/root/doris/be/src/vec/functions/function.h
Line | Count | Source (jump to first uncovered line) |
1 | | // Licensed to the Apache Software Foundation (ASF) under one |
2 | | // or more contributor license agreements. See the NOTICE file |
3 | | // distributed with this work for additional information |
4 | | // regarding copyright ownership. The ASF licenses this file |
5 | | // to you under the Apache License, Version 2.0 (the |
6 | | // "License"); you may not use this file except in compliance |
7 | | // with the License. You may obtain a copy of the License at |
8 | | // |
9 | | // http://www.apache.org/licenses/LICENSE-2.0 |
10 | | // |
11 | | // Unless required by applicable law or agreed to in writing, |
12 | | // software distributed under the License is distributed on an |
13 | | // "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY |
14 | | // KIND, either express or implied. See the License for the |
15 | | // specific language governing permissions and limitations |
16 | | // under the License. |
17 | | // This file is copied from |
18 | | // https://github.com/ClickHouse/ClickHouse/blob/master/src/Functions/IFunction.h |
19 | | // and modified by Doris |
20 | | |
21 | | #pragma once |
22 | | |
23 | | #include <fmt/format.h> |
24 | | #include <glog/logging.h> |
25 | | #include <stddef.h> |
26 | | |
27 | | #include <memory> |
28 | | #include <ostream> |
29 | | #include <string> |
30 | | #include <utility> |
31 | | |
32 | | #include "common/exception.h" |
33 | | #include "common/status.h" |
34 | | #include "olap/rowset/segment_v2/inverted_index_reader.h" |
35 | | #include "udf/udf.h" |
36 | | #include "vec/core/block.h" |
37 | | #include "vec/core/column_numbers.h" |
38 | | #include "vec/core/column_with_type_and_name.h" |
39 | | #include "vec/core/columns_with_type_and_name.h" |
40 | | #include "vec/core/types.h" |
41 | | #include "vec/data_types/data_type.h" |
42 | | #include "vec/data_types/data_type_nullable.h" |
43 | | |
44 | | namespace doris::segment_v2 { |
45 | | struct FuncExprParams; |
46 | | } // namespace doris::segment_v2 |
47 | | |
48 | | namespace doris::vectorized { |
49 | | |
50 | | struct FunctionAttr { |
51 | | bool enable_decimal256 {false}; |
52 | | bool new_is_ip_address_in_range {false}; |
53 | | }; |
54 | | |
55 | | #define RETURN_REAL_TYPE_FOR_DATEV2_FUNCTION(TYPE) \ |
56 | 99 | bool is_nullable = false; \ |
57 | 99 | bool is_datev2 = false; \ |
58 | 150 | for (auto it : arguments) { \ |
59 | 150 | is_nullable = is_nullable || it.type->is_nullable(); \ |
60 | 150 | is_datev2 = is_datev2 || WhichDataType(remove_nullable(it.type)).is_date_v2() || \ |
61 | 150 | WhichDataType(remove_nullable(it.type)).is_date_time_v2(); \ |
62 | 150 | } \ |
63 | 103 | return is_nullable || !is_datev2 ? make_nullable(std::make_shared<TYPE>()) \ |
64 | 99 | : std::make_shared<TYPE>(); |
65 | | |
66 | | #define SET_NULLMAP_IF_FALSE(EXPR) \ |
67 | 0 | if (!EXPR) [[unlikely]] { \ |
68 | 0 | null_map[i] = true; \ |
69 | 0 | } |
70 | | |
71 | | class Field; |
72 | | class VExpr; |
73 | | |
74 | | // Only use dispose the variadic argument |
75 | | template <typename T> |
76 | | auto has_variadic_argument_types(T&& arg) -> decltype(T::get_variadic_argument_types()) {}; |
77 | | void has_variadic_argument_types(...); |
78 | | |
79 | | template <typename T> |
80 | | concept HasGetVariadicArgumentTypesImpl = requires(T t) { |
81 | | { t.get_variadic_argument_types_impl() } -> std::same_as<DataTypes>; |
82 | | }; |
83 | | |
84 | | bool have_null_column(const Block& block, const ColumnNumbers& args); |
85 | | bool have_null_column(const ColumnsWithTypeAndName& args); |
86 | | |
87 | | /// The simplest executable object. |
88 | | /// Motivation: |
89 | | /// * Prepare something heavy once before main execution loop instead of doing it for each block. |
90 | | /// * Provide const interface for IFunctionBase (later). |
91 | | class IPreparedFunction { |
92 | | public: |
93 | 30.2k | virtual ~IPreparedFunction() = default; |
94 | | |
95 | | /// Get the main function name. |
96 | | virtual String get_name() const = 0; |
97 | | |
98 | | virtual Status execute(FunctionContext* context, Block& block, const ColumnNumbers& arguments, |
