OLD | NEW |
(Empty) | |
| 1 // Copyright (c) 2014, the Dart project authors. Please see the AUTHORS file |
| 2 // for details. All rights reserved. Use of this source code is governed by a |
| 3 // BSD-style license that can be found in the LICENSE file. |
| 4 |
| 5 #include "vm/regexp.h" |
| 6 |
| 7 #include "vm/dart_entry.h" |
| 8 #include "vm/regexp_assembler.h" |
| 9 #include "vm/regexp_ast.h" |
| 10 #include "vm/unibrow-inl.h" |
| 11 #include "vm/unicode.h" |
| 12 #include "vm/symbols.h" |
| 13 |
| 14 #define I isolate() |
| 15 #define CI compiler->isolate() |
| 16 |
| 17 namespace dart { |
| 18 |
| 19 DECLARE_FLAG(bool, trace_irregexp); |
| 20 |
| 21 #define DEFINE_ACCEPT(Type) \ |
| 22 void Type##Node::Accept(NodeVisitor* visitor) { \ |
| 23 visitor->Visit##Type(this); \ |
| 24 } |
| 25 FOR_EACH_NODE_TYPE(DEFINE_ACCEPT) |
| 26 #undef DEFINE_ACCEPT |
| 27 |
| 28 |
| 29 // Default to generating optimized regexp code. |
| 30 static const bool kRegexpOptimization = true; |
| 31 |
| 32 // More makes code generation slower, less makes V8 benchmark score lower. |
| 33 static const intptr_t kMaxLookaheadForBoyerMoore = 8; |
| 34 |
| 35 // In a 3-character pattern you can maximally step forwards 3 characters |
| 36 // at a time, which is not always enough to pay for the extra logic. |
| 37 static const intptr_t kPatternTooShortForBoyerMoore = 2; |
| 38 |
| 39 // The '2' variant has inclusive from and exclusive to. |
| 40 // This covers \s as defined in ECMA-262 5.1, 15.10.2.12, |
| 41 // which include WhiteSpace (7.2) or LineTerminator (7.3) values. |
| 42 static const intptr_t kSpaceRanges[] = { '\t', '\r' + 1, ' ', ' ' + 1, |
| 43 0x00A0, 0x00A1, 0x1680, 0x1681, 0x180E, 0x180F, 0x2000, 0x200B, |
| 44 0x2028, 0x202A, 0x202F, 0x2030, 0x205F, 0x2060, 0x3000, 0x3001, |
| 45 0xFEFF, 0xFF00, 0x10000 }; |
| 46 static const intptr_t kSpaceRangeCount = ARRAY_SIZE(kSpaceRanges); |
| 47 static const intptr_t kWordRanges[] = { |
| 48 '0', '9' + 1, 'A', 'Z' + 1, '_', '_' + 1, 'a', 'z' + 1, 0x10000 }; |
| 49 static const intptr_t kWordRangeCount = ARRAY_SIZE(kWordRanges); |
| 50 static const intptr_t kDigitRanges[] = { '0', '9' + 1, 0x10000 }; |
| 51 static const intptr_t kDigitRangeCount = ARRAY_SIZE(kDigitRanges); |
| 52 static const intptr_t kSurrogateRanges[] = { 0xd800, 0xe000, 0x10000 }; |
| 53 static const intptr_t kSurrogateRangeCount = ARRAY_SIZE(kSurrogateRanges); |
| 54 static const intptr_t kLineTerminatorRanges[] = { |
| 55 0x000A, 0x000B, 0x000D, 0x000E, 0x2028, 0x202A, 0x10000 }; |
| 56 static const intptr_t kLineTerminatorRangeCount = |
| 57 ARRAY_SIZE(kLineTerminatorRanges); |
| 58 |
| 59 |
| 60 static inline void PrintUtf16(uint16_t c) { |
| 61 const char* format = (0x20 <= c && c <= 0x7F) ? |
| 62 "%c" : (c <= 0xff) ? "\\x%02x" : "\\u%04x"; |
| 63 OS::Print(format, c); |
| 64 } |
| 65 |
| 66 |
| 67 // We need to check for the following characters: 0x39c 0x3bc 0x178. |
| 68 static inline bool RangeContainsLatin1Equivalents(CharacterRange range) { |
| 69 // TODO(dcarney): this could be a lot more efficient. |
| 70 return range.Contains(0x39c) || |
| 71 range.Contains(0x3bc) || range.Contains(0x178); |
| 72 } |
| 73 |
| 74 |
| 75 static bool RangesContainLatin1Equivalents( |
| 76 ZoneGrowableArray<CharacterRange>* ranges) { |
| 77 for (intptr_t i = 0; i < ranges->length(); i++) { |
| 78 // TODO(dcarney): this could be a lot more efficient. |
| 79 if (RangeContainsLatin1Equivalents(ranges->At(i))) return true; |
| 80 } |
| 81 return false; |
| 82 } |
| 83 |
| 84 static uint16_t ConvertNonLatin1ToLatin1(uint16_t c) { |
| 85 ASSERT(c > Symbols::kMaxOneCharCodeSymbol); |
| 86 switch (c) { |
| 87 // This are equivalent characters in unicode. |
| 88 case 0x39c: |
| 89 case 0x3bc: |
| 90 return 0xb5; |
| 91 // This is an uppercase of a Latin-1 character |
| 92 // outside of Latin-1. |
| 93 case 0x178: |
| 94 return 0xff; |
| 95 } |
| 96 return 0; |
| 97 } |
| 98 |
| 99 |
| 100 class VisitMarker : public ValueObject { |
| 101 public: |
| 102 explicit VisitMarker(NodeInfo* info) : info_(info) { |
| 103 ASSERT(!info->visited); |
| 104 info->visited = true; |
| 105 } |
| 106 ~VisitMarker() { |
| 107 info_->visited = false; |
| 108 } |
| 109 private: |
| 110 NodeInfo* info_; |
| 111 }; |
| 112 |
| 113 |
| 114 class FrequencyCollator : public ValueObject { |
| 115 public: |
| 116 FrequencyCollator() : total_samples_(0) { |
