1 | #include <IO/ReadHelpers.h> |
2 | #include <Interpreters/evaluateConstantExpression.h> |
3 | #include <Interpreters/Context.h> |
4 | #include <Interpreters/convertFieldToType.h> |
5 | #include <Parsers/TokenIterator.h> |
6 | #include <Parsers/ExpressionListParsers.h> |
7 | #include <Processors/Formats/Impl/ValuesBlockInputFormat.h> |
8 | #include <Formats/FormatFactory.h> |
9 | #include <Common/FieldVisitors.h> |
10 | #include <Core/Block.h> |
11 | #include <Common/typeid_cast.h> |
12 | #include <common/find_symbols.h> |
13 | #include <Parsers/ASTLiteral.h> |
14 | #include <DataTypes/DataTypeNullable.h> |
15 | |
16 | |
17 | namespace DB |
18 | { |
19 | |
20 | namespace ErrorCodes |
21 | { |
22 | extern const int CANNOT_PARSE_INPUT_ASSERTION_FAILED; |
23 | extern const int CANNOT_PARSE_QUOTED_STRING; |
24 | extern const int CANNOT_PARSE_NUMBER; |
25 | extern const int CANNOT_PARSE_DATE; |
26 | extern const int CANNOT_PARSE_DATETIME; |
27 | extern const int CANNOT_READ_ARRAY_FROM_TEXT; |
28 | extern const int CANNOT_PARSE_DATE; |
29 | extern const int SYNTAX_ERROR; |
30 | extern const int TYPE_MISMATCH; |
31 | extern const int SUPPORT_IS_DISABLED; |
32 | } |
33 | |
34 | |
35 | ValuesBlockInputFormat::ValuesBlockInputFormat(ReadBuffer & in_, const Block & , const RowInputFormatParams & params_, |
36 | const FormatSettings & format_settings_) |
37 | : IInputFormat(header_, buf), buf(in_), params(params_), |
38 | format_settings(format_settings_), num_columns(header_.columns()), |
39 | parser_type_for_column(num_columns, ParserType::Streaming), |
40 | attempts_to_deduce_template(num_columns), attempts_to_deduce_template_cached(num_columns), |
41 | rows_parsed_using_template(num_columns), templates(num_columns), types(header_.getDataTypes()) |
42 | { |
43 | /// In this format, BOM at beginning of stream cannot be confused with value, so it is safe to skip it. |
44 | skipBOMIfExists(buf); |
45 | } |
46 | |
47 | Chunk ValuesBlockInputFormat::generate() |
48 | { |
49 | const Block & = getPort().getHeader(); |
50 | MutableColumns columns = header.cloneEmptyColumns(); |
51 | block_missing_values.clear(); |
52 | |
53 | for (size_t rows_in_block = 0; rows_in_block < params.max_block_size; ++rows_in_block) |
54 | { |
55 | try |
56 | { |
57 | skipWhitespaceIfAny(buf); |
58 | if (buf.eof() || *buf.position() == ';') |
59 | break; |
60 | readRow(columns, rows_in_block); |
61 | if (params.callback) |
62 | params.callback(); |
63 | } |
64 | catch (Exception & e) |
65 | { |
66 | if (isParseError(e.code())) |
67 | e.addMessage(" at row " + std::to_string(total_rows)); |
68 | throw; |
69 | } |
70 | } |
71 | |
72 | /// Evaluate expressions, which were parsed using templates, if any |
73 | for (size_t i = 0; i < columns.size(); ++i) |
74 | { |
75 | if (!templates[i] || !templates[i]->rowsCount()) |
76 | continue; |
77 | if (columns[i]->empty()) |
78 | columns[i] = std::move(*templates[i]->evaluateAll(block_missing_values, i)).mutate(); |
79 | else |
80 | { |
81 | ColumnPtr evaluated = templates[i]->evaluateAll(block_missing_values, i, columns[i]->size()); |
82 | columns[i]->insertRangeFrom(*evaluated, 0, evaluated->size()); |
83 | } |
84 | } |
85 | |
86 | if (columns.empty() || columns[0]->empty()) |
87 | { |
88 | readSuffix(); |
89 | return {}; |
90 | } |
91 | |