99 | | size_t result, size_t input_rows_count, bool dry_run) const = 0; |
100 | | }; |
101 | | |
102 | | using PreparedFunctionPtr = std::shared_ptr<IPreparedFunction>; |
103 | | |
104 | | class PreparedFunctionImpl : public IPreparedFunction { |
105 | | public: |
106 | | Status execute(FunctionContext* context, Block& block, const ColumnNumbers& arguments, |
107 | | size_t result, size_t input_rows_count, bool dry_run = false) const final; |
108 | | |
109 | | /** If the function have non-zero number of arguments, |
110 | | * and if all arguments are constant, that we could automatically provide default implementation: |
111 | | * arguments are converted to ordinary columns with single value which is not const, then function is executed as usual, |
112 | | * and then the result is converted to constant column. |
113 | | */ |
114 | 20.2k | virtual bool use_default_implementation_for_constants() const { return true; } |
115 | | |
116 | | /** If use_default_implementation_for_nulls() is true, after execute the function, |
117 | | * whether need to replace the nested data of null data to the default value. |
118 | | * E.g. for binary arithmetic exprs, need return true to avoid false overflow. |
119 | | */ |
120 | 0 | virtual bool need_replace_null_data_to_default() const { return false; } |
121 | | |
122 | | protected: |
123 | | virtual Status execute_impl_dry_run(FunctionContext* context, Block& block, |
124 | | const ColumnNumbers& arguments, size_t result, |
125 | 0 | size_t input_rows_count) const { |
126 | 0 | return execute_impl(context, block, arguments, result, input_rows_count); |
127 | 0 | } |
128 | | |
129 | | virtual Status execute_impl(FunctionContext* context, Block& block, |
130 | | const ColumnNumbers& arguments, size_t result, |
131 | | size_t input_rows_count) const = 0; |
132 | | |
133 | | /** Default implementation in presence of Nullable arguments or NULL constants as arguments is the following: |
134 | | * if some of arguments are NULL constants then return NULL constant, |
135 | | * if some of arguments are Nullable, then execute function as usual for block, |
136 | | * where Nullable columns are substituted with nested columns (they have arbitrary values in rows corresponding to NULL value) |
137 | | * and wrap result in Nullable column where NULLs are in all rows where any of arguments are NULL. |
138 | | */ |
139 | 0 | virtual bool use_default_implementation_for_nulls() const { return true; } |
140 | | |
141 | | /** If function arguments has single low cardinality column and all other arguments are constants, call function on nested column. |
142 | | * Otherwise, convert all low cardinality columns to ordinary columns. |
143 | | * Returns ColumnLowCardinality if at least one argument is ColumnLowCardinality. |
144 | | */ |
145 | 0 | virtual bool use_default_implementation_for_low_cardinality_columns() const { return true; } |
146 | | |
147 | | /** Some arguments could remain constant during this implementation. |
148 | | * Every argument required const must write here and no checks elsewhere. |
149 | | */ |
150 | 0 | virtual ColumnNumbers get_arguments_that_are_always_constant() const { return {}; } |
151 | | |
152 | | private: |
153 | | Status default_implementation_for_nulls(FunctionContext* context, Block& block, |
154 | | const ColumnNumbers& args, size_t result, |
155 | | size_t input_rows_count, bool dry_run, |
156 | | bool* executed) const; |
157 | | Status default_implementation_for_constant_arguments(FunctionContext* context, Block& block, |
158 | | const ColumnNumbers& args, size_t result, |
159 | | size_t input_rows_count, bool dry_run, |
160 | | bool* executed) const; |
161 | | Status execute_without_low_cardinality_columns(FunctionContext* context, Block& block, |
162 | | const ColumnNumbers& arguments, size_t result, |
163 | | size_t input_rows_count, bool dry_run) const; |
164 | | Status _execute_skipped_constant_deal(FunctionContext* context, Block& block, |
165 | | const ColumnNumbers& args, size_t result, |
166 | | size_t input_rows_count, bool dry_run) const; |
167 | | }; |
168 | | |
169 | | /// Function with known arguments and return type. |
170 | | class IFunctionBase { |
171 | | public: |
172 | 30.1k | virtual ~IFunctionBase() = default; |
