| 1 | #include "duckdb/function/scalar/string_functions.hpp" |
| 2 | |
| 3 | #include "duckdb/common/exception.hpp" |
| 4 | #include "duckdb/common/types/date.hpp" |
| 5 | #include "duckdb/common/vector_operations/vector_operations.hpp" |
| 6 | #include "duckdb/common/vector_operations/binary_executor.hpp" |
| 7 | |
| 8 | #include <string.h> |
| 9 | |
| 10 | using namespace std; |
| 11 | |
| 12 | namespace duckdb { |
| 13 | |
| 14 | static void concat_function(DataChunk &args, ExpressionState &state, Vector &result) { |
| 15 | result.vector_type = VectorType::CONSTANT_VECTOR; |
| 16 | // iterate over the vectors to count how large the final string will be |
| 17 | idx_t constant_lengths = 0; |
| 18 | vector<idx_t> result_lengths(args.size(), 0); |
| 19 | for (idx_t col_idx = 0; col_idx < args.column_count(); col_idx++) { |
| 20 | auto &input = args.data[col_idx]; |
| 21 | assert(input.type == TypeId::VARCHAR); |
| 22 | if (input.vector_type == VectorType::CONSTANT_VECTOR) { |
| 23 | if (ConstantVector::IsNull(input)) { |
| 24 | // constant null, skip |
| 25 | continue; |
| 26 | } |
| 27 | auto input_data = ConstantVector::GetData<string_t>(input); |
| 28 | constant_lengths += input_data->GetSize(); |
| 29 | } else { |
| 30 | // non-constant vector: set the result type to a flat vector |
| 31 | result.vector_type = VectorType::FLAT_VECTOR; |
| 32 | // now get the lengths of each of the input elements |
| 33 | VectorData vdata; |
| 34 | input.Orrify(args.size(), vdata); |
| 35 | |
| 36 | auto input_data = (string_t *)vdata.data; |
| 37 | // now add the length of each vector to the result length |
| 38 | for (idx_t i = 0; i < args.size(); i++) { |
| 39 | auto idx = vdata.sel->get_index(i); |
| 40 | if ((*vdata.nullmask)[idx]) { |
| 41 | continue; |
| 42 | } |
| 43 | result_lengths[i] += input_data[idx].GetSize(); |
| 44 | } |
| 45 | } |
| 46 | } |
| 47 | |
| 48 | // first we allocate the empty strings for each of the values |
| 49 | auto result_data = FlatVector::GetData<string_t>(result); |
| 50 | for (idx_t i = 0; i < args.size(); i++) { |
| 51 | // allocate an empty string of the required size |
| 52 | idx_t str_length = constant_lengths + result_lengths[i]; |
| 53 | result_data[i] = StringVector::EmptyString(result, str_length); |
| 54 | // we reuse the result_lengths vector to store the currently appended size |
| 55 | result_lengths[i] = 0; |
| 56 | } |
| 57 | |
| 58 | // now that the empty space for the strings has been allocated, perform the concatenation |
| 59 | for (idx_t col_idx = 0; col_idx < args.column_count(); col_idx++) { |
| 60 | auto &input = args.data[col_idx]; |
| 61 | |
| 62 | // loop over the vector and concat to all results |
| 63 | if (input.vector_type == VectorType::CONSTANT_VECTOR) { |
| 64 | // constant vector |
| 65 | if (ConstantVector::IsNull(input)) { |
| 66 | // constant null, skip |
| 67 | continue; |
| 68 | } |
| 69 | // append the constant vector to each of the strings |
| 70 | auto input_data = ConstantVector::GetData<string_t>(input); |
| 71 | auto input_ptr = input_data->GetData(); |
| 72 | auto input_len = input_data->GetSize(); |
| 73 | for (idx_t i = 0; i < args.size(); i++) { |
| 74 | memcpy(result_data[i].GetData() + result_lengths[i], input_ptr, input_len); |
| 75 | result_lengths[i] += input_len; |
| 76 | } |
| 77 | } else { |
| 78 | // standard vector |
| 79 | VectorData idata; |
| 80 | input.Orrify(args.size(), idata); |
| 81 | |
| 82 | auto input_data = (string_t *)idata.data; |
| 83 | for (idx_t i = 0; i < args.size(); i++) { |
| 84 | auto idx = idata.sel->get_index(i); |
| 85 | if ((*idata.nullmask)[idx]) { |
| 86 | continue; |
| 87 | } |
| 88 | auto input_ptr = input_data[idx].GetData(); |
| 89 | auto input_len = input_data[idx].GetSize(); |
| 90 | memcpy(result_data[i].GetData() + result_lengths[i], input_ptr, input_len); |
| 91 | result_lengths[i] += input_len; |
| 92 | } |
