DataTypePtr FieldToDataType::operator() (const Array & x) const
{
DataTypes element_types;
element_types.reserve(x.size());
for (const Field & elem : x)
element_types.emplace_back(applyVisitor(FieldToDataType(), elem));
return std::make_shared<DataTypeArray>(getLeastSupertype(element_types));
}
DataTypePtr getReturnTypeImpl(const DataTypes & arguments) const override
{
/// Skip all NULL arguments. If any argument is non-Nullable, skip all next arguments.
DataTypes filtered_args;
filtered_args.reserve(arguments.size());
for (const auto & arg : arguments)
{
if (arg->onlyNull())
continue;
filtered_args.push_back(arg);
if (!arg->isNullable())
break;
}
DataTypes new_args;
for (size_t i = 0; i < filtered_args.size(); ++i)
{
bool is_last = i + 1 == filtered_args.size();
if (is_last)
new_args.push_back(filtered_args[i]);
else
new_args.push_back(removeNullable(filtered_args[i]));
}
if (new_args.empty())
return std::make_shared<DataTypeNullable>(std::make_shared<DataTypeNothing>());
if (new_args.size() == 1)
return new_args.front();
auto res = getLeastSupertype(new_args);
/// if last argument is not nullable, result should be also not nullable
if (!new_args.back()->isNullable() && res->isNullable())
res = removeNullable(res);
return res;
}
DataTypePtr getReturnTypeImpl(const DataTypes & args) const override
{
/// Arguments are the following: cond1, then1, cond2, then2, ... condN, thenN, else.
auto for_conditions = [&args](auto && f)
{
size_t conditions_end = args.size() - 1;
for (size_t i = 0; i < conditions_end; i += 2)
f(args[i]);
};
auto for_branches = [&args](auto && f)
{
size_t branches_end = args.size();
for (size_t i = 1; i < branches_end; i += 2)
f(args[i]);
f(args.back());
};
if (!(args.size() >= 3 && args.size() % 2 == 1))
throw Exception{"Invalid number of arguments for function " + getName(),
ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH};
/// Conditions must be UInt8, Nullable(UInt8) or Null. If one of conditions is Nullable, the result is also Nullable.
bool have_nullable_condition = false;
for_conditions([&](const DataTypePtr & arg)
{
const IDataType * nested_type;
if (arg->isNullable())
{
have_nullable_condition = true;
if (arg->onlyNull())
return;
const DataTypeNullable & nullable_type = static_cast<const DataTypeNullable &>(*arg);
nested_type = nullable_type.getNestedType().get();
}
else
{
nested_type = arg.get();
}
if (!WhichDataType(nested_type).isUInt8())
throw Exception{"Illegal type " + arg->getName() + " of argument (condition) "
"of function " + getName() + ". Must be UInt8.",
ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT};
});
DataTypes types_of_branches;
types_of_branches.reserve(args.size() / 2 + 1);
for_branches([&](const DataTypePtr & arg)
{
types_of_branches.emplace_back(arg);
});
DataTypePtr common_type_of_branches = getLeastSupertype(types_of_branches);
return have_nullable_condition
? makeNullable(common_type_of_branches)
: common_type_of_branches;
}
InterpreterSelectWithUnionQuery::InterpreterSelectWithUnionQuery(
const ASTPtr & query_ptr_,
const Context & context_,
const Names & required_result_column_names,
QueryProcessingStage::Enum to_stage_,
size_t subquery_depth_,
bool only_analyze,
bool modify_inplace)
: query_ptr(query_ptr_),
context(context_),
to_stage(to_stage_),
subquery_depth(subquery_depth_)
{
const ASTSelectWithUnionQuery & ast = typeid_cast<const ASTSelectWithUnionQuery &>(*query_ptr);
size_t num_selects = ast.list_of_selects->children.size();
if (!num_selects)
throw Exception("Logical error: no children in ASTSelectWithUnionQuery", ErrorCodes::LOGICAL_ERROR);
/// Initialize interpreters for each SELECT query.
/// Note that we pass 'required_result_column_names' to first SELECT.