92 | size_t rows_in_block = columns[0]->size(); |
93 | return Chunk{std::move(columns), rows_in_block}; |
94 | } |
95 | |
96 | void ValuesBlockInputFormat::readRow(MutableColumns & columns, size_t row_num) |
97 | { |
98 | assertChar('(', buf); |
99 | |
100 | for (size_t column_idx = 0; column_idx < num_columns; ++column_idx) |
101 | { |
102 | skipWhitespaceIfAny(buf); |
103 | PeekableReadBufferCheckpoint checkpoint{buf}; |
104 | bool read; |
105 | |
106 | /// Parse value using fast streaming parser for literals and slow SQL parser for expressions. |
107 | /// If there is SQL expression in some row, template of this expression will be deduced, |
108 | /// so it makes possible to parse the following rows much faster |
109 | /// if expressions in the following rows have the same structure |
110 | if (parser_type_for_column[column_idx] == ParserType::Streaming) |
111 | read = tryReadValue(*columns[column_idx], column_idx); |
112 | else if (parser_type_for_column[column_idx] == ParserType::BatchTemplate) |
113 | read = tryParseExpressionUsingTemplate(columns[column_idx], column_idx); |
114 | else /// if (parser_type_for_column[column_idx] == ParserType::SingleExpressionEvaluation) |
115 | read = parseExpression(*columns[column_idx], column_idx); |
116 | |
117 | if (!read) |
118 | block_missing_values.setBit(column_idx, row_num); |
119 | /// If read is true, value still may be missing. Bit mask for these values will be copied from ConstantExpressionTemplate later. |
120 | } |
121 | |
122 | skipWhitespaceIfAny(buf); |
123 | if (!buf.eof() && *buf.position() == ',') |
124 | ++buf.position(); |
125 | |
126 | ++total_rows; |
127 | } |
128 | |
129 | bool ValuesBlockInputFormat::tryParseExpressionUsingTemplate(MutableColumnPtr & column, size_t column_idx) |
130 | { |
131 | /// Try to parse expression using template if one was successfully deduced while parsing the first row |
132 | if (templates[column_idx]->parseExpression(buf, format_settings)) |
133 | { |
134 | ++rows_parsed_using_template[column_idx]; |
135 | return true; |
136 | } |
137 | |
138 | /// Expression in the current row is not match template deduced on the first row. |
139 | /// Evaluate expressions, which were parsed using this template. |
140 | if (column->empty()) |
141 | column = std::move(*templates[column_idx]->evaluateAll(block_missing_values, column_idx)).mutate(); |
142 | else |
143 | { |
144 | ColumnPtr evaluated = templates[column_idx]->evaluateAll(block_missing_values, column_idx, column->size()); |
145 | column->insertRangeFrom(*evaluated, 0, evaluated->size()); |
146 | } |
147 | /// Do not use this template anymore |
148 | templates[column_idx].reset(); |
149 | buf.rollbackToCheckpoint(); |
150 | |
151 | /// It will deduce new template or fallback to slow SQL parser |
152 | return parseExpression(*column, column_idx); |
153 | } |
154 | |
155 | bool ValuesBlockInputFormat::tryReadValue(IColumn & column, size_t column_idx) |
156 | { |
157 | bool rollback_on_exception = false; |
158 | try |
159 | { |
160 | bool read = true; |
161 | const auto & type = types[column_idx]; |
162 | if (format_settings.null_as_default && !type->isNullable()) |
163 | read = DataTypeNullable::deserializeTextQuoted(column, buf, format_settings, type); |
164 | else |
165 | type->deserializeAsTextQuoted(column, buf, format_settings); |
