HIVE-16511: CBO looses inner casts on constants of complex type (Vine…

…et Garg, reviewed by Ashutosh Chauhan)

itests/src/test/resources/testconfiguration.properties

@@ -30,7 +30,6 @@ disabled.query.files=ql_rewrite_gbtoidx.q,\
   cbo_rp_subq_not_in.q,\
   cbo_rp_subq_exists.q,\
   orc_llap.q,\
-  min_structvalue.q,\
   ql_rewrite_gbtoidx_cbo_2.q,\
   rcfile_merge1.q,\
   smb_mapjoin_8.q,\

ql/src/java/org/apache/hadoop/hive/ql/exec/vector/VectorizationContext.java

@@ -2351,11 +2351,10 @@ private VectorExpression getBetweenFilterExpression(List<ExprNodeDesc> childExpr
   }
 
   private boolean isNullConst(ExprNodeDesc exprNodeDesc) {
-    if (exprNodeDesc instanceof ExprNodeConstantDesc) {
-      String typeString = exprNodeDesc.getTypeString();
-      if (typeString.equalsIgnoreCase("void")) {
+    //null constant could be typed so we need to check the value
+    if (exprNodeDesc instanceof ExprNodeConstantDesc &&
+        ((ExprNodeConstantDesc) exprNodeDesc).getValue() == null) {
         return true;
-      }
     }
     return false;
   }

ql/src/java/org/apache/hadoop/hive/ql/optimizer/calcite/translator/ASTConverter.java

@@ -19,9 +19,12 @@
 
 import java.math.BigDecimal;
 import java.util.ArrayList;
+import java.util.Comparator;
 import java.util.LinkedList;
 import java.util.List;
 import java.util.Map;
+import java.util.TreeMap;
+
 
 import org.apache.calcite.adapter.druid.DruidQuery;
 import org.apache.calcite.rel.RelFieldCollation;
@@ -40,6 +43,7 @@
 import org.apache.calcite.rel.core.TableScan;
 import org.apache.calcite.rel.core.Union;
 import org.apache.calcite.rel.type.RelDataTypeField;
+import org.apache.calcite.rex.RexBuilder;
 import org.apache.calcite.rex.RexCall;
 import org.apache.calcite.rex.RexFieldAccess;
 import org.apache.calcite.rex.RexFieldCollation;
@@ -205,12 +209,8 @@ else if (aggregateType == Group.CUBE) {
         int i = 0;
 
         for (RexNode r : select.getChildExps()) {
-          if (RexUtil.isNull(r) && r.getType().getSqlTypeName() != SqlTypeName.NULL) {
-            // It is NULL value with different type, we need to introduce a CAST
-            // to keep it
-            r = select.getCluster().getRexBuilder().makeAbstractCast(r.getType(), r);
-          }
-          ASTNode expr = r.accept(new RexVisitor(schema, r instanceof RexLiteral));
+          ASTNode expr = r.accept(new RexVisitor(schema, r instanceof RexLiteral,
+              select.getCluster().getRexBuilder()));
           String alias = select.getRowType().getFieldNames().get(i++);
           ASTNode selectExpr = ASTBuilder.selectExpr(expr, alias);
           b.add(selectExpr);
@@ -223,12 +223,8 @@ else if (aggregateType == Group.CUBE) {
       List<ASTNode> children = new ArrayList<>();
       RexCall call = (RexCall) udtf.getCall();
       for (RexNode r : call.getOperands()) {
-        if (RexUtil.isNull(r) && r.getType().getSqlTypeName() != SqlTypeName.NULL) {
-          // It is NULL value with different type, we need to introduce a CAST
-          // to keep it
-          r = select.getCluster().getRexBuilder().makeAbstractCast(r.getType(), r);
-        }
-        ASTNode expr = r.accept(new RexVisitor(schema, r instanceof RexLiteral));
+        ASTNode expr = r.accept(new RexVisitor(schema, r instanceof RexLiteral,
+            select.getCluster().getRexBuilder()));
         children.add(expr);
       }
       ASTBuilder sel = ASTBuilder.construct(HiveParser.TOK_SELEXPR, "TOK_SELEXPR");
@@ -460,19 +456,41 @@ public void visit(RelNode node, int ordinal, RelNode parent) {
 
