Backport #46856 to 22.12: Fix a bug in Map data type

src/Functions/FunctionsHashing.h

@@ -1110,7 +1110,7 @@ class FunctionAnyHash : public IFunction
         {
             /// NOTE: here, of course, you can do without the materialization of the column.
             ColumnPtr full_column = col_from_const->convertToFullColumn();
-            executeArray<first>(type, &*full_column, vec_to);
+            executeArray<first>(type, full_column.get(), vec_to);
         }
         else
             throw Exception("Illegal column " + column->getName()
@@ -1123,6 +1123,10 @@ class FunctionAnyHash : public IFunction
     {
         WhichDataType which(from_type);
 
+        if (icolumn->size() != vec_to.size())
+            throw Exception(ErrorCodes::LOGICAL_ERROR, "Argument column '{}' size {} doesn't match result column size {} of function {}",
+                    icolumn->getName(), icolumn->size(), vec_to.size(), getName());
+
         if      (which.isUInt8()) executeIntType<UInt8, first>(icolumn, vec_to);
         else if (which.isUInt16()) executeIntType<UInt16, first>(icolumn, vec_to);
         else if (which.isUInt32()) executeIntType<UInt32, first>(icolumn, vec_to);
@@ -1181,10 +1185,9 @@ class FunctionAnyHash : public IFunction
             const auto & type_map = assert_cast<const DataTypeMap &>(*type);
             executeForArgument(type_map.getNestedType().get(), map->getNestedColumnPtr().get(), vec_to, is_first);
         }
-        else if (const auto * const_map = checkAndGetColumnConstData<ColumnMap>(column))
+        else if (const auto * const_map = checkAndGetColumnConst<ColumnMap>(column))
         {
-            const auto & type_map = assert_cast<const DataTypeMap &>(*type);
-            executeForArgument(type_map.getNestedType().get(), const_map->getNestedColumnPtr().get(), vec_to, is_first);
+            executeForArgument(type, const_map->convertToFullColumnIfConst().get(), vec_to, is_first);
         }
         else
         {
@@ -1220,15 +1223,14 @@ class FunctionAnyHash : public IFunction
 
     ColumnPtr executeImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr &, size_t input_rows_count) const override
     {
-        size_t rows = input_rows_count;
-        auto col_to = ColumnVector<ToType>::create(rows);
+        auto col_to = ColumnVector<ToType>::create(input_rows_count);
 
         typename ColumnVector<ToType>::Container & vec_to = col_to->getData();
 
         if (arguments.empty())
         {
-            /// Constant random number from /dev/urandom is used as a hash value of empty list of arguments.
-            vec_to.assign(rows, static_cast<ToType>(0xe28dbde7fe22e41c));
+            /// Return a fixed random-looking magic number when input is empty
+            vec_to.assign(input_rows_count, static_cast<ToType>(0xe28dbde7fe22e41c));
         }
 
         /// The function supports arbitrary number of arguments of arbitrary types.

src/Functions/IFunction.cpp

@@ -26,6 +26,7 @@
 #    pragma GCC diagnostic pop
 #endif
 
+
 namespace DB
 {
 
@@ -66,12 +67,12 @@ ColumnPtr replaceLowCardinalityColumnsByNestedAndGetDictionaryIndexes(
 
             if (!low_cardinality_type)
                 throw Exception(ErrorCodes::LOGICAL_ERROR,
-                    "Incompatible type for low cardinality column: {}",
+                    "Incompatible type for LowCardinality column: {}",
                     column.type->getName());
 
             if (can_be_executed_on_default_arguments)
             {
-                /// Normal case, when function can be executed on values's default.
+                /// Normal case, when function can be executed on values' default.
                 column.column = low_cardinality_column->getDictionary().getNestedColumn();
                 indexes = low_cardinality_column->getIndexesPtr();
             }
@@ -280,6 +281,7 @@ ColumnPtr IExecutableFunction::executeWithoutSparseColumns(const ColumnsWithType
 
             auto res = executeWithoutLowCardinalityColumns(columns_without_low_cardinality, dictionary_type, new_input_rows_count, dry_run);
             bool res_is_constant = isColumnConst(*res);
+
             auto keys = res_is_constant
                 ? res->cloneResized(1)->convertToFullColumnIfConst()
                 : res;

src/Functions/map.cpp

@@ -334,7 +334,7 @@ class FunctionMapContainsKeyLike : public IFunction
             }
 
             size_t col_key_size = sub_map_column->size();
-            auto column = is_const? ColumnConst::create(std::move(sub_map_column), std::move(col_key_size)) : std::move(sub_map_column);
+            auto column = is_const ? ColumnConst::create(std::move(sub_map_column), std::move(col_key_size)) : std::move(sub_map_column);
 
             ColumnsWithTypeAndName new_arguments =
                 {

tests/queries/0_stateless/02673_map_hashing_msan.reference

@@ -0,0 +1,6 @@
+4786021384179797717
+5368498105280294197
+42	15687122600100720591
+42	15687122600100720591
+42	15687122600100720591
+\N

tests/queries/0_stateless/02673_map_hashing_msan.sql

@@ -0,0 +1,7 @@
+SELECT cityHash64(map(1, 'Hello'), CAST(materialize('World') AS LowCardinality(String)));
+SELECT cityHash64(map(), CAST(materialize('') AS LowCardinality(Nullable(String))));
+SELECT materialize(42) as last_element, cityHash64(map(), CAST(materialize('') AS LowCardinality(Nullable(String))), last_element) from numbers(3);
+
+SET allow_suspicious_low_cardinality_types = 1;
+CREATE TEMPORARY TABLE datetime__fuzz_14 (`d` LowCardinality(Nullable(UInt128)));
+SELECT max(mapPopulateSeries(mapPopulateSeries(map(toInt64(1048), toInt64(9223), 3, -2147))), toInt64(1048), map('11', 257, '', NULL), cityHash64(*)) > NULL FROM (SELECT max(cityHash64(mapPopulateSeries(mapPopulateSeries(map(toInt64(1048), toInt64(2147), 655, -2147))), *)) > NULL, map(toInt64(-2147), toInt64(100.0001), -2147, NULL), mapPopulateSeries(map(toInt64(1024), toInt64(1048), 1048, -1)), map(toInt64(256), toInt64(NULL), -1, NULL), quantile(0.0001)(d) FROM datetime__fuzz_14 WITH TOTALS);