166 | rollback_on_exception = true; |
167 | |
168 | skipWhitespaceIfAny(buf); |
169 | assertDelimiterAfterValue(column_idx); |
170 | return read; |
171 | } |
172 | catch (const Exception & e) |
173 | { |
174 | if (!isParseError(e.code()) && e.code() != ErrorCodes::CANNOT_PARSE_INPUT_ASSERTION_FAILED) |
175 | throw; |
176 | if (rollback_on_exception) |
177 | column.popBack(1); |
178 | |
179 | /// Switch to SQL parser and don't try to use streaming parser for complex expressions |
180 | /// Note: Throwing exceptions for each expression may be very slow because of stacktraces |
181 | buf.rollbackToCheckpoint(); |
182 | return parseExpression(column, column_idx); |
183 | } |
184 | } |
185 | |
186 | bool ValuesBlockInputFormat::parseExpression(IColumn & column, size_t column_idx) |
187 | { |
188 | const Block & = getPort().getHeader(); |
189 | const IDataType & type = *header.getByPosition(column_idx).type; |
190 | |
191 | /// We need continuous memory containing the expression to use Lexer |
192 | skipToNextRow(0, 1); |
193 | buf.makeContinuousMemoryFromCheckpointToPos(); |
194 | buf.rollbackToCheckpoint(); |
195 | |
196 | Expected expected; |
197 | Tokens tokens(buf.position(), buf.buffer().end()); |
198 | IParser::Pos token_iterator(tokens); |
199 | ASTPtr ast; |
200 | |
201 | bool parsed = parser.parse(token_iterator, ast, expected); |
202 | |
203 | /// Consider delimiter after value (',' or ')') as part of expression |
204 | if (column_idx + 1 != num_columns) |
205 | parsed &= token_iterator->type == TokenType::Comma; |
206 | else |
207 | parsed &= token_iterator->type == TokenType::ClosingRoundBracket; |
208 | |
209 | if (!parsed) |
210 | throw Exception("Cannot parse expression of type " + type.getName() + " here: " |
211 | + String(buf.position(), std::min(SHOW_CHARS_ON_SYNTAX_ERROR, buf.buffer().end() - buf.position())), |
212 | ErrorCodes::SYNTAX_ERROR); |
213 | ++token_iterator; |
214 | |
215 | if (parser_type_for_column[column_idx] != ParserType::Streaming && dynamic_cast<const ASTLiteral *>(ast.get())) |
216 | { |
217 | /// It's possible that streaming parsing has failed on some row (e.g. because of '+' sign before integer), |
218 | /// but it still can parse the following rows |
219 | /// Check if we can use fast streaming parser instead if using templates |
220 | bool rollback_on_exception = false; |
221 | bool ok = false; |
222 | try |
223 | { |
224 | header.getByPosition(column_idx).type->deserializeAsTextQuoted(column, buf, format_settings); |
225 | rollback_on_exception = true; |
226 | skipWhitespaceIfAny(buf); |
227 | if (checkDelimiterAfterValue(column_idx)) |
228 | ok = true; |
229 | } |
230 | catch (const Exception & e) |
231 | { |
232 | if (!isParseError(e.code())) |
233 | throw; |
234 | } |
235 | if (ok) |
236 | { |
237 | parser_type_for_column[column_idx] = ParserType::Streaming; |
238 | return true; |
239 | } |
240 | else if (rollback_on_exception) |
241 | column.popBack(1); |
242 | } |
243 | |
244 | parser_type_for_column[column_idx] = ParserType::SingleExpressionEvaluation; |
245 | |
246 | /// Try to deduce template of expression and use it to parse the following rows |
247 | if (shouldDeduceNewTemplate(column_idx)) |
248 | { |
249 | if (templates[column_idx]) |