   }
 
+
   static class RexVisitor extends RexVisitorImpl<ASTNode> {
 
     private final Schema schema;
     private final boolean useTypeQualInLiteral;
+    private final RexBuilder rexBuilder;
+    // this is to keep track of null literal which already has been visited
+    private Map<RexLiteral, Boolean> nullLiteralMap ;
+
+
+    protected RexVisitor(Schema schema, boolean useTypeQualInLiteral) {
+      this(schema, useTypeQualInLiteral, null);
 
+    }
     protected RexVisitor(Schema schema) {
       this(schema, false);
     }
 
-    protected RexVisitor(Schema schema, boolean useTypeQualInLiteral) {
+    protected RexVisitor(Schema schema, boolean useTypeQualInLiteral, RexBuilder rexBuilder) {
       super(true);
       this.schema = schema;
       this.useTypeQualInLiteral = useTypeQualInLiteral;
+      this.rexBuilder = rexBuilder;
+
+      this.nullLiteralMap =
+          new TreeMap<>(new Comparator<RexLiteral>(){
+            // RexLiteral's equal only consider value and type which isn't sufficient
+            // so providing custom comparator which distinguishes b/w objects irrespective
+            // of value/type
+            @Override
+            public int compare(RexLiteral o1, RexLiteral o2) {
+              if(o1 == o2) return 0;
+              else return 1;
+            }
+          });
     }
 
     @Override
@@ -497,6 +515,19 @@ public ASTNode visitInputRef(RexInputRef inputRef) {
 
     @Override
     public ASTNode visitLiteral(RexLiteral literal) {
+
+      if (RexUtil.isNull(literal) && literal.getType().getSqlTypeName() != SqlTypeName.NULL
+          && rexBuilder != null) {
+        // It is NULL value with different type, we need to introduce a CAST
+        // to keep it
+        if(nullLiteralMap.containsKey(literal)) {
+          return ASTBuilder.literal(literal, useTypeQualInLiteral);
+        }
+        nullLiteralMap.put(literal, true);
+        RexNode r = rexBuilder.makeAbstractCast(literal.getType(), literal);
+
+        return r.accept(this);
+      }
       return ASTBuilder.literal(literal, useTypeQualInLiteral);
     }
 

ql/src/test/results/clientpositive/llap/vector_coalesce.q.out

@@ -523,7 +523,7 @@ STAGE PLANS:
                           className: VectorSelectOperator
                           native: true
                           projectedOutputColumns: [12, 0, 14]
-                          selectExpressions: ConstantVectorExpression(val null) -> 12:bigint, VectorCoalesce(columns [13, 0])(children: ConstantVectorExpression(val null) -> 13:tinyint, col 0) -> 14:tinyint
+                          selectExpressions: ConstantVectorExpression(val null) -> 12:bigint, VectorCoalesce(columns [13, 0])(children: ConstantVectorExpression(val null) -> 13:bigint, col 0) -> 14:bigint
                         Limit Vectorization:
                             className: VectorLimitOperator
                             native: true

ql/src/test/results/clientpositive/llap/vector_groupby_grouping_sets_grouping.q.out