| 117 for (intptr_t i = 0; i < RegExpMacroAssembler::kTableSize; i++) { |
| 118 frequencies_[i] = CharacterFrequency(i); |
| 119 } |
| 120 } |
| 121 |
| 122 void CountCharacter(intptr_t character) { |
| 123 intptr_t index = (character & RegExpMacroAssembler::kTableMask); |
| 124 frequencies_[index].Increment(); |
| 125 total_samples_++; |
| 126 } |
| 127 |
| 128 // Does not measure in percent, but rather per-128 (the table size from the |
| 129 // regexp macro assembler). |
| 130 intptr_t Frequency(intptr_t in_character) { |
| 131 ASSERT((in_character & RegExpMacroAssembler::kTableMask) == in_character); |
| 132 if (total_samples_ < 1) return 1; // Division by zero. |
| 133 intptr_t freq_in_per128 = |
| 134 (frequencies_[in_character].counter() * 128) / total_samples_; |
| 135 return freq_in_per128; |
| 136 } |
| 137 |
| 138 private: |
| 139 class CharacterFrequency { |
| 140 public: |
| 141 CharacterFrequency() : counter_(0), character_(-1) { } |
| 142 explicit CharacterFrequency(intptr_t character) |
| 143 : counter_(0), character_(character) { } |
| 144 |
| 145 void Increment() { counter_++; } |
| 146 intptr_t counter() { return counter_; } |
| 147 intptr_t character() { return character_; } |
| 148 |
| 149 private: |
| 150 intptr_t counter_; |
| 151 intptr_t character_; |
| 152 |
| 153 DISALLOW_ALLOCATION(); |
| 154 }; |
| 155 |
| 156 |
| 157 private: |
| 158 CharacterFrequency frequencies_[RegExpMacroAssembler::kTableSize]; |
| 159 intptr_t total_samples_; |
| 160 }; |
| 161 |
| 162 |
| 163 class RegExpCompiler : public ValueObject { |
| 164 public: |
| 165 RegExpCompiler(intptr_t capture_count, |
| 166 bool ignore_case, |
| 167 intptr_t specialization_cid); |
| 168 |
| 169 intptr_t AllocateRegister() { |
| 170 return next_register_++; |
| 171 } |
| 172 |
| 173 RegExpEngine::CompilationResult Assemble(IRRegExpMacroAssembler* assembler, |
| 174 RegExpNode* start, |
| 175 intptr_t capture_count, |
| 176 const String& pattern); |
| 177 |
| 178 inline void AddWork(RegExpNode* node) { work_list_->Add(node); } |
| 179 |
| 180 static const intptr_t kImplementationOffset = 0; |
| 181 static const intptr_t kNumberOfRegistersOffset = 0; |
| 182 static const intptr_t kCodeOffset = 1; |
| 183 |
| 184 IRRegExpMacroAssembler* macro_assembler() { return macro_assembler_; } |
| 185 EndNode* accept() { return accept_; } |
| 186 |
| 187 static const intptr_t kMaxRecursion = 100; |
| 188 inline intptr_t recursion_depth() { return recursion_depth_; } |
| 189 inline void IncrementRecursionDepth() { recursion_depth_++; } |
| 190 inline void DecrementRecursionDepth() { recursion_depth_--; } |
| 191 |
| 192 void SetRegExpTooBig() { reg_exp_too_big_ = true; } |
| 193 |
| 194 inline bool ignore_case() { return ignore_case_; } |
| 195 inline bool ascii() const { |
| 196 return (specialization_cid_ == kOneByteStringCid || |
| 197 specialization_cid_ == kExternalOneByteStringCid); |
| 198 } |
| 199 inline intptr_t specialization_cid() { return specialization_cid_; } |
| 200 FrequencyCollator* frequency_collator() { return &frequency_collator_; } |
| 201 |
| 202 intptr_t current_expansion_factor() { return current_expansion_factor_; } |
| 203 void set_current_expansion_factor(intptr_t value) { |
| 204 current_expansion_factor_ = value; |
| 205 } |
| 206 |
| 207 Isolate* isolate() const { return isolate_; } |
| 208 |
| 209 static const intptr_t kNoRegister = -1; |
| 210 |
| 211 private: |
| 212 EndNode* accept_; |
| 213 intptr_t next_register_; |
| 214 ZoneGrowableArray<RegExpNode*>* work_list_; |
| 215 intptr_t recursion_depth_; |
| 216 IRRegExpMacroAssembler* macro_assembler_; |
| 217 bool ignore_case_; |
| 218 intptr_t specialization_cid_; |
| 219 bool reg_exp_too_big_; |
| 220 intptr_t current_expansion_factor_; |
| 221 FrequencyCollator frequency_collator_; |
| 222 Isolate* isolate_; |
| 223 }; |
| 224 |
| 225 |
| 226 class RecursionCheck : public ValueObject { |
| 227 public: |
| 228 explicit RecursionCheck(RegExpCompiler* compiler) : compiler_(compiler) { |
| 229 compiler->IncrementRecursionDepth(); |
| 230 } |
| 231 ~RecursionCheck() { compiler_->DecrementRecursionDepth(); } |
| 232 private: |
| 233 RegExpCompiler* compiler_; |
| 234 }; |
| 235 |
| 236 |
| 237 // Scoped object to keep track of how much we unroll quantifier loops in the |
| 238 // regexp graph generator. |
| 239 class RegExpExpansionLimiter : public ValueObject { |
| 240 public: |
| 241 static const intptr_t kMaxExpansionFactor = 6; |