250 | throw DB::Exception("Template for column " + std::to_string(column_idx) + " already exists and it was not evaluated yet" , |
251 | ErrorCodes::LOGICAL_ERROR); |
252 | std::exception_ptr exception; |
253 | try |
254 | { |
255 | bool found_in_cache = false; |
256 | const auto & result_type = header.getByPosition(column_idx).type; |
257 | const char * delimiter = (column_idx + 1 == num_columns) ? ")" : "," ; |
258 | auto structure = templates_cache.getFromCacheOrConstruct(result_type, format_settings.null_as_default, |
259 | TokenIterator(tokens), token_iterator, |
260 | ast, *context, &found_in_cache, delimiter); |
261 | templates[column_idx].emplace(structure); |
262 | if (found_in_cache) |
263 | ++attempts_to_deduce_template_cached[column_idx]; |
264 | else |
265 | ++attempts_to_deduce_template[column_idx]; |
266 | |
267 | buf.rollbackToCheckpoint(); |
268 | if (templates[column_idx]->parseExpression(buf, format_settings)) |
269 | { |
270 | ++rows_parsed_using_template[column_idx]; |
271 | parser_type_for_column[column_idx] = ParserType::BatchTemplate; |
272 | return true; |
273 | } |
274 | } |
275 | catch (...) |
276 | { |
277 | exception = std::current_exception(); |
278 | } |
279 | if (!format_settings.values.interpret_expressions) |
280 | { |
281 | if (exception) |
282 | std::rethrow_exception(exception); |
283 | else |
284 | { |
285 | buf.rollbackToCheckpoint(); |
286 | size_t len = const_cast<char *>(token_iterator->begin) - buf.position(); |
287 | throw Exception("Cannot deduce template of expression: " + std::string(buf.position(), len), ErrorCodes::SYNTAX_ERROR); |
288 | } |
289 | } |
290 | /// Continue parsing without template |
291 | templates[column_idx].reset(); |
292 | } |
293 | |
294 | if (!format_settings.values.interpret_expressions) |
295 | throw Exception("Interpreting expressions is disabled" , ErrorCodes::SUPPORT_IS_DISABLED); |
296 | |
297 | /// Try to evaluate single expression if other parsers don't work |
298 | buf.position() = const_cast<char *>(token_iterator->begin); |
299 | |
300 | std::pair<Field, DataTypePtr> value_raw = evaluateConstantExpression(ast, *context); |
301 | Field value = convertFieldToType(value_raw.first, type, value_raw.second.get()); |
302 | |
303 | /// Check that we are indeed allowed to insert a NULL. |
304 | if (value.isNull() && !type.isNullable()) |
305 | { |
306 | if (format_settings.null_as_default) |
307 | { |
308 | type.insertDefaultInto(column); |
309 | return false; |
310 | } |
311 | buf.rollbackToCheckpoint(); |
312 | throw Exception{"Cannot insert NULL value into a column of type '" + type.getName() + "'" |
313 | + " at: " + |
314 | String(buf.position(), std::min(SHOW_CHARS_ON_SYNTAX_ERROR, buf.buffer().end() - buf.position())), |
315 | ErrorCodes::TYPE_MISMATCH}; |
316 | } |
317 | |
318 | column.insert(value); |
319 | return true; |
320 | } |
321 | |
322 | /// Can be used in fileSegmentationEngine for parallel parsing of Values |
323 | bool ValuesBlockInputFormat::skipToNextRow(size_t min_chunk_bytes, int balance) |
324 | { |
325 | skipWhitespaceIfAny(buf); |
326 | if (buf.eof() || *buf.position() == ';') |
327 | return false; |
328 | bool quoted = false; |
329 | |
330 | size_t chunk_begin_buf_count = buf.count(); |
331 | while (!buf.eof() && (balance || buf.count() - chunk_begin_buf_count < min_chunk_bytes)) |