@@ -772,7 +772,7 @@ STAGE PLANS:
             Execution mode: vectorized, llap
             LLAP IO: all inputs
         Reducer 2 
-            Execution mode: llap
+            Execution mode: vectorized, llap
             Reduce Operator Tree:
               Group By Operator
                 keys: KEY._col0 (type: int), KEY._col1 (type: int), KEY._col2 (type: int)

ql/src/test/results/clientpositive/llap/vector_ptf_part_simple.q.out

@@ -5613,9 +5613,9 @@ STAGE PLANS:
                       native: true
                       projectedOutputColumns: [0, 1, 2]
                   Reduce Output Operator
-                    key expressions: p_mfgr (type: string), CASE WHEN ((p_mfgr = 'Manufacturer#2')) THEN (2000-01-01 00:00:00.0) ELSE (null) END (type: timestamp)
+                    key expressions: p_mfgr (type: string), CASE WHEN ((p_mfgr = 'Manufacturer#2')) THEN (2000-01-01 00:00:00.0) ELSE (CAST( null AS TIMESTAMP)) END (type: timestamp)
                     sort order: ++
-                    Map-reduce partition columns: p_mfgr (type: string), CASE WHEN ((p_mfgr = 'Manufacturer#2')) THEN (2000-01-01 00:00:00.0) ELSE (null) END (type: timestamp)
+                    Map-reduce partition columns: p_mfgr (type: string), CASE WHEN ((p_mfgr = 'Manufacturer#2')) THEN (2000-01-01 00:00:00.0) ELSE (CAST( null AS TIMESTAMP)) END (type: timestamp)
                     Reduce Sink Vectorization:
                         className: VectorReduceSinkMultiKeyOperator
                         keyColumns: [0, 5]
@@ -5662,13 +5662,13 @@ STAGE PLANS:
                       Windowing table definition
                         input alias: ptf_1
                         name: windowingtablefunction
-                        order by: _col0 ASC NULLS FIRST, CASE WHEN ((_col0 = 'Manufacturer#2')) THEN (2000-01-01 00:00:00.0) ELSE (null) END ASC NULLS FIRST
-                        partition by: _col0, CASE WHEN ((_col0 = 'Manufacturer#2')) THEN (2000-01-01 00:00:00.0) ELSE (null) END
+                        order by: _col0 ASC NULLS FIRST, CASE WHEN ((_col0 = 'Manufacturer#2')) THEN (2000-01-01 00:00:00.0) ELSE (CAST( null AS TIMESTAMP)) END ASC NULLS FIRST
+                        partition by: _col0, CASE WHEN ((_col0 = 'Manufacturer#2')) THEN (2000-01-01 00:00:00.0) ELSE (CAST( null AS TIMESTAMP)) END
                         raw input shape:
                         window functions:
                             window function definition
                               alias: rank_window_0
-                              arguments: _col0, CASE WHEN ((_col0 = 'Manufacturer#2')) THEN (2000-01-01 00:00:00.0) ELSE (null) END
+                              arguments: _col0, CASE WHEN ((_col0 = 'Manufacturer#2')) THEN (2000-01-01 00:00:00.0) ELSE (CAST( null AS TIMESTAMP)) END
                               name: rank
                               window function: GenericUDAFRankEvaluator
                               window frame: ROWS PRECEDING(MAX)~FOLLOWING(MAX)
@@ -5781,9 +5781,9 @@ STAGE PLANS:
                       native: true
                       projectedOutputColumns: [0, 1, 2]
                   Reduce Output Operator
-                    key expressions: p_mfgr (type: string), CASE WHEN ((p_mfgr = 'Manufacturer#2')) THEN (2000-01-01 00:00:00.0) ELSE (null) END (type: timestamp), p_name (type: string)
+                    key expressions: p_mfgr (type: string), CASE WHEN ((p_mfgr = 'Manufacturer#2')) THEN (2000-01-01 00:00:00.0) ELSE (CAST( null AS TIMESTAMP)) END (type: timestamp), p_name (type: string)
                     sort order: +++
-                    Map-reduce partition columns: p_mfgr (type: string), CASE WHEN ((p_mfgr = 'Manufacturer#2')) THEN (2000-01-01 00:00:00.0) ELSE (null) END (type: timestamp)
+                    Map-reduce partition columns: p_mfgr (type: string), CASE WHEN ((p_mfgr = 'Manufacturer#2')) THEN (2000-01-01 00:00:00.0) ELSE (CAST( null AS TIMESTAMP)) END (type: timestamp)
                     Reduce Sink Vectorization:
                         className: VectorReduceSinkObjectHashOperator
                         keyColumns: [0, 5, 1]
@@ -5845,7 +5845,7 @@ STAGE PLANS:
                         input alias: ptf_1
                         name: windowingtablefunction
                         order by: _col1 ASC NULLS FIRST
-                        partition by: _col0, CASE WHEN ((_col0 = 'Manufacturer#2')) THEN (2000-01-01 00:00:00.0) ELSE (null) END
+                        partition by: _col0, CASE WHEN ((_col0 = 'Manufacturer#2')) THEN (2000-01-01 00:00:00.0) ELSE (CAST( null AS TIMESTAMP)) END
                         raw input shape:
                         window functions:
                             window function definition