173 | | |
174 | | /// Get the main function name. |
175 | | virtual String get_name() const = 0; |
176 | | |
177 | | virtual const DataTypes& get_argument_types() const = 0; |
178 | | virtual const DataTypePtr& get_return_type() const = 0; |
179 | | |
180 | | /// Do preparations and return executable. |
181 | | /// sample_block should contain data types of arguments and values of constants, if relevant. |
182 | | virtual PreparedFunctionPtr prepare(FunctionContext* context, const Block& sample_block, |
183 | | const ColumnNumbers& arguments, size_t result) const = 0; |
184 | | |
185 | | /// Override this when function need to store state in the `FunctionContext`, or do some |
186 | | /// preparation work according to information from `FunctionContext`. |
187 | 260 | virtual Status open(FunctionContext* context, FunctionContext::FunctionStateScope scope) { |
188 | 260 | return Status::OK(); |
189 | 260 | } |
190 | | |
191 | | /// TODO: make const |
192 | | virtual Status execute(FunctionContext* context, Block& block, const ColumnNumbers& arguments, |
193 | 14.2k | size_t result, size_t input_rows_count, bool dry_run = false) const { |
194 | 14.2k | return prepare(context, block, arguments, result) |
195 | 14.2k | ->execute(context, block, arguments, result, input_rows_count, dry_run); |
196 | 14.2k | } |
197 | | |
198 | | virtual Status evaluate_inverted_index( |
199 | | const ColumnsWithTypeAndName& arguments, |
200 | | const std::vector<vectorized::IndexFieldNameAndTypePair>& data_type_with_names, |
201 | | std::vector<segment_v2::InvertedIndexIterator*> iterators, uint32_t num_rows, |
202 | 0 | segment_v2::InvertedIndexResultBitmap& bitmap_result) const { |
203 | 0 | return Status::OK(); |
204 | 0 | } |
205 | | |
206 | | /// Do cleaning work when function is finished, i.e., release state variables in the |
207 | | /// `FunctionContext` which are registered in `prepare` phase. |
208 | 21.8k | virtual Status close(FunctionContext* context, FunctionContext::FunctionStateScope scope) { |
209 | 21.8k | return Status::OK(); |
210 | 21.8k | } |
211 | | |
212 | | virtual bool is_use_default_implementation_for_constants() const = 0; |
213 | | |
214 | 0 | virtual bool is_udf_function() const { return false; } |
215 | | |
216 | | /// The property of monotonicity for a certain range. |
217 | | struct Monotonicity { |
218 | | bool is_monotonic = false; /// Is the function monotonous (nondecreasing or nonincreasing). |
219 | | bool is_positive = |
220 | | true; /// true if the function is nondecreasing, false, if notincreasing. If is_monotonic = false, then it does not matter. |
221 | | bool is_always_monotonic = |
222 | | false; /// Is true if function is monotonic on the whole input range I |
223 | | |
224 | | Monotonicity(bool is_monotonic_ = false, bool is_positive_ = true, |
225 | | bool is_always_monotonic_ = false) |
226 | | : is_monotonic(is_monotonic_), |
227 | | is_positive(is_positive_), |
228 | 0 | is_always_monotonic(is_always_monotonic_) {} |
229 | | }; |
230 | | |
231 | | /** Get information about monotonicity on a range of values. Call only if hasInformationAboutMonotonicity. |
232 | | * NULL can be passed as one of the arguments. This means that the corresponding range is unlimited on the left or on the right. |
233 | | */ |
234 | | virtual Monotonicity get_monotonicity_for_range(const IDataType& /*type*/, |
235 | | const Field& /*left*/, |
236 | 0 | const Field& /*right*/) const { |
237 | 0 | throw doris::Exception(ErrorCode::INTERNAL_ERROR, |
238 | 0 | "Function {} has no information about its monotonicity.", |
239 | 0 | get_name()); |
240 | 0 | return Monotonicity {}; |
241 | 0 | } |
242 | | |
243 | 0 | virtual bool can_push_down_to_index() const { return false; } |
244 | | }; |
245 | | |
246 | | using FunctionBasePtr = std::shared_ptr<IFunctionBase>; |
247 | | |
248 | | /// Creates IFunctionBase from argument types list. |
249 | | class IFunctionBuilder { |
250 | | public: |
251 | 32.1k | virtual ~IFunctionBuilder() = default; |
252 | | |
253 | | /// Get the main function name. |
254 | | virtual String get_name() const = 0; |
255 | | |
256 | | /// Override and return true if function could take different number of arguments. |
257 | | virtual bool is_variadic() const = 0; |