332 | { |
333 | buf.position() = find_first_symbols<'\\', '\'', ')', '('>(buf.position(), buf.buffer().end()); |
334 | if (buf.position() == buf.buffer().end()) |
335 | continue; |
336 | if (*buf.position() == '\\') |
337 | { |
338 | ++buf.position(); |
339 | if (!buf.eof()) |
340 | ++buf.position(); |
341 | } |
342 | else if (*buf.position() == '\'') |
343 | { |
344 | quoted ^= true; |
345 | ++buf.position(); |
346 | } |
347 | else if (*buf.position() == ')') |
348 | { |
349 | ++buf.position(); |
350 | if (!quoted) |
351 | --balance; |
352 | } |
353 | else if (*buf.position() == '(') |
354 | { |
355 | ++buf.position(); |
356 | if (!quoted) |
357 | ++balance; |
358 | } |
359 | } |
360 | |
361 | if (!buf.eof() && *buf.position() == ',') |
362 | ++buf.position(); |
363 | return true; |
364 | } |
365 | |
366 | void ValuesBlockInputFormat::assertDelimiterAfterValue(size_t column_idx) |
367 | { |
368 | if (unlikely(!checkDelimiterAfterValue(column_idx))) |
369 | throwAtAssertionFailed((column_idx + 1 == num_columns) ? ")" : "," , buf); |
370 | } |
371 | |
372 | bool ValuesBlockInputFormat::checkDelimiterAfterValue(size_t column_idx) |
373 | { |
374 | skipWhitespaceIfAny(buf); |
375 | |
376 | if (likely(column_idx + 1 != num_columns)) |
377 | return checkChar(',', buf); |
378 | else |
379 | return checkChar(')', buf); |
380 | } |
381 | |
382 | bool ValuesBlockInputFormat::shouldDeduceNewTemplate(size_t column_idx) |
383 | { |
384 | if (!format_settings.values.deduce_templates_of_expressions) |
385 | return false; |
386 | |
387 | /// TODO better heuristic |
388 | |
389 | /// Using template from cache is approx 2x faster, than evaluating single expression |
390 | /// Construction of new template is approx 1.5x slower, than evaluating single expression |
391 | float attempts_weighted = 1.5 * attempts_to_deduce_template[column_idx] + 0.5 * attempts_to_deduce_template_cached[column_idx]; |
392 | |
393 | constexpr size_t max_attempts = 100; |
394 | if (attempts_weighted < max_attempts) |
395 | return true; |
396 | |
397 | if (rows_parsed_using_template[column_idx] / attempts_weighted > 1) |
398 | { |
399 | /// Try again |
400 | attempts_to_deduce_template[column_idx] = 0; |
401 | attempts_to_deduce_template_cached[column_idx] = 0; |
402 | rows_parsed_using_template[column_idx] = 0; |
403 | return true; |
404 | } |
405 | return false; |
406 | } |
407 | |
408 | void ValuesBlockInputFormat::readSuffix() |
409 | { |
410 | if (buf.hasUnreadData()) |
411 | throw Exception("Unread data in PeekableReadBuffer will be lost. Most likely it's a bug." , ErrorCodes::LOGICAL_ERROR); |
412 | } |
413 | |
414 | void ValuesBlockInputFormat::resetParser() |
415 | { |
416 | IInputFormat::resetParser(); |
417 | // I'm not resetting parser modes here. |
418 | // There is a good chance that all messages have the same format. |
419 | total_rows = 0; |
420 | } |
421 | |
422 | void registerInputFormatProcessorValues(FormatFactory & factory) |
423 | { |
424 | factory.registerInputFormatProcessor("Values" , []( |
425 | ReadBuffer & buf, |
426 | const Block & , |
427 | const RowInputFormatParams & params, |
428 | const FormatSettings & settings) |
429 | { |
430 | return std::make_shared<ValuesBlockInputFormat>(buf, header, params, settings); |
431 | }); |
432 | } |
433 | |
434 | } |
435 | |