| 242 RegExpExpansionLimiter(RegExpCompiler* compiler, intptr_t factor) |
| 243 : compiler_(compiler), |
| 244 saved_expansion_factor_(compiler->current_expansion_factor()), |
| 245 ok_to_expand_(saved_expansion_factor_ <= kMaxExpansionFactor) { |
| 246 ASSERT(factor > 0); |
| 247 if (ok_to_expand_) { |
| 248 if (factor > kMaxExpansionFactor) { |
| 249 // Avoid integer overflow of the current expansion factor. |
| 250 ok_to_expand_ = false; |
| 251 compiler->set_current_expansion_factor(kMaxExpansionFactor + 1); |
| 252 } else { |
| 253 intptr_t new_factor = saved_expansion_factor_ * factor; |
| 254 ok_to_expand_ = (new_factor <= kMaxExpansionFactor); |
| 255 compiler->set_current_expansion_factor(new_factor); |
| 256 } |
| 257 } |
| 258 } |
| 259 |
| 260 ~RegExpExpansionLimiter() { |
| 261 compiler_->set_current_expansion_factor(saved_expansion_factor_); |
| 262 } |
| 263 |
| 264 bool ok_to_expand() { return ok_to_expand_; } |
| 265 |
| 266 private: |
| 267 RegExpCompiler* compiler_; |
| 268 intptr_t saved_expansion_factor_; |
| 269 bool ok_to_expand_; |
| 270 |
| 271 DISALLOW_IMPLICIT_CONSTRUCTORS(RegExpExpansionLimiter); |
| 272 }; |
| 273 |
| 274 |
| 275 // Node generation ------------------------------------------------------------- |
| 276 |
| 277 RegExpNode* RegExpAtom::ToNode(RegExpCompiler* compiler, |
| 278 RegExpNode* on_success) { |
| 279 ZoneGrowableArray<TextElement>* elms = |
| 280 new(CI) ZoneGrowableArray<TextElement>(1); |
| 281 elms->Add(TextElement::Atom(this)); |
| 282 return new(CI) TextNode(elms, on_success); |
| 283 } |
| 284 |
| 285 |
| 286 RegExpNode* RegExpText::ToNode(RegExpCompiler* compiler, |
| 287 RegExpNode* on_success) { |
| 288 ZoneGrowableArray<TextElement>* elms = |
| 289 new(CI) ZoneGrowableArray<TextElement>(1); |
| 290 for (intptr_t i = 0; i < elements()->length(); i++) { |
| 291 elms->Add(elements()->At(i)); |
| 292 } |
| 293 return new(CI) TextNode(elms, on_success); |
| 294 } |
| 295 |
| 296 |
| 297 RegExpNode* RegExpCharacterClass::ToNode(RegExpCompiler* compiler, |
| 298 RegExpNode* on_success) { |
| 299 return new(CI) TextNode(this, on_success); |
| 300 } |
| 301 |
| 302 |
| 303 RegExpNode* RegExpDisjunction::ToNode(RegExpCompiler* compiler, |
| 304 RegExpNode* on_success) { |
| 305 ZoneGrowableArray<RegExpTree*>* alternatives = this->alternatives(); |
| 306 intptr_t length = alternatives->length(); |
| 307 ChoiceNode* result = |
| 308 new(CI) ChoiceNode(length, CI); |
| 309 for (intptr_t i = 0; i < length; i++) { |
| 310 GuardedAlternative alternative(alternatives->At(i)->ToNode(compiler, |
| 311 on_success)); |
| 312 result->AddAlternative(alternative); |
| 313 } |
| 314 return result; |
| 315 } |
| 316 |
| 317 |
| 318 RegExpNode* RegExpQuantifier::ToNode(RegExpCompiler* compiler, |
| 319 RegExpNode* on_success) { |
| 320 return ToNode(min(), |
| 321 max(), |
| 322 is_greedy(), |
| 323 body(), |
| 324 compiler, |
| 325 on_success); |
| 326 } |
| 327 |
| 328 |
| 329 RegExpNode* RegExpQuantifier::ToNode(intptr_t min, |
| 330 intptr_t max, |
| 331 bool is_greedy, |
| 332 RegExpTree* body, |
| 333 RegExpCompiler* compiler, |
| 334 RegExpNode* on_success, |
| 335 bool not_at_start) { |
| 336 // x{f, t} becomes this: |
| 337 // |
| 338 // (r++)<-. |
| 339 // | ` |
| 340 // | (x) |
| 341 // v ^ |
| 342 // (r=0)-->(?)---/ [if r < t] |
| 343 // | |
| 344 // [if r >= f] \----> ... |
| 345 // |
| 346 |
| 347 // 15.10.2.5 RepeatMatcher algorithm. |
| 348 // The parser has already eliminated the case where max is 0. In the case |
| 349 // where max_match is zero the parser has removed the quantifier if min was |
| 350 // > 0 and removed the atom if min was 0. See AddQuantifierToAtom. |
| 351 |
| 352 // If we know that we cannot match zero length then things are a little |
| 353 // simpler since we don't need to make the special zero length match check |
| 354 // from step 2.1. If the min and max are small we can unroll a little in |
| 355 // this case. |
| 356 // Unroll (foo)+ and (foo){3,} |
| 357 static const intptr_t kMaxUnrolledMinMatches = 3; |
| 358 // Unroll (foo)? and (foo){x,3} |
| 359 static const intptr_t kMaxUnrolledMaxMatches = 3; |
| 360 if (max == 0) return on_success; // This can happen due to recursion. |
| 361 bool body_can_be_empty = (body->min_match() == 0); |
| 362 intptr_t body_start_reg = RegExpCompiler::kNoRegister; |
| 363 Interval capture_registers = body->CaptureRegisters(); |