258 | | |
259 | | /// For non-variadic functions, return number of arguments; otherwise return zero (that should be ignored). |
260 | | virtual size_t get_number_of_arguments() const = 0; |
261 | | |
262 | | /// Throw if number of arguments is incorrect. Default implementation will check only in non-variadic case. |
263 | | virtual void check_number_of_arguments(size_t number_of_arguments) const = 0; |
264 | | |
265 | | /// Check arguments and return IFunctionBase. |
266 | | virtual FunctionBasePtr build(const ColumnsWithTypeAndName& arguments, |
267 | | const DataTypePtr& return_type) const = 0; |
268 | | |
269 | | /// For higher-order functions (functions, that have lambda expression as at least one argument). |
270 | | /// You pass data types with empty DataTypeFunction for lambda arguments. |
271 | | /// This function will replace it with DataTypeFunction containing actual types. |
272 | | virtual DataTypes get_variadic_argument_types() const = 0; |
273 | | |
274 | | /// Returns indexes of arguments, that must be ColumnConst |
275 | | virtual ColumnNumbers get_arguments_that_are_always_constant() const = 0; |
276 | | }; |
277 | | |
278 | | using FunctionBuilderPtr = std::shared_ptr<IFunctionBuilder>; |
279 | | |
280 | 1 | inline std::string get_types_string(const ColumnsWithTypeAndName& arguments) { |
281 | 1 | std::string types; |
282 | 1 | for (const auto& argument : arguments) { |
283 | 0 | if (!types.empty()) { |
284 | 0 | types += ", "; |
285 | 0 | } |
286 | 0 | types += argument.type->get_name(); |
287 | 0 | } |
288 | 1 | return types; |
289 | 1 | } |
290 | | |
291 | | /// used in function_factory. when we register a function, save a builder. to get a function, to get a builder. |
292 | | /// will use DefaultFunctionBuilder as the default builder in function's registration if we didn't explicitly specify. |
293 | | class FunctionBuilderImpl : public IFunctionBuilder { |
294 | | public: |
295 | | FunctionBasePtr build(const ColumnsWithTypeAndName& arguments, |
296 | 14.1k | const DataTypePtr& return_type) const final { |
297 | 14.1k | const DataTypePtr& func_return_type = get_return_type(arguments); |
298 | 14.1k | if (func_return_type == nullptr) { |
299 | 1 | throw doris::Exception( |
300 | 1 | ErrorCode::INTERNAL_ERROR, |
301 | 1 | "function return type check failed, function_name={}, " |
302 | 1 | "expect_return_type={}, real_return_type is nullptr, input_arguments={}", |
303 | 1 | get_name(), return_type->get_name(), get_types_string(arguments)); |
304 | 1 | } |
305 | | |
306 | | // check return types equal. |
307 | 14.1k | if (!(return_type->equals(*func_return_type) || |
308 | | // For null constant argument, `get_return_type` would return |
309 | | // Nullable<DataTypeNothing> when `use_default_implementation_for_nulls` is true. |
310 | 14.1k | (return_type->is_nullable() && func_return_type->is_nullable() && |
311 | 2 | is_nothing(((DataTypeNullable*)func_return_type.get())->get_nested_type())) || |
312 | 14.1k | is_date_or_datetime_or_decimal(return_type, func_return_type) || |
313 | 14.1k | is_array_nested_type_date_or_datetime_or_decimal(return_type, func_return_type))) { |
314 | 0 | LOG_WARNING( |
315 | 0 | "function return type check failed, function_name={}, " |
316 | 0 | "expect_return_type={}, real_return_type={}, input_arguments={}", |
317 | 0 | get_name(), return_type->get_name(), func_return_type->get_name(), |
318 | 0 | get_types_string(arguments)); |
319 | 0 | return nullptr; |
320 | 0 | } |
321 | 14.1k | return build_impl(arguments, return_type); |
322 | 14.1k | } |
323 | | |
324 | 10.0k | bool is_variadic() const override { return false; } |
325 | | |
326 | | // Default implementation. Will check only in non-variadic case. |
327 | | void check_number_of_arguments(size_t number_of_arguments) const override; |
328 | | // the return type should be same with what FE plans. |
329 | | // it returns: `get_return_type_impl` if `use_default_implementation_for_nulls` = false |
330 | | // `get_return_type_impl` warpped in NULL if `use_default_implementation_for_nulls` = true and input has NULL |
331 | | DataTypePtr get_return_type(const ColumnsWithTypeAndName& arguments) const; |
332 | | |
