attr)))
    +            }
    +            None
    +          }
    +        case And(left, right) =>
    +          (translate(left) ++ translate(right)).reduceOption(sources.And)
    +        case EqualTo(a: Attribute, Literal(v, t)) =>
    +          Some(sources.EqualTo(a.name, v))
    +        case EqualTo(l@Literal(v, t), a: Attribute) =>
    +          Some(sources.EqualTo(a.name, v))
    +        case c@EqualTo(Cast(a: Attribute, _), Literal(v, t)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case c@EqualTo(Literal(v, t), Cast(a: Attribute, _)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case Not(EqualTo(a: Attribute, Literal(v, t))) =>
    +          Some(sources.Not(sources.EqualTo(a.name, v)))
    +        case Not(EqualTo(Literal(v, t), a: Attribute)) =>
    +          Some(sources.Not(sources.EqualTo(a.name, v)))
    +        case c@Not(EqualTo(Cast(a: Attribute, _), Literal(v, t))) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case c@Not(EqualTo(Literal(v, t), Cast(a: Attribute, _))) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case IsNotNull(a: Attribute) =>
    +          Some(sources.IsNotNull(a.name))
    +        case IsNull(a: Attribute) =>
    +          Some(sources.IsNull(a.name))
    +        case Not(In(a: Attribute, list)) if !list.exists(!_.isInstanceOf[Literal]) =>
    +          val hSet = list.map(e => e.eval(EmptyRow))
    +          Some(sources.Not(sources.In(a.name, hSet.toArray)))
    +        case In(a: Attribute, list) if !list.exists(!_.isInstanceOf[Literal]) =>
    +          val hSet = list.map(e => e.eval(EmptyRow))
    +          Some(sources.In(a.name, hSet.toArray))
    +        case c@Not(In(Cast(a: Attribute, _), list)) if !list.exists(!_.isInstanceOf[Literal]) =>
    +          Some(CastExpr(c))
    +        case c@In(Cast(a: Attribute, _), list) if !list.exists(!_.isInstanceOf[Literal]) =>
    +            Some(CastExpr(c))
    +        case GreaterThan(a: Attribute, Literal(v, t)) =>
    +          Some(sources.GreaterThan(a.name, v))
    +        case GreaterThan(Literal(v, t), a: Attribute) =>
    +          Some(sources.LessThan(a.name, v))
    +        case c@GreaterThan(Cast(a: Attribute, _), Literal(v, t)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case c@GreaterThan(Literal(v, t), Cast(a: Attribute, _)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case LessThan(a: Attribute, Literal(v, t)) =>
    +          Some(sources.LessThan(a.name, v))
    +        case LessThan(Literal(v, t), a: Attribute) =>
    +          Some(sources.GreaterThan(a.name, v))
    +        case c@LessThan(Cast(a: Attribute, _), Literal(v, t)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case c@LessThan(Literal(v, t), Cast(a: Attribute, _)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case GreaterThanOrEqual(a: Attribute, Literal(v, t)) =>
    +          Some(sources.GreaterThanOrEqual(a.name, v))
    +        case GreaterThanOrEqual(Literal(v, t), a: Attribute) =>
    +          Some(sources.LessThanOrEqual(a.name, v))
    +        case c@GreaterThanOrEqual(Cast(a: Attribute, _), Literal(v, t)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case c@GreaterThanOrEqual(Literal(v, t), Cast(a: Attribute, _)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case LessThanOrEqual(a: Attribute, Literal(v, t)) =>
    +          Some(sources.LessThanOrEqual(a.name, v))
    +        case LessThanOrEqual(Literal(v, t), a: Attribute) =>
    +          Some(sources.GreaterThanOrEqual(a.name, v))
    +        case c@LessThanOrEqual(Cast(a: Attribute, _), Literal(v, t)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case c@LessThanOrEqual(Literal(v, t), Cast(a: Attribute, _)) =>
    +          CastExpressionOptimization.checkIfCastCanBeRemove(c)
    +        case c@Cast(a: Attribute, _) =>
    +          Some(CastExpr(c))
    +        case others =>
    +          if (!or) {
    +            others.collect {
    +              case attr: AttributeReference =>
    +                attrList.add(AttributeReferenceWrapper(aliasMap.getOrElse(attr, attr)))
    +            }
    +          }
    +          None
    +      }
    +    }
    +
    +    filters.flatMap(translate(_, false)).toArray
    +  }
    +
    +  def transformExpression(expr: Expression): CarbonExpression = {
    +    expr match {
    +      case Or(left, right)
    +        if (isCarbonSupportedDataTypes(left) && isCarbonSupportedDataTypes(right)) => new
    --- End diff --
   
    Done


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [hidden email] or file a JIRA ticket
with INFRA.
---

[qiuchenjian-2]