| 364 bool needs_capture_clearing = !capture_registers.is_empty(); |
| 365 Isolate* isolate = compiler->isolate(); |
| 366 |
| 367 if (body_can_be_empty) { |
| 368 body_start_reg = compiler->AllocateRegister(); |
| 369 } else if (kRegexpOptimization && !needs_capture_clearing) { |
| 370 // Only unroll if there are no captures and the body can't be |
| 371 // empty. |
| 372 { |
| 373 RegExpExpansionLimiter limiter( |
| 374 compiler, min + ((max != min) ? 1 : 0)); |
| 375 if (min > 0 && min <= kMaxUnrolledMinMatches && limiter.ok_to_expand()) { |
| 376 intptr_t new_max = (max == kInfinity) ? max : max - min; |
| 377 // Recurse once to get the loop or optional matches after the fixed |
| 378 // ones. |
| 379 RegExpNode* answer = ToNode( |
| 380 0, new_max, is_greedy, body, compiler, on_success, true); |
| 381 // Unroll the forced matches from 0 to min. This can cause chains of |
| 382 // TextNodes (which the parser does not generate). These should be |
| 383 // combined if it turns out they hinder good code generation. |
| 384 for (intptr_t i = 0; i < min; i++) { |
| 385 answer = body->ToNode(compiler, answer); |
| 386 } |
| 387 return answer; |
| 388 } |
| 389 } |
| 390 if (max <= kMaxUnrolledMaxMatches && min == 0) { |
| 391 ASSERT(max > 0); // Due to the 'if' above. |
| 392 RegExpExpansionLimiter limiter(compiler, max); |
| 393 if (limiter.ok_to_expand()) { |
| 394 // Unroll the optional matches up to max. |
| 395 RegExpNode* answer = on_success; |
| 396 for (intptr_t i = 0; i < max; i++) { |
| 397 ChoiceNode* alternation = new(isolate) ChoiceNode(2, isolate); |
| 398 if (is_greedy) { |
| 399 alternation->AddAlternative( |
| 400 GuardedAlternative(body->ToNode(compiler, answer))); |
| 401 alternation->AddAlternative(GuardedAlternative(on_success)); |
| 402 } else { |
| 403 alternation->AddAlternative(GuardedAlternative(on_success)); |
| 404 alternation->AddAlternative( |
| 405 GuardedAlternative(body->ToNode(compiler, answer))); |
| 406 } |
| 407 answer = alternation; |
| 408 if (not_at_start) alternation->set_not_at_start(); |
| 409 } |
| 410 return answer; |
| 411 } |
| 412 } |
| 413 } |
| 414 bool has_min = min > 0; |
| 415 bool has_max = max < RegExpTree::kInfinity; |
| 416 bool needs_counter = has_min || has_max; |
| 417 intptr_t reg_ctr = needs_counter |
| 418 ? compiler->AllocateRegister() |
| 419 : RegExpCompiler::kNoRegister; |
| 420 LoopChoiceNode* center = new(isolate) LoopChoiceNode(body->min_match() == 0, |
| 421 isolate); |
| 422 if (not_at_start) center->set_not_at_start(); |
| 423 RegExpNode* loop_return = needs_counter |
| 424 ? static_cast<RegExpNode*>(ActionNode::IncrementRegister(reg_ctr, center)) |
| 425 : static_cast<RegExpNode*>(center); |
| 426 if (body_can_be_empty) { |
| 427 // If the body can be empty we need to check if it was and then |
| 428 // backtrack. |
| 429 loop_return = ActionNode::EmptyMatchCheck(body_start_reg, |
| 430 reg_ctr, |
| 431 min, |
| 432 loop_return); |
| 433 } |
| 434 RegExpNode* body_node = body->ToNode(compiler, loop_return); |
| 435 if (body_can_be_empty) { |
| 436 // If the body can be empty we need to store the start position |
| 437 // so we can bail out if it was empty. |
| 438 body_node = ActionNode::StorePosition(body_start_reg, false, body_node); |
| 439 } |
| 440 if (needs_capture_clearing) { |
| 441 // Before entering the body of this loop we need to clear captures. |
| 442 body_node = ActionNode::ClearCaptures(capture_registers, body_node); |
| 443 } |
| 444 GuardedAlternative body_alt(body_node); |
| 445 if (has_max) { |
| 446 Guard* body_guard = |
| 447 new(isolate) Guard(reg_ctr, Guard::LT, max); |
| 448 body_alt.AddGuard(body_guard, isolate); |
| 449 } |
| 450 GuardedAlternative rest_alt(on_success); |
| 451 if (has_min) { |
| 452 Guard* rest_guard = new(isolate) Guard(reg_ctr, Guard::GEQ, min); |
| 453 rest_alt.AddGuard(rest_guard, isolate); |
| 454 } |
| 455 if (is_greedy) { |
| 456 center->AddLoopAlternative(body_alt); |
| 457 center->AddContinueAlternative(rest_alt); |
| 458 } else { |
| 459 center->AddContinueAlternative(rest_alt); |
| 460 center->AddLoopAlternative(body_alt); |
| 461 } |
| 462 if (needs_counter) { |
| 463 return ActionNode::SetRegister(reg_ctr, 0, center); |
| 464 } else { |
| 465 return center; |
| 466 } |
| 467 } |
| 468 |
| 469 |
| 470 RegExpNode* RegExpAssertion::ToNode(RegExpCompiler* compiler, |
| 471 RegExpNode* on_success) { |