333 | 1.02k | DataTypes get_variadic_argument_types() const override { |
334 | 1.02k | return get_variadic_argument_types_impl(); |
335 | 1.02k | } |
336 | | |
337 | 0 | ColumnNumbers get_arguments_that_are_always_constant() const override { return {}; } |
338 | | |
339 | | protected: |
340 | | // Get the result type by argument type. If the function does not apply to these arguments, throw an exception. |
341 | | // the get_return_type_impl and its overrides should only return the nested type if `use_default_implementation_for_nulls` is true. |
342 | | // whether to wrap in nullable type will be automatically decided. |
343 | 13.9k | virtual DataTypePtr get_return_type_impl(const ColumnsWithTypeAndName& arguments) const { |
344 | 13.9k | DataTypes data_types(arguments.size()); |
345 | 47.0k | for (size_t i = 0; i < arguments.size(); ++i) { |
346 | 33.1k | data_types[i] = arguments[i].type; |
347 | 33.1k | } |
348 | 13.9k | return get_return_type_impl(data_types); |
349 | 13.9k | } |
350 | | |
351 | 0 | virtual DataTypePtr get_return_type_impl(const DataTypes& /*arguments*/) const { |
352 | 0 | throw doris::Exception(ErrorCode::NOT_IMPLEMENTED_ERROR, |
353 | 0 | "get_return_type is not implemented for {}", get_name()); |
354 | 0 | return nullptr; |
355 | 0 | } |
356 | | |
357 | | /** If use_default_implementation_for_nulls() is true, than change arguments for get_return_type() and build_impl(): |
358 | | * if some of arguments are Nullable(Nothing) then don't call get_return_type(), call build_impl() with return_type = Nullable(Nothing), |
359 | | * if some of arguments are Nullable, then: |
360 | | * - Nullable types are substituted with nested types for get_return_type() function |
361 | | * - WRAP get_return_type() RESULT IN NULLABLE type and pass to build_impl |
362 | | * |
363 | | * Otherwise build returns build_impl(arguments, get_return_type(arguments)); |
364 | | */ |
365 | 0 | virtual bool use_default_implementation_for_nulls() const { return true; } |
366 | | |
367 | 0 | virtual bool need_replace_null_data_to_default() const { return false; } |
368 | | |
369 | | /** If use_default_implementation_for_nulls() is true, than change arguments for get_return_type() and build_impl(). |
370 | | * If function arguments has low cardinality types, convert them to ordinary types. |
371 | | * get_return_type returns ColumnLowCardinality if at least one argument type is ColumnLowCardinality. |
372 | | */ |
373 | 0 | virtual bool use_default_implementation_for_low_cardinality_columns() const { return true; } |
374 | | |
375 | | /// return a real function object to execute. called in build(...). |
376 | | virtual FunctionBasePtr build_impl(const ColumnsWithTypeAndName& arguments, |
377 | | const DataTypePtr& return_type) const = 0; |
378 | | |
379 | 303 | virtual DataTypes get_variadic_argument_types_impl() const { return DataTypes(); } |
380 | | |
381 | | private: |
382 | | DataTypePtr get_return_type_without_low_cardinality( |
383 | | const ColumnsWithTypeAndName& arguments) const; |
384 | | |
385 | | bool is_date_or_datetime_or_decimal(const DataTypePtr& return_type, |
386 | | const DataTypePtr& func_return_type) const; |
387 | | bool is_array_nested_type_date_or_datetime_or_decimal( |
388 | | const DataTypePtr& return_type, const DataTypePtr& func_return_type) const; |
389 | | }; |
390 | | |
391 | | /// Previous function interface. |
392 | | class IFunction : public std::enable_shared_from_this<IFunction>, |
393 | | public FunctionBuilderImpl, |
394 | | public IFunctionBase, |
395 | | public PreparedFunctionImpl { |
396 | | public: |
397 | | String get_name() const override = 0; |
398 | | |
399 | | /// Notice: We should not change the column in the block, because the column may be shared by multiple expressions or exec nodes. |
400 | | virtual Status execute_impl(FunctionContext* context, Block& block, |
401 | | const ColumnNumbers& arguments, size_t result, |
402 | | size_t input_rows_count) const override = 0; |
403 | | |
404 | | /// Override this functions to change default implementation behavior. See details in IMyFunction. |
405 | 30.1k | bool use_default_implementation_for_nulls() const override { return true; } |
406 | | |
407 | 5.76k | bool need_replace_null_data_to_default() const override { return false; } |