ql/src/test/results/clientpositive/min_structvalue.q.out

@@ -0,0 +1,45 @@
+PREHOOK: query: select max(a), min(a) FROM (select named_struct("field",1) as a union all select named_struct("field",2) as a union all select named_struct("field",cast(null as int)) as a) tmp
+PREHOOK: type: QUERY
+PREHOOK: Input: _dummy_database@_dummy_table
+#### A masked pattern was here ####
+POSTHOOK: query: select max(a), min(a) FROM (select named_struct("field",1) as a union all select named_struct("field",2) as a union all select named_struct("field",cast(null as int)) as a) tmp
+POSTHOOK: type: QUERY
+POSTHOOK: Input: _dummy_database@_dummy_table
+#### A masked pattern was here ####
+{"field":2}	{"field":1}
+PREHOOK: query: select min(a) FROM (select named_struct("field",1) as a union all select named_struct("field",-2) as a union all select named_struct("field",cast(null as int)) as a) tmp
+PREHOOK: type: QUERY
+PREHOOK: Input: _dummy_database@_dummy_table
+#### A masked pattern was here ####
+POSTHOOK: query: select min(a) FROM (select named_struct("field",1) as a union all select named_struct("field",-2) as a union all select named_struct("field",cast(null as int)) as a) tmp
+POSTHOOK: type: QUERY
+POSTHOOK: Input: _dummy_database@_dummy_table
+#### A masked pattern was here ####
+{"field":-2}
+PREHOOK: query: select min(a) FROM (select named_struct("field",1) as a union all select named_struct("field",2) as a union all select named_struct("field",cast(5 as int)) as a) tmp
+PREHOOK: type: QUERY
+PREHOOK: Input: _dummy_database@_dummy_table
+#### A masked pattern was here ####
+POSTHOOK: query: select min(a) FROM (select named_struct("field",1) as a union all select named_struct("field",2) as a union all select named_struct("field",cast(5 as int)) as a) tmp
+POSTHOOK: type: QUERY
+POSTHOOK: Input: _dummy_database@_dummy_table
+#### A masked pattern was here ####
+{"field":1}
+PREHOOK: query: select min(a) FROM (select named_struct("field",1, "secf", cast(null as int) ) as a union all select named_struct("field",2, "secf", 3) as a union all select named_struct("field",cast(5 as int), "secf", 4) as a) tmp
+PREHOOK: type: QUERY
+PREHOOK: Input: _dummy_database@_dummy_table
+#### A masked pattern was here ####
+POSTHOOK: query: select min(a) FROM (select named_struct("field",1, "secf", cast(null as int) ) as a union all select named_struct("field",2, "secf", 3) as a union all select named_struct("field",cast(5 as int), "secf", 4) as a) tmp
+POSTHOOK: type: QUERY
+POSTHOOK: Input: _dummy_database@_dummy_table
+#### A masked pattern was here ####
+{"field":1,"secf":null}
+PREHOOK: query: select min(a) FROM (select named_struct("field",1, "secf", 2) as a union all select named_struct("field",-2, "secf", 3) as a union all select named_struct("field",cast(null as int), "secf", 1) as a) tmp
+PREHOOK: type: QUERY
+PREHOOK: Input: _dummy_database@_dummy_table
+#### A masked pattern was here ####
+POSTHOOK: query: select min(a) FROM (select named_struct("field",1, "secf", 2) as a union all select named_struct("field",-2, "secf", 3) as a union all select named_struct("field",cast(null as int), "secf", 1) as a) tmp
+POSTHOOK: type: QUERY
+POSTHOOK: Input: _dummy_database@_dummy_table
+#### A masked pattern was here ####
+{"field":-2,"secf":3}