| 472 switch (assertion_type()) { |
| 473 case START_OF_LINE: |
| 474 return AssertionNode::AfterNewline(on_success); |
| 475 case START_OF_INPUT: |
| 476 return AssertionNode::AtStart(on_success); |
| 477 case BOUNDARY: |
| 478 return AssertionNode::AtBoundary(on_success); |
| 479 case NON_BOUNDARY: |
| 480 return AssertionNode::AtNonBoundary(on_success); |
| 481 case END_OF_INPUT: |
| 482 return AssertionNode::AtEnd(on_success); |
| 483 case END_OF_LINE: { |
| 484 // Compile $ in multiline regexps as an alternation with a positive |
| 485 // lookahead in one side and an end-of-input on the other side. |
| 486 // We need two registers for the lookahead. |
| 487 intptr_t stack_pointer_register = compiler->AllocateRegister(); |
| 488 intptr_t position_register = compiler->AllocateRegister(); |
| 489 // The ChoiceNode to distinguish between a newline and end-of-input. |
| 490 ChoiceNode* result = new ChoiceNode(2, on_success->isolate()); |
| 491 // Create a newline atom. |
| 492 ZoneGrowableArray<CharacterRange>* newline_ranges = |
| 493 new ZoneGrowableArray<CharacterRange>(3); |
| 494 CharacterRange::AddClassEscape('n', newline_ranges); |
| 495 RegExpCharacterClass* newline_atom = new RegExpCharacterClass('n'); |
| 496 TextNode* newline_matcher = new TextNode( |
| 497 newline_atom, |
| 498 ActionNode::PositiveSubmatchSuccess(stack_pointer_register, |
| 499 position_register, |
| 500 0, // No captures inside. |
| 501 -1, // Ignored if no captures. |
| 502 on_success)); |
| 503 // Create an end-of-input matcher. |
| 504 RegExpNode* end_of_line = ActionNode::BeginSubmatch( |
| 505 stack_pointer_register, |
| 506 position_register, |
| 507 newline_matcher); |
| 508 // Add the two alternatives to the ChoiceNode. |
| 509 GuardedAlternative eol_alternative(end_of_line); |
| 510 result->AddAlternative(eol_alternative); |
| 511 GuardedAlternative end_alternative(AssertionNode::AtEnd(on_success)); |
| 512 result->AddAlternative(end_alternative); |
| 513 return result; |
| 514 } |
| 515 default: |
| 516 UNREACHABLE(); |
| 517 } |
| 518 return on_success; |
| 519 } |
| 520 |
| 521 |
| 522 RegExpNode* RegExpBackReference::ToNode(RegExpCompiler* compiler, |
| 523 RegExpNode* on_success) { |
| 524 return new(CI) |
| 525 BackReferenceNode(RegExpCapture::StartRegister(index()), |
| 526 RegExpCapture::EndRegister(index()), |
| 527 on_success); |
| 528 } |
| 529 |
| 530 |
| 531 RegExpNode* RegExpEmpty::ToNode(RegExpCompiler* compiler, |
| 532 RegExpNode* on_success) { |
| 533 return on_success; |
| 534 } |
| 535 |
| 536 |
| 537 RegExpNode* RegExpLookahead::ToNode(RegExpCompiler* compiler, |
| 538 RegExpNode* on_success) { |
| 539 intptr_t stack_pointer_register = compiler->AllocateRegister(); |
| 540 intptr_t position_register = compiler->AllocateRegister(); |
| 541 |
| 542 const intptr_t registers_per_capture = 2; |
| 543 const intptr_t register_of_first_capture = 2; |
| 544 intptr_t register_count = capture_count_ * registers_per_capture; |
| 545 intptr_t register_start = |
| 546 register_of_first_capture + capture_from_ * registers_per_capture; |
| 547 |
| 548 RegExpNode* success; |
| 549 if (is_positive()) { |
| 550 RegExpNode* node = ActionNode::BeginSubmatch( |
| 551 stack_pointer_register, |
| 552 position_register, |
| 553 body()->ToNode( |
| 554 compiler, |
| 555 ActionNode::PositiveSubmatchSuccess(stack_pointer_register, |
| 556 position_register, |
| 557 register_count, |
| 558 register_start, |
| 559 on_success))); |
| 560 return node; |
| 561 } else { |
| 562 // We use a ChoiceNode for a negative lookahead because it has most of |
| 563 // the characteristics we need. It has the body of the lookahead as its |
| 564 // first alternative and the expression after the lookahead of the second |
| 565 // alternative. If the first alternative succeeds then the |
| 566 // NegativeSubmatchSuccess will unwind the stack including everything the |
| 567 // choice node set up and backtrack. If the first alternative fails then |
| 568 // the second alternative is tried, which is exactly the desired result |
| 569 // for a negative lookahead. The NegativeLookaheadChoiceNode is a special |
| 570 // ChoiceNode that knows to ignore the first exit when calculating quick |
| 571 // checks. |
| 572 |
| 573 GuardedAlternative body_alt( |
| 574 body()->ToNode( |
| 575 compiler, |
| 576 success = new(CI) NegativeSubmatchSuccess(stack_pointer_register, |
| 577 position_register, |