Github user sounakr commented on a diff in the pull request:

    https://github.com/apache/incubator-carbondata/pull/805#discussion_r112674762
 
    --- Diff: integration/spark2/src/main/scala/org/apache/spark/sql/execution/CastExpressionOptimization.scala ---
    @@ -0,0 +1,387 @@
    +/*
    + * 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.execution
    +
    +import java.text.{ParseException, SimpleDateFormat}
    +import java.util
    +import java.util.{Date, Locale, TimeZone}
    +
    +import scala.collection.JavaConverters._
    +
    +import org.apache.spark.sql.catalyst.expressions.{Attribute, Cast, EmptyRow, EqualTo, Expression, GreaterThan, GreaterThanOrEqual, In, LessThan, LessThanOrEqual, Literal, Not}
    +import org.apache.spark.sql.sources
    +import org.apache.spark.sql.types.{DoubleType, IntegerType, StringType, TimestampType}
    +
    +import org.apache.carbondata.core.constants.CarbonCommonConstants
    +import org.apache.carbondata.core.util.CarbonProperties
    +import org.apache.carbondata.spark.CastExpr
    +
    +object CastExpressionOptimization {
    +
    +
    +  def typeCastStringToLong(v: Any): Any = {
    +    val parser: SimpleDateFormat = new SimpleDateFormat(CarbonProperties.getInstance
    +      .getProperty(CarbonCommonConstants.CARBON_TIMESTAMP_FORMAT,
    +        CarbonCommonConstants.CARBON_TIMESTAMP_DEFAULT_FORMAT))
    +    try {
    +      val value = parser.parse(v.toString).getTime() * 1000L
    +      value
    +    } catch {
    +      case e: ParseException =>
    +        try {
    +          val parsenew: SimpleDateFormat = new SimpleDateFormat("yyyy-MM-dd'T'HH:mm:ss.SSSz")
    +          parsenew.parse(v.toString).getTime() * 1000L
    +        } catch {
    +          case e: ParseException =>
    +            val gmtDay = new SimpleDateFormat("yyyy-MM-dd", Locale.US)
    +            gmtDay.setTimeZone(TimeZone.getTimeZone("GMT"))
    +            try {
    +              gmtDay.parse(v.toString).getTime()
    +            } catch {
    +              case e: ParseException =>
    +                v
    +              case e: Exception =>
    +                v
    +            }
    +          case e: Exception =>
    +            v
    +        }
    +      case e: Exception =>
    +        v
    +    }
    +  }
    +
    +  def typeCastStringToLongList(list: Seq[Expression]): Seq[Expression] = {
    +    val tempList = new util.ArrayList[Expression]()
    +    list.foreach { value =>
    +      val output = typeCastStringToLong(value)
    +      if (!output.equals(value)) {
    +        tempList.add(output.asInstanceOf[Expression])
    +      }
    +    }
    +    if (tempList.size() != list.size) {
    +      list
    +    } else {
    +      tempList.asScala
    +    }
    +  }
    +
    +  def typeCastDoubleToIntList(list: Seq[Expression]): Seq[Expression] = {
    +    val tempList = new util.ArrayList[Expression]()
    +    list.foreach { value =>
    +      val output = value.asInstanceOf[Double].toInt
    +      if (value.asInstanceOf[Double].toInt.equals(output)) {
    +        tempList.add(output.asInstanceOf[Expression])
    +      }
    +    }
    +    if (tempList.size() != list.size) {
    +      list
    +    } else {
    +      tempList.asScala
    +    }
    +  }
    +
    +  /**
    +   * This routines tries to apply rules on Cast Filter Predicates and if the rules applied and the
    +   * values can be toss back to native datatypes the cast is removed. Current two rules are applied
    +   * a) Left : timestamp column      Right : String Value
    +   * Input from Spark : cast (col as string) <> 'String Literal'
    +   * Change to        : Column <> 'Long value of Timestamp String'
    +   *
    +   * b) Left : Integer Column        Right : String Value
    +   * Input from Spark : cast (col as double) <> 'Double Literal'
    +   * Change to        : Column <> 'Int value'
    +   *
    +   * @param expr
    +   * @return
    +   */
    +  def checkIfCastCanBeRemove(expr: Expression): Option[sources.Filter] = {
    +    expr match {
    +      case c@EqualTo(Cast(a: Attribute, _), Literal(v, t)) =>
    +        if ((a.dataType.isInstanceOf[TimestampType]) && (t.sameType(StringType))) {
    --- End diff --
   
    Done


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [hidden email] or file a JIRA ticket
with INFRA.
---

[qiuchenjian-2]

Github user CarbonDataQA commented on the issue:

    https://github.com/apache/incubator-carbondata/pull/805
 
    Build Success with Spark 1.6.2, Please check CI http://136.243.101.176:8080/job/ApacheCarbonPRBuilder/1742/



---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [hidden email] or file a JIRA ticket
with INFRA.
---

[qiuchenjian-2]

Github user ravipesala commented on the issue:

    https://github.com/apache/incubator-carbondata/pull/805
 
    LGTM


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [hidden email] or file a JIRA ticket
with INFRA.
---

[qiuchenjian-2]

Github user asfgit closed the pull request at:

    https://github.com/apache/incubator-carbondata/pull/805


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [hidden email] or file a JIRA ticket
with INFRA.
---