/// And for the rest, we pass names at the corresponding positions of 'required_result_column_names' in the result of first SELECT,
/// because names could be different.
nested_interpreters.reserve(num_selects);
std::vector<Names> required_result_column_names_for_other_selects(num_selects);
if (!required_result_column_names.empty() && num_selects > 1)
{
/// Result header if there are no filtering by 'required_result_column_names'.
/// We use it to determine positions of 'required_result_column_names' in SELECT clause.
Block full_result_header = InterpreterSelectQuery(
ast.list_of_selects->children.at(0), context, Names(), to_stage, subquery_depth, true).getSampleBlock();
std::vector<size_t> positions_of_required_result_columns(required_result_column_names.size());
for (size_t required_result_num = 0, size = required_result_column_names.size(); required_result_num < size; ++required_result_num)
positions_of_required_result_columns[required_result_num] = full_result_header.getPositionByName(required_result_column_names[required_result_num]);
for (size_t query_num = 1; query_num < num_selects; ++query_num)
{
Block full_result_header_for_current_select = InterpreterSelectQuery(
ast.list_of_selects->children.at(query_num), context, Names(), to_stage, subquery_depth, true).getSampleBlock();
if (full_result_header_for_current_select.columns() != full_result_header.columns())
throw Exception("Different number of columns in UNION ALL elements:\n"
+ full_result_header.dumpNames()
+ "\nand\n"
+ full_result_header_for_current_select.dumpNames() + "\n",
ErrorCodes::UNION_ALL_RESULT_STRUCTURES_MISMATCH);
required_result_column_names_for_other_selects[query_num].reserve(required_result_column_names.size());
for (const auto & pos : positions_of_required_result_columns)
required_result_column_names_for_other_selects[query_num].push_back(full_result_header_for_current_select.getByPosition(pos).name);
}
}
for (size_t query_num = 0; query_num < num_selects; ++query_num)
{
const Names & current_required_result_column_names
= query_num == 0 ? required_result_column_names : required_result_column_names_for_other_selects[query_num];
nested_interpreters.emplace_back(std::make_unique<InterpreterSelectQuery>(
ast.list_of_selects->children.at(query_num),
context,
current_required_result_column_names,
to_stage,
subquery_depth,
only_analyze,
modify_inplace));
}
/// Determine structure of the result.
if (num_selects == 1)
{
result_header = nested_interpreters.front()->getSampleBlock();
}
else
{
Blocks headers(num_selects);
for (size_t query_num = 0; query_num < num_selects; ++query_num)
headers[query_num] = nested_interpreters[query_num]->getSampleBlock();
result_header = headers.front();
size_t num_columns = result_header.columns();
for (size_t query_num = 1; query_num < num_selects; ++query_num)
if (headers[query_num].columns() != num_columns)
throw Exception("Different number of columns in UNION ALL elements:\n"
+ result_header.dumpNames()
+ "\nand\n"
+ headers[query_num].dumpNames() + "\n",
ErrorCodes::UNION_ALL_RESULT_STRUCTURES_MISMATCH);
for (size_t column_num = 0; column_num < num_columns; ++column_num)
{
ColumnWithTypeAndName & result_elem = result_header.getByPosition(column_num);
/// Determine common type.
DataTypes types(num_selects);
for (size_t query_num = 0; query_num < num_selects; ++query_num)
types[query_num] = headers[query_num].getByPosition(column_num).type;
result_elem.type = getLeastSupertype(types);
/// If there are different constness or different values of constants, the result must be non-constant.
if (result_elem.column->isColumnConst())
{
bool need_materialize = false;
for (size_t query_num = 1; query_num < num_selects; ++query_num)
{
const ColumnWithTypeAndName & source_elem = headers[query_num].getByPosition(column_num);
if (!source_elem.column->isColumnConst()
|| (static_cast<const ColumnConst &>(*result_elem.column).getField()
!= static_cast<const ColumnConst &>(*source_elem.column).getField()))
{
need_materialize = true;
break;
}
}
if (need_materialize)
result_elem.column = result_elem.type->createColumn();
}
/// BTW, result column names are from first SELECT.
}
}
}
DataTypePtr getReturnTypeImpl(const DataTypes & arguments) const override
{
return std::make_shared<DataTypeArray>(getLeastSupertype(arguments));
}