| 578 register_count, |
| 579 register_start, |
| 580 CI))); |
| 581 ChoiceNode* choice_node = |
| 582 new(CI) NegativeLookaheadChoiceNode(body_alt, |
| 583 GuardedAlternative(on_success), |
| 584 CI); |
| 585 return ActionNode::BeginSubmatch(stack_pointer_register, |
| 586 position_register, |
| 587 choice_node); |
| 588 } |
| 589 } |
| 590 |
| 591 |
| 592 RegExpNode* RegExpCapture::ToNode(RegExpCompiler* compiler, |
| 593 RegExpNode* on_success) { |
| 594 return ToNode(body(), index(), compiler, on_success); |
| 595 } |
| 596 |
| 597 |
| 598 RegExpNode* RegExpCapture::ToNode(RegExpTree* body, |
| 599 intptr_t index, |
| 600 RegExpCompiler* compiler, |
| 601 RegExpNode* on_success) { |
| 602 intptr_t start_reg = RegExpCapture::StartRegister(index); |
| 603 intptr_t end_reg = RegExpCapture::EndRegister(index); |
| 604 RegExpNode* store_end = ActionNode::StorePosition(end_reg, true, on_success); |
| 605 RegExpNode* body_node = body->ToNode(compiler, store_end); |
| 606 return ActionNode::StorePosition(start_reg, true, body_node); |
| 607 } |
| 608 |
| 609 |
| 610 RegExpNode* RegExpAlternative::ToNode(RegExpCompiler* compiler, |
| 611 RegExpNode* on_success) { |
| 612 ZoneGrowableArray<RegExpTree*>* children = nodes(); |
| 613 RegExpNode* current = on_success; |
| 614 for (intptr_t i = children->length() - 1; i >= 0; i--) { |
| 615 current = children->At(i)->ToNode(compiler, current); |
| 616 } |
| 617 return current; |
| 618 } |
| 619 |
| 620 |
| 621 // ASCII filtering ------------------------------------------------------------- |
| 622 |
| 623 |
| 624 RegExpNode* SeqRegExpNode::FilterSuccessor(intptr_t depth, bool ignore_case) { |
| 625 RegExpNode* next = on_success_->FilterASCII(depth - 1, ignore_case); |
| 626 if (next == NULL) return set_replacement(NULL); |
| 627 on_success_ = next; |
| 628 return set_replacement(this); |
| 629 } |
| 630 |
| 631 |
| 632 RegExpNode* SeqRegExpNode::FilterASCII(intptr_t depth, bool ignore_case) { |
| 633 if (info()->replacement_calculated) return replacement(); |
| 634 if (depth < 0) return this; |
| 635 ASSERT(!info()->visited); |
| 636 VisitMarker marker(info()); |
| 637 return FilterSuccessor(depth - 1, ignore_case); |
| 638 } |
| 639 |
| 640 |
| 641 RegExpNode* TextNode::FilterASCII(intptr_t depth, bool ignore_case) { |
| 642 if (info()->replacement_calculated) return replacement(); |
| 643 if (depth < 0) return this; |
| 644 ASSERT(!info()->visited); |
| 645 VisitMarker marker(info()); |
| 646 intptr_t element_count = elms_->length(); |
| 647 for (intptr_t i = 0; i < element_count; i++) { |
| 648 TextElement elm = elms_->At(i); |
| 649 if (elm.text_type() == TextElement::ATOM) { |
| 650 ZoneGrowableArray<uint16_t>* quarks = elm.atom()->data(); |
| 651 for (intptr_t j = 0; j < quarks->length(); j++) { |
| 652 uint16_t c = quarks->At(j); |
| 653 if (c <= Symbols::kMaxOneCharCodeSymbol) continue; |
| 654 if (!ignore_case) return set_replacement(NULL); |
| 655 // Here, we need to check for characters whose upper and lower cases |
| 656 // are outside the Latin-1 range. |
| 657 uint16_t converted = ConvertNonLatin1ToLatin1(c); |
| 658 // Character is outside Latin-1 completely |
| 659 if (converted == 0) return set_replacement(NULL); |
| 660 // Convert quark to Latin-1 in place. |
| 661 (*quarks)[0] = converted; |
| 662 } |
| 663 } else { |
| 664 ASSERT(elm.text_type() == TextElement::CHAR_CLASS); |
| 665 RegExpCharacterClass* cc = elm.char_class(); |
| 666 ZoneGrowableArray<CharacterRange>* ranges = cc->ranges(); |
| 667 if (!CharacterRange::IsCanonical(ranges)) { |
| 668 CharacterRange::Canonicalize(ranges); |
| 669 } |
| 670 // Now they are in order so we only need to look at the first. |
| 671 intptr_t range_count = ranges->length(); |
| 672 if (cc->is_negated()) { |
| 673 if (range_count != 0 && |
| 674 ranges->At(0).from() == 0 && |
| 675 ranges->At(0).to() >= Symbols::kMaxOneCharCodeSymbol) { |
| 676 // This will be handled in a later filter. |
| 677 if (ignore_case && RangesContainLatin1Equivalents(ranges)) continue; |
| 678 return set_replacement(NULL); |
| 679 } |
| 680 } else { |
| 681 if (range_count == 0 || |
| 682 ranges->At(0).from() > Symbols::kMaxOneCharCodeSymbol) { |
| 683 // This will be handled in a later filter. |
| 684 if (ignore_case && RangesContainLatin1Equivalents(ranges)) continue; |
| 685 return set_replacement(NULL); |
| 686 } |
| 687 } |
| 688 } |
| 689 } |