ql/src/test/results/clientpositive/perf/spark/query36.q.out

@@ -215,9 +215,9 @@ STAGE PLANS:
                   outputColumnNames: _col0, _col1, _col4, _col5, _col6
                   Statistics: Num rows: 1149975358 Data size: 101451159969 Basic stats: COMPLETE Column stats: NONE
                   Reduce Output Operator
-                    key expressions: (grouping(_col6, 1) + grouping(_col6, 0)) (type: int), CASE WHEN ((grouping(_col6, 0) = 0)) THEN (_col0) ELSE (null) END (type: string), (_col4 / _col5) (type: decimal(37,20))
+                    key expressions: (grouping(_col6, 1) + grouping(_col6, 0)) (type: int), CASE WHEN ((grouping(_col6, 0) = 0)) THEN (_col0) ELSE (CAST( null AS varchar(65535))) END (type: string), (_col4 / _col5) (type: decimal(37,20))
                     sort order: +++
-                    Map-reduce partition columns: (grouping(_col6, 1) + grouping(_col6, 0)) (type: int), CASE WHEN ((grouping(_col6, 0) = 0)) THEN (_col0) ELSE (null) END (type: string)
+                    Map-reduce partition columns: (grouping(_col6, 1) + grouping(_col6, 0)) (type: int), CASE WHEN ((grouping(_col6, 0) = 0)) THEN (_col0) ELSE (CAST( null AS varchar(65535))) END (type: string)
                     Statistics: Num rows: 1149975358 Data size: 101451159969 Basic stats: COMPLETE Column stats: NONE
                     value expressions: _col0 (type: string), _col1 (type: string), _col4 (type: decimal(17,2)), _col5 (type: decimal(17,2)), _col6 (type: int)
         Reducer 5 
@@ -236,7 +236,7 @@ STAGE PLANS:
                         input alias: ptf_1
                         name: windowingtablefunction
                         order by: (_col4 / _col5) ASC NULLS FIRST
-                        partition by: (grouping(_col6, 1) + grouping(_col6, 0)), CASE WHEN ((grouping(_col6, 0) = 0)) THEN (_col0) ELSE (null) END
+                        partition by: (grouping(_col6, 1) + grouping(_col6, 0)), CASE WHEN ((grouping(_col6, 0) = 0)) THEN (_col0) ELSE (CAST( null AS varchar(65535))) END
                         raw input shape:
                         window functions:
                             window function definition