| 93 | } |
| 94 | } |
| 95 | for (idx_t i = 0; i < args.size(); i++) { |
| 96 | result_data[i].Finalize(); |
| 97 | } |
| 98 | } |
| 99 | |
| 100 | static void concat_operator(DataChunk &args, ExpressionState &state, Vector &result) { |
| 101 | BinaryExecutor::Execute<string_t, string_t, string_t, true>( |
| 102 | args.data[0], args.data[1], result, args.size(), [&](string_t a, string_t b) { |
| 103 | auto a_data = a.GetData(); |
| 104 | auto b_data = b.GetData(); |
| 105 | auto a_length = a.GetSize(); |
| 106 | auto b_length = b.GetSize(); |
| 107 | |
| 108 | auto target_length = a_length + b_length; |
| 109 | auto target = StringVector::EmptyString(result, target_length); |
| 110 | auto target_data = target.GetData(); |
| 111 | |
| 112 | memcpy(target_data, a_data, a_length); |
| 113 | memcpy(target_data + a_length, b_data, b_length); |
| 114 | target.Finalize(); |
| 115 | return target; |
| 116 | }); |
| 117 | } |
| 118 | |
| 119 | static void templated_concat_ws(DataChunk &args, string_t *sep_data, const SelectionVector &sep_sel, |
| 120 | const SelectionVector &rsel, idx_t count, Vector &result) { |
| 121 | vector<idx_t> result_lengths(args.size(), 0); |
| 122 | vector<bool> has_results(args.size(), false); |
| 123 | auto orrified_data = unique_ptr<VectorData[]>(new VectorData[args.column_count() - 1]); |
| 124 | for (idx_t col_idx = 1; col_idx < args.column_count(); col_idx++) { |
| 125 | args.data[col_idx].Orrify(args.size(), orrified_data[col_idx - 1]); |
| 126 | } |
| 127 | |
| 128 | // first figure out the lengths |
| 129 | for (idx_t col_idx = 1; col_idx < args.column_count(); col_idx++) { |
| 130 | auto &idata = orrified_data[col_idx - 1]; |
| 131 | |
| 132 | auto input_data = (string_t *)idata.data; |
| 133 | for (idx_t i = 0; i < count; i++) { |
| 134 | auto ridx = rsel.get_index(i); |
| 135 | auto sep_idx = sep_sel.get_index(ridx); |
| 136 | auto idx = idata.sel->get_index(ridx); |
| 137 | if ((*idata.nullmask)[idx]) { |
| 138 | continue; |
| 139 | } |
| 140 | if (has_results[ridx]) { |
| 141 | result_lengths[ridx] += sep_data[sep_idx].GetSize(); |
| 142 | } |
| 143 | result_lengths[ridx] += input_data[idx].GetSize(); |
| 144 | has_results[ridx] = true; |
| 145 | } |
| 146 | } |
| 147 | |
| 148 | // first we allocate the empty strings for each of the values |
| 149 | auto result_data = FlatVector::GetData<string_t>(result); |
| 150 | for (idx_t i = 0; i < count; i++) { |
| 151 | auto ridx = rsel.get_index(i); |
| 152 | // allocate an empty string of the required size |
| 153 | result_data[ridx] = StringVector::EmptyString(result, result_lengths[ridx]); |
| 154 | // we reuse the result_lengths vector to store the currently appended size |
| 155 | result_lengths[ridx] = 0; |
| 156 | has_results[ridx] = false; |
| 157 | } |
| 158 | |
| 159 | // now that the empty space for the strings has been allocated, perform the concatenation |
| 160 | for (idx_t col_idx = 1; col_idx < args.column_count(); col_idx++) { |
| 161 | auto &idata = orrified_data[col_idx - 1]; |
| 162 | auto input_data = (string_t *)idata.data; |
| 163 | for (idx_t i = 0; i < count; i++) { |
| 164 | auto ridx = rsel.get_index(i); |
| 165 | auto sep_idx = sep_sel.get_index(ridx); |
| 166 | auto idx = idata.sel->get_index(ridx); |
| 167 | if ((*idata.nullmask)[idx]) { |
| 168 | continue; |
| 169 | } |
| 170 | if (has_results[ridx]) { |
| 171 | auto sep_size = sep_data[sep_idx].GetSize(); |
| 172 | auto sep_ptr = sep_data[sep_idx].GetData(); |
| 173 | memcpy(result_data[ridx].GetData() + result_lengths[ridx], sep_ptr, sep_size); |
| 174 | result_lengths[ridx] += sep_size; |
| 175 | } |
| 176 | auto input_ptr = input_data[idx].GetData(); |
| 177 | auto input_len = input_data[idx].GetSize(); |
| 178 | memcpy(result_data[ridx].GetData() + result_lengths[ridx], input_ptr, input_len); |
| 179 | result_lengths[ridx] += input_len; |
| 180 | has_results[ridx] = true; |
| 181 | } |
| 182 | } |
| 183 | for (idx_t i = 0; i < count; i++) { |
| 184 | auto ridx = rsel.get_index(i); |
| 185 | result_data[ridx].Finalize(); |
| 186 | } |
| 187 | } |
| 188 | |
| 189 | static void concat_ws_function(DataChunk &args, ExpressionState &state, Vector &result) { |
| 190 | auto &separator = args.data[0]; |
| 191 | VectorData vdata; |
| 192 | separator.Orrify(args.size(), vdata); |
| 193 | |
| 194 | result.vector_type = VectorType::CONSTANT_VECTOR; |
| 195 | for (idx_t col_idx = 0; col_idx < args.column_count(); col_idx++) { |
| 196 | if (args.data[col_idx].vector_type != VectorType::CONSTANT_VECTOR) { |
| 197 | result.vector_type = VectorType::FLAT_VECTOR; |
| 198 | break; |
| 199 | } |
| 200 | } |
| 201 | switch (separator.vector_type) { |
| 202 | case VectorType::CONSTANT_VECTOR: |
| 203 | if (ConstantVector::IsNull(separator)) { |
| 204 | // constant NULL as separator: return constant NULL vector |
| 205 | result.vector_type = VectorType::CONSTANT_VECTOR; |
| 206 | ConstantVector::SetNull(result, true); |
| 207 | return; |
| 208 | } |
| 209 | // no null values |
| 210 | templated_concat_ws(args, (string_t *)vdata.data, *vdata.sel, FlatVector::IncrementalSelectionVector, |
| 211 | args.size(), result); |
| 212 | return; |
| 213 | default: { |
| 214 | // default case: loop over nullmask and create a non-null selection vector |
| 215 | idx_t not_null_count = 0; |
| 216 | SelectionVector not_null_vector(STANDARD_VECTOR_SIZE); |
| 217 | auto &result_nullmask = FlatVector::Nullmask(result); |
| 218 | for (idx_t i = 0; i < args.size(); i++) { |
| 219 | if ((*vdata.nullmask)[vdata.sel->get_index(i)]) { |
| 220 | result_nullmask[i] = true; |
| 221 | } else { |
| 222 | not_null_vector.set_index(not_null_count++, i); |
| 223 | } |
| 224 | } |
| 225 | templated_concat_ws(args, (string_t *)vdata.data, *vdata.sel, not_null_vector, not_null_count, result); |
| 226 | return; |
| 227 | } |
| 228 | } |
| 229 | } |
| 230 | |
| 231 | void ConcatFun::RegisterFunction(BuiltinFunctions &set) { |
| 232 | // the concat operator and concat function have different behavior regarding NULLs |
| 233 | // this is strange but seems consistent with postgresql and mysql |
| 234 | // (sqlite does not support the concat function, only the concat operator) |
| 235 | |
| 236 | // the concat operator behaves as one would expect: any NULL value present results in a NULL |
| 237 | // i.e. NULL || 'hello' = NULL |
| 238 | // the concat function, however, treats NULL values as an empty string |
| 239 | // i.e. concat(NULL, 'hello') = 'hello' |
| 240 | // concat_ws functions similarly to the concat function, except the result is NULL if the separator is NULL |
| 241 | // if the separator is not NULL, however, NULL values are counted as empty string |
| 242 | // there is one separate rule: there are no separators added between NULL values |
| 243 | // so the NULL value and empty string are different! |
| 244 | // e.g.: |
| 245 | // concat_ws(',', NULL, NULL) = "" |
| 246 | // concat_ws(',', '', '') = "," |
| 247 | ScalarFunction concat = ScalarFunction("concat" , {SQLType::VARCHAR}, SQLType::VARCHAR, concat_function); |
| 248 | concat.varargs = SQLType::VARCHAR; |
| 249 | set.AddFunction(concat); |
| 250 | |
| 251 | ScalarFunctionSet concat_op("||" ); |
| 252 | concat_op.AddFunction(ScalarFunction({SQLType::VARCHAR, SQLType::VARCHAR}, SQLType::VARCHAR, concat_operator)); |
| 253 | concat_op.AddFunction(ScalarFunction({SQLType::BLOB, SQLType::BLOB}, SQLType::BLOB, concat_operator)); |
| 254 | set.AddFunction(concat_op); |
| 255 | |
| 256 | |
| 257 | ScalarFunction concat_ws = |
| 258 | ScalarFunction("concat_ws" , {SQLType::VARCHAR, SQLType::VARCHAR}, SQLType::VARCHAR, concat_ws_function); |
| 259 | concat_ws.varargs = SQLType::VARCHAR; |
| 260 | set.AddFunction(concat_ws); |
| 261 | } |
| 262 | |
| 263 | } // namespace duckdb |
| 264 | |