408 | | |
409 | 14.0k | bool use_default_implementation_for_low_cardinality_columns() const override { return true; } |
410 | | |
411 | | /// all constancy check should use this function to do automatically |
412 | 19.9k | ColumnNumbers get_arguments_that_are_always_constant() const override { return {}; } |
413 | | |
414 | 2 | bool is_use_default_implementation_for_constants() const override { |
415 | 2 | return use_default_implementation_for_constants(); |
416 | 2 | } |
417 | | |
418 | | using PreparedFunctionImpl::execute; |
419 | | using PreparedFunctionImpl::execute_impl_dry_run; |
420 | | using FunctionBuilderImpl::get_return_type_impl; |
421 | | using FunctionBuilderImpl::get_variadic_argument_types_impl; |
422 | | using FunctionBuilderImpl::get_return_type; |
423 | | |
424 | | [[noreturn]] PreparedFunctionPtr prepare(FunctionContext* context, |
425 | | const Block& /*sample_block*/, |
426 | | const ColumnNumbers& /*arguments*/, |
427 | 0 | size_t /*result*/) const final { |
428 | 0 | throw doris::Exception(ErrorCode::NOT_IMPLEMENTED_ERROR, |
429 | 0 | "prepare is not implemented for IFunction {}", get_name()); |
430 | 0 | __builtin_unreachable(); |
431 | 0 | } |
432 | | |
433 | 21.1k | Status open(FunctionContext* context, FunctionContext::FunctionStateScope scope) override { |
434 | 21.1k | return Status::OK(); |
435 | 21.1k | } |
436 | | |
437 | 0 | [[noreturn]] const DataTypes& get_argument_types() const final { |
438 | 0 | throw doris::Exception(ErrorCode::NOT_IMPLEMENTED_ERROR, |
439 | 0 | "get_argument_types is not implemented for IFunction {}", |
440 | 0 | get_name()); |
441 | 0 | __builtin_unreachable(); |
442 | 0 | } |
443 | | |
444 | 0 | [[noreturn]] const DataTypePtr& get_return_type() const final { |
445 | 0 | throw doris::Exception(ErrorCode::NOT_IMPLEMENTED_ERROR, |
446 | 0 | "get_return_type is not implemented for IFunction {}", get_name()); |
447 | 0 | __builtin_unreachable(); |
448 | 0 | } |
449 | | |
450 | | protected: |
451 | | FunctionBasePtr build_impl(const ColumnsWithTypeAndName& /*arguments*/, |
452 | 0 | const DataTypePtr& /*return_type*/) const final { |
453 | 0 | throw doris::Exception(ErrorCode::NOT_IMPLEMENTED_ERROR, |
454 | 0 | "build_impl is not implemented for IFunction {}", get_name()); |
455 | 0 | __builtin_unreachable(); |
456 | 0 | return {}; |
457 | 0 | } |
458 | | }; |
459 | | |
460 | | /// Wrappers over IFunction. If we (default)use DefaultFunction as wrapper, all function execution will go through this. |
461 | | |
462 | | class DefaultExecutable final : public PreparedFunctionImpl { |
463 | | public: |
464 | | explicit DefaultExecutable(std::shared_ptr<IFunction> function_) |
465 | 14.1k | : function(std::move(function_)) {} |
466 | | |
467 | 0 | String get_name() const override { return function->get_name(); } |
468 | | |
469 | | protected: |
470 | | Status execute_impl(FunctionContext* context, Block& block, const ColumnNumbers& arguments, |
471 | 11.0k | size_t result, size_t input_rows_count) const final { |
472 | 11.0k | return function->execute_impl(context, block, arguments, result, input_rows_count); |
473 | 11.0k | } |
474 | | |
475 | | Status evaluate_inverted_index( |
476 | | const ColumnsWithTypeAndName& arguments, |
477 | | const std::vector<vectorized::IndexFieldNameAndTypePair>& data_type_with_names, |
478 | | std::vector<segment_v2::InvertedIndexIterator*> iterators, uint32_t num_rows, |
479 | 0 | segment_v2::InvertedIndexResultBitmap& bitmap_result) const { |
480 | 0 | return function->evaluate_inverted_index(arguments, data_type_with_names, iterators, |
481 | 0 | num_rows, bitmap_result); |
482 | 0 | } |
483 | | |
484 | | Status execute_impl_dry_run(FunctionContext* context, Block& block, |
485 | | const ColumnNumbers& arguments, size_t result, |
486 | 0 | size_t input_rows_count) const final { |
487 | 0 | return function->execute_impl_dry_run(context, block, arguments, result, input_rows_count); |
488 | 0 | } |
489 | 20.1k | bool use_default_implementation_for_nulls() const final { |
490 | 20.1k | return function->use_default_implementation_for_nulls(); |
491 | 20.1k | } |
492 | 5.95k | bool need_replace_null_data_to_default() const final { |
493 | 5.95k | return function->need_replace_null_data_to_default(); |
494 | 5.95k | } |
495 | 20.1k | bool use_default_implementation_for_constants() const final { |
496 | 20.1k | return function->use_default_implementation_for_constants(); |
497 | 20.1k | } |
498 | 0 | bool use_default_implementation_for_low_cardinality_columns() const final { |
499 | 0 | return function->use_default_implementation_for_low_cardinality_columns(); |
500 | 0 | } |
501 | 20.1k | ColumnNumbers get_arguments_that_are_always_constant() const final { |
502 | 20.1k | return function->get_arguments_that_are_always_constant(); |
503 | 20.1k | } |
504 | | |
505 | | private: |
506 | | std::shared_ptr<IFunction> function; |
507 | | }; |
508 | | |
509 | | /* |
510 | | * when we register a function which didn't specify its base(i.e. inherited from IFunction), actually we use this as a wrapper. |
511 | | * it saves real implementation as `function`. |
512 | | */ |
513 | | class DefaultFunction final : public IFunctionBase { |
514 | | public: |
515 | | DefaultFunction(std::shared_ptr<IFunction> function_, DataTypes arguments_, |
516 | | DataTypePtr return_type_) |
517 | | : function(std::move(function_)), |
518 | | arguments(std::move(arguments_)), |
519 | 14.0k | return_type(std::move(return_type_)) {} |
520 | | |
521 | 0 | String get_name() const override { return function->get_name(); } |
522 | | |
523 | 0 | const DataTypes& get_argument_types() const override { return arguments; } |
524 | 0 | const DataTypePtr& get_return_type() const override { return return_type; } |
525 | | |
526 | | // return a default wrapper for IFunction. |
527 | | PreparedFunctionPtr prepare(FunctionContext* context, const Block& /*sample_block*/, |
528 | | const ColumnNumbers& /*arguments*/, |
529 | 14.1k | size_t /*result*/) const override { |
530 | 14.1k | return std::make_shared<DefaultExecutable>(function); |
531 | 14.1k | } |
532 | | |
533 | 21.8k | Status open(FunctionContext* context, FunctionContext::FunctionStateScope scope) override { |
534 | 21.8k | return function->open(context, scope); |
535 | 21.8k | } |
536 | | |
537 | 21.8k | Status close(FunctionContext* context, FunctionContext::FunctionStateScope scope) override { |
538 | 21.8k | return function->close(context, scope); |
539 | 21.8k | } |
540 | | |
541 | | Status evaluate_inverted_index( |
542 | | const ColumnsWithTypeAndName& args, |
543 | | const std::vector<vectorized::IndexFieldNameAndTypePair>& data_type_with_names, |
544 | | std::vector<segment_v2::InvertedIndexIterator*> iterators, uint32_t num_rows, |
545 | 0 | segment_v2::InvertedIndexResultBitmap& bitmap_result) const override { |
546 | 0 | return function->evaluate_inverted_index(args, data_type_with_names, iterators, num_rows, |
547 | 0 | bitmap_result); |
548 | 0 | } |
549 | | |
550 | | IFunctionBase::Monotonicity get_monotonicity_for_range(const IDataType& type, const Field& left, |
551 | 0 | const Field& right) const override { |
552 | 0 | return function->get_monotonicity_for_range(type, left, right); |
553 | 0 | } |
554 | | |
555 | 2 | bool is_use_default_implementation_for_constants() const override { |
556 | 2 | return function->is_use_default_implementation_for_constants(); |
557 | 2 | } |
558 | | |
559 | 0 | bool can_push_down_to_index() const override { return function->can_push_down_to_index(); } |
560 | | |
561 | | private: |
562 | | std::shared_ptr<IFunction> function; |
563 | | DataTypes arguments; |
564 | | DataTypePtr return_type; |
565 | | }; |
566 | | |
567 | | class DefaultFunctionBuilder : public FunctionBuilderImpl { |
568 | | public: |
569 | | explicit DefaultFunctionBuilder(std::shared_ptr<IFunction> function_) |
570 | 15.9k | : function(std::move(function_)) {} |
571 | | |
572 | 14.0k | void check_number_of_arguments(size_t number_of_arguments) const override { |
573 | 14.0k | return function->check_number_of_arguments(number_of_arguments); |
574 | 14.0k | } |
575 | | |
576 | 333 | String get_name() const override { return function->get_name(); } |
577 | 882 | bool is_variadic() const override { return function->is_variadic(); } |
578 | 0 | size_t get_number_of_arguments() const override { return function->get_number_of_arguments(); } |
579 | | |
580 | 0 | ColumnNumbers get_arguments_that_are_always_constant() const override { |
581 | 0 | return function->get_arguments_that_are_always_constant(); |
582 | 0 | } |
583 | | |
584 | | protected: |
585 | 0 | DataTypePtr get_return_type_impl(const DataTypes& arguments) const override { |
586 | 0 | return function->get_return_type_impl(arguments); |
587 | 0 | } |
588 | 14.0k | DataTypePtr get_return_type_impl(const ColumnsWithTypeAndName& arguments) const override { |
589 | 14.0k | return function->get_return_type_impl(arguments); |
590 | 14.0k | } |
591 | | |
592 | 14.0k | bool use_default_implementation_for_nulls() const override { |
593 | 14.0k | return function->use_default_implementation_for_nulls(); |
594 | 14.0k | } |
595 | | |
596 | 0 | bool need_replace_null_data_to_default() const override { |
597 | 0 | return function->need_replace_null_data_to_default(); |
598 | 0 | } |
599 | 14.0k | bool use_default_implementation_for_low_cardinality_columns() const override { |
600 | 14.0k | return function->use_default_implementation_for_low_cardinality_columns(); |
601 | 14.0k | } |
602 | | |
603 | | FunctionBasePtr build_impl(const ColumnsWithTypeAndName& arguments, |
604 | 14.0k | const DataTypePtr& return_type) const override { |
605 | 14.0k | DataTypes data_types(arguments.size()); |
606 | 47.3k | for (size_t i = 0; i < arguments.size(); ++i) { |
607 | 33.3k | data_types[i] = arguments[i].type; |
608 | 33.3k | } |
609 | 14.0k | return std::make_shared<DefaultFunction>(function, data_types, return_type); |
610 | 14.0k | } |
611 | | |
612 | 1.01k | DataTypes get_variadic_argument_types_impl() const override { |
613 | 1.01k | return function->get_variadic_argument_types_impl(); |
614 | 1.01k | } |
615 | | |
616 | | private: |
617 | | std::shared_ptr<IFunction> function; |
618 | | }; |
619 | | |
620 | | using FunctionPtr = std::shared_ptr<IFunction>; |
621 | | |
622 | | /** Return ColumnNullable of src, with null map as OR-ed null maps of args columns in blocks. |
623 | | * Or ColumnConst(ColumnNullable) if the result is always NULL or if the result is constant and always not NULL. |
624 | | */ |
625 | | ColumnPtr wrap_in_nullable(const ColumnPtr& src, const Block& block, const ColumnNumbers& args, |
626 | | size_t result, size_t input_rows_count); |
627 | | |
628 | | #define NUMERIC_TYPE_TO_COLUMN_TYPE(M) \ |
629 | 165 | M(UInt8, ColumnUInt8) \ |
630 | 165 | M(Int8, ColumnInt8) \ |
631 | 163 | M(Int16, ColumnInt16) \ |
632 | 159 | M(Int32, ColumnInt32) \ |
633 | 155 | M(Int64, ColumnInt64) \ |
634 | 86 | M(Int128, ColumnInt128) \ |
635 | 82 | M(Float32, ColumnFloat32) \ |
636 | 80 | M(Float64, ColumnFloat64) |
637 | | |
638 | | #define DECIMAL_TYPE_TO_COLUMN_TYPE(M) \ |
639 | 80 | M(Decimal32, ColumnDecimal<Decimal32>) \ |
640 | 80 | M(Decimal64, ColumnDecimal<Decimal64>) \ |
641 | 80 | M(Decimal128V2, ColumnDecimal<Decimal128V2>) \ |
642 | 78 | M(Decimal128V3, ColumnDecimal<Decimal128V3>) \ |
643 | 76 | M(Decimal256, ColumnDecimal<Decimal256>) |
644 | | |
645 | | #define STRING_TYPE_TO_COLUMN_TYPE(M) \ |
646 | 74 | M(String, ColumnString) \ |
647 | 74 | M(JSONB, ColumnString) |
648 | | |
649 | | #define TIME_TYPE_TO_COLUMN_TYPE(M) \ |
650 | 70 | M(Date, ColumnInt64) \ |
651 | 68 | M(DateTime, ColumnInt64) \ |
652 | 63 | M(DateV2, ColumnUInt32) \ |
653 | 50 | M(DateTimeV2, ColumnUInt64) |
654 | | |
655 | | #define IP_TYPE_TO_COLUMN_TYPE(M) \ |
656 | 44 | M(IPv4, ColumnIPv4) \ |
657 | 44 | M(IPv6, ColumnIPv6) |
658 | | |
659 | | #define COMPLEX_TYPE_TO_COLUMN_TYPE(M) \ |
660 | 44 | M(Array, ColumnArray) \ |
661 | 44 | M(Map, ColumnMap) \ |
662 | 8 | M(Struct, ColumnStruct) \ |
663 | 8 | M(VARIANT, ColumnObject) \ |
664 | 8 | M(BitMap, ColumnBitmap) \ |
665 | 8 | M(HLL, ColumnHLL) \ |
666 | 4 | M(QuantileState, ColumnQuantileState) |
667 | | |
668 | | #define TYPE_TO_BASIC_COLUMN_TYPE(M) \ |
669 | 105 | NUMERIC_TYPE_TO_COLUMN_TYPE(M) \ |
670 | 78 | DECIMAL_TYPE_TO_COLUMN_TYPE(M) \ |
671 | 74 | STRING_TYPE_TO_COLUMN_TYPE(M) \ |
672 | 68 | TIME_TYPE_TO_COLUMN_TYPE(M) \ |
673 | 105 | IP_TYPE_TO_COLUMN_TYPE(M) |
674 | | |
675 | | #define TYPE_TO_COLUMN_TYPE(M) \ |
676 | 105 | TYPE_TO_BASIC_COLUMN_TYPE(M) \ |
677 | 105 | COMPLEX_TYPE_TO_COLUMN_TYPE(M) |
678 | | |
679 | | } // namespace doris::vectorized |