ql/src/test/results/clientpositive/perf/spark/query70.q.out

@@ -362,9 +362,9 @@ STAGE PLANS:
                   outputColumnNames: _col0, _col1, _col4, _col5
                   Statistics: Num rows: 1149975358 Data size: 101451159969 Basic stats: COMPLETE Column stats: NONE
                   Reduce Output Operator
-                    key expressions: (grouping(_col5, 1) + grouping(_col5, 0)) (type: int), CASE WHEN ((grouping(_col5, 0) = 0)) THEN (_col0) ELSE (null) END (type: string), _col4 (type: decimal(17,2))
+                    key expressions: (grouping(_col5, 1) + grouping(_col5, 0)) (type: int), CASE WHEN ((grouping(_col5, 0) = 0)) THEN (_col0) ELSE (CAST( null AS varchar(65535))) END (type: string), _col4 (type: decimal(17,2))
                     sort order: ++-
-                    Map-reduce partition columns: (grouping(_col5, 1) + grouping(_col5, 0)) (type: int), CASE WHEN ((grouping(_col5, 0) = 0)) THEN (_col0) ELSE (null) END (type: string)
+                    Map-reduce partition columns: (grouping(_col5, 1) + grouping(_col5, 0)) (type: int), CASE WHEN ((grouping(_col5, 0) = 0)) THEN (_col0) ELSE (CAST( null AS varchar(65535))) END (type: string)
                     Statistics: Num rows: 1149975358 Data size: 101451159969 Basic stats: COMPLETE Column stats: NONE
                     value expressions: _col0 (type: string), _col1 (type: string), _col5 (type: int)
         Reducer 5 
@@ -383,7 +383,7 @@ STAGE PLANS:
                         input alias: ptf_1
                         name: windowingtablefunction
                         order by: _col4 DESC NULLS LAST
-                        partition by: (grouping(_col5, 1) + grouping(_col5, 0)), CASE WHEN ((grouping(_col5, 0) = 0)) THEN (_col0) ELSE (null) END
+                        partition by: (grouping(_col5, 1) + grouping(_col5, 0)), CASE WHEN ((grouping(_col5, 0) = 0)) THEN (_col0) ELSE (CAST( null AS varchar(65535))) END
                         raw input shape:
                         window functions:
                             window function definition

ql/src/test/results/clientpositive/perf/spark/query86.q.out

@@ -173,9 +173,9 @@ STAGE PLANS:
                   outputColumnNames: _col0, _col1, _col4, _col5
                   Statistics: Num rows: 261364852 Data size: 35538061226 Basic stats: COMPLETE Column stats: NONE
                   Reduce Output Operator
-                    key expressions: (grouping(_col5, 1) + grouping(_col5, 0)) (type: int), CASE WHEN ((grouping(_col5, 0) = 0)) THEN (_col0) ELSE (null) END (type: string), _col4 (type: decimal(17,2))
+                    key expressions: (grouping(_col5, 1) + grouping(_col5, 0)) (type: int), CASE WHEN ((grouping(_col5, 0) = 0)) THEN (_col0) ELSE (CAST( null AS varchar(65535))) END (type: string), _col4 (type: decimal(17,2))
                     sort order: ++-
-                    Map-reduce partition columns: (grouping(_col5, 1) + grouping(_col5, 0)) (type: int), CASE WHEN ((grouping(_col5, 0) = 0)) THEN (_col0) ELSE (null) END (type: string)
+                    Map-reduce partition columns: (grouping(_col5, 1) + grouping(_col5, 0)) (type: int), CASE WHEN ((grouping(_col5, 0) = 0)) THEN (_col0) ELSE (CAST( null AS varchar(65535))) END (type: string)
                     Statistics: Num rows: 261364852 Data size: 35538061226 Basic stats: COMPLETE Column stats: NONE
                     value expressions: _col0 (type: string), _col1 (type: string), _col5 (type: int)
         Reducer 4 
@@ -194,7 +194,7 @@ STAGE PLANS:
                         input alias: ptf_1
                         name: windowingtablefunction
                         order by: _col4 DESC NULLS LAST
-                        partition by: (grouping(_col5, 1) + grouping(_col5, 0)), CASE WHEN ((grouping(_col5, 0) = 0)) THEN (_col0) ELSE (null) END
+                        partition by: (grouping(_col5, 1) + grouping(_col5, 0)), CASE WHEN ((grouping(_col5, 0) = 0)) THEN (_col0) ELSE (CAST( null AS varchar(65535))) END
                         raw input shape:
                         window functions:
                             window function definition