| 690 return FilterSuccessor(depth - 1, ignore_case); |
| 691 } |
| 692 |
| 693 |
| 694 RegExpNode* LoopChoiceNode::FilterASCII(intptr_t depth, bool ignore_case) { |
| 695 if (info()->replacement_calculated) return replacement(); |
| 696 if (depth < 0) return this; |
| 697 if (info()->visited) return this; |
| 698 { |
| 699 VisitMarker marker(info()); |
| 700 |
| 701 RegExpNode* continue_replacement = |
| 702 continue_node_->FilterASCII(depth - 1, ignore_case); |
| 703 // If we can't continue after the loop then there is no sense in doing the |
| 704 // loop. |
| 705 if (continue_replacement == NULL) return set_replacement(NULL); |
| 706 } |
| 707 |
| 708 return ChoiceNode::FilterASCII(depth - 1, ignore_case); |
| 709 } |
| 710 |
| 711 |
| 712 RegExpNode* ChoiceNode::FilterASCII(intptr_t depth, bool ignore_case) { |
| 713 if (info()->replacement_calculated) return replacement(); |
| 714 if (depth < 0) return this; |
| 715 if (info()->visited) return this; |
| 716 VisitMarker marker(info()); |
| 717 intptr_t choice_count = alternatives_->length(); |
| 718 |
| 719 for (intptr_t i = 0; i < choice_count; i++) { |
| 720 GuardedAlternative alternative = alternatives_->At(i); |
| 721 if (alternative.guards() != NULL && alternative.guards()->length() != 0) { |
| 722 set_replacement(this); |
| 723 return this; |
| 724 } |
| 725 } |
| 726 |
| 727 intptr_t surviving = 0; |
| 728 RegExpNode* survivor = NULL; |
| 729 for (intptr_t i = 0; i < choice_count; i++) { |
| 730 GuardedAlternative alternative = alternatives_->At(i); |
| 731 RegExpNode* replacement = |
| 732 alternative.node()->FilterASCII(depth - 1, ignore_case); |
| 733 ASSERT(replacement != this); // No missing EMPTY_MATCH_CHECK. |
| 734 if (replacement != NULL) { |
| 735 (*alternatives_)[i].set_node(replacement); |
| 736 surviving++; |
| 737 survivor = replacement; |
| 738 } |
| 739 } |
| 740 if (surviving < 2) return set_replacement(survivor); |
| 741 |
| 742 set_replacement(this); |
| 743 if (surviving == choice_count) { |
| 744 return this; |
| 745 } |
| 746 // Only some of the nodes survived the filtering. We need to rebuild the |
| 747 // alternatives list. |
| 748 ZoneGrowableArray<GuardedAlternative>* new_alternatives = |
| 749 new(I) ZoneGrowableArray<GuardedAlternative>(surviving); |
| 750 for (intptr_t i = 0; i < choice_count; i++) { |
| 751 RegExpNode* replacement = |
| 752 (*alternatives_)[i].node()->FilterASCII(depth - 1, ignore_case); |
| 753 if (replacement != NULL) { |
| 754 (*alternatives_)[i].set_node(replacement); |
| 755 new_alternatives->Add((*alternatives_)[i]); |
| 756 } |
| 757 } |
| 758 alternatives_ = new_alternatives; |
| 759 return this; |
| 760 } |
| 761 |
| 762 |
| 763 RegExpNode* NegativeLookaheadChoiceNode::FilterASCII(intptr_t depth, |
| 764 bool ignore_case) { |
| 765 if (info()->replacement_calculated) return replacement(); |
| 766 if (depth < 0) return this; |
| 767 if (info()->visited) return this; |
| 768 VisitMarker marker(info()); |
| 769 // Alternative 0 is the negative lookahead, alternative 1 is what comes |
| 770 // afterwards. |
| 771 RegExpNode* node = (*alternatives_)[1].node(); |
| 772 RegExpNode* replacement = node->FilterASCII(depth - 1, ignore_case); |
| 773 if (replacement == NULL) return set_replacement(NULL); |
| 774 (*alternatives_)[1].set_node(replacement); |
| 775 |
| 776 RegExpNode* neg_node = (*alternatives_)[0].node(); |
| 777 RegExpNode* neg_replacement = neg_node->FilterASCII(depth - 1, ignore_case); |
| 778 // If the negative lookahead is always going to fail then |
| 779 // we don't need to check it. |
| 780 if (neg_replacement == NULL) return set_replacement(replacement); |
| 781 (*alternatives_)[0].set_node(neg_replacement); |
| 782 return set_replacement(this); |
| 783 } |
| 784 |
| 785 |
| 786 // Code emission --------------------------------------------------------------- |
| 787 |
| 788 |
| 789 void ChoiceNode::GenerateGuard(RegExpMacroAssembler* macro_assembler, |
| 790 Guard* guard, |
| 791 Trace* trace) { |
| 792 switch (guard->op()) { |
| 793 case Guard::LT: |
| 794 ASSERT(!trace->mentions_reg(guard->reg())); |
| 795 macro_assembler->IfRegisterGE(guard->reg(), |
| 796 guard->value(), |
| 797 trace->backtrack()); |
| 798 break; |
| 799 case Guard::GEQ: |
| 800 ASSERT(!trace->mentions_reg(guard->reg())); |
| 801 macro_assembler->IfRegisterLT(guard->reg(), |
| 802 guard->value(), |
| 803 trace->backtrack()); |
| 804 break; |
| 805 } |
| 806 } |
| 807 |
| 808 |
| 809 void NegativeSubmatchSuccess::Emit(RegExpCompiler* compiler, Trace* trace) { |
| 810 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 811 |
| 812 // Omit flushing the trace. We discard the entire stack frame anyway. |
| 813 |
| 814 if (!label()->IsBound()) { |
| 815 // We are completely independent of the trace, since we ignore it, |
| 816 // so this code can be used as the generic version. |
| 817 assembler->BindBlock(label()); |
| 818 } |
| 819 |
| 820 // Throw away everything on the backtrack stack since the start |
| 821 // of the negative submatch and restore the character position. |
| 822 assembler->ReadCurrentPositionFromRegister(current_position_register_); |
| 823 assembler->ReadStackPointerFromRegister(stack_pointer_register_); |
| 824 if (clear_capture_count_ > 0) { |
| 825 // Clear any captures that might have been performed during the success |
| 826 // of the body of the negative look-ahead. |
| 827 intptr_t clear_capture_end = |
| 828 clear_capture_start_ + clear_capture_count_ - 1; |
| 829 assembler->ClearRegisters(clear_capture_start_, clear_capture_end); |
| 830 } |
| 831 // Now that we have unwound the stack we find at the top of the stack the |
| 832 // backtrack that the BeginSubmatch node got. |
| 833 assembler->Backtrack(); |
| 834 } |
| 835 |
| 836 |
| 837 bool Trace::GetStoredPosition(intptr_t reg, intptr_t* cp_offset) { |
| 838 ASSERT(*cp_offset == 0); |
| 839 for (DeferredAction* action = actions_; |
| 840 action != NULL; |
| 841 action = action->next()) { |
| 842 if (action->Mentions(reg)) { |
| 843 if (action->action_type() == ActionNode::STORE_POSITION) { |
| 844 *cp_offset = static_cast<DeferredCapture*>(action)->cp_offset(); |
| 845 return true; |
| 846 } else { |
| 847 return false; |
| 848 } |
| 849 } |
| 850 } |
| 851 return false; |
| 852 } |
| 853 |
| 854 |
| 855 // This is called as we come into a loop choice node and some other tricky |
| 856 // nodes. It normalizes the state of the code generator to ensure we can |
| 857 // generate generic code. |
| 858 intptr_t Trace::FindAffectedRegisters(OutSet* affected_registers, |
| 859 Isolate* isolate) { |
| 860 intptr_t max_register = RegExpCompiler::kNoRegister; |
| 861 for (DeferredAction* action = actions_; |
| 862 action != NULL; |
| 863 action = action->next()) { |
| 864 if (action->action_type() == ActionNode::CLEAR_CAPTURES) { |
| 865 Interval range = static_cast<DeferredClearCaptures*>(action)->range(); |
| 866 for (intptr_t i = range.from(); i <= range.to(); i++) |
| 867 affected_registers->Set(i, isolate); |
| 868 if (range.to() > max_register) max_register = range.to(); |
| 869 } else { |
| 870 affected_registers->Set(action->reg(), isolate); |
| 871 if (action->reg() > max_register) max_register = action->reg(); |
| 872 } |
| 873 } |
| 874 return max_register; |
| 875 } |
| 876 |
| 877 |
| 878 bool Trace::DeferredAction::Mentions(intptr_t that) { |
| 879 if (action_type() == ActionNode::CLEAR_CAPTURES) { |
| 880 Interval range = static_cast<DeferredClearCaptures*>(this)->range(); |
| 881 return range.Contains(that); |
| 882 } else { |
| 883 return reg() == that; |
| 884 } |
| 885 } |
| 886 |
| 887 |
| 888 bool Trace::mentions_reg(intptr_t reg) { |
| 889 for (DeferredAction* action = actions_; |
| 890 action != NULL; |
| 891 action = action->next()) { |
| 892 if (action->Mentions(reg)) |
| 893 return true; |
| 894 } |
| 895 return false; |
| 896 } |
| 897 |
| 898 |
| 899 void Trace::RestoreAffectedRegisters(RegExpMacroAssembler* assembler, |
| 900 intptr_t max_register, |
| 901 const OutSet& registers_to_pop, |
| 902 const OutSet& registers_to_clear) { |
| 903 for (intptr_t reg = max_register; reg >= 0; reg--) { |
| 904 if (registers_to_pop.Get(reg)) { |
| 905 assembler->PopRegister(reg); |
| 906 } else if (registers_to_clear.Get(reg)) { |
| 907 intptr_t clear_to = reg; |
| 908 while (reg > 0 && registers_to_clear.Get(reg - 1)) { |
| 909 reg--; |
| 910 } |
| 911 assembler->ClearRegisters(reg, clear_to); |
| 912 } |
| 913 } |
| 914 } |
| 915 |
| 916 |
| 917 void Trace::PerformDeferredActions(RegExpMacroAssembler* assembler, |
| 918 intptr_t max_register, |
| 919 const OutSet& affected_registers, |
| 920 OutSet* registers_to_pop, |
| 921 OutSet* registers_to_clear, |
| 922 Isolate* isolate) { |
| 923 for (intptr_t reg = 0; reg <= max_register; reg++) { |
| 924 if (!affected_registers.Get(reg)) { |
| 925 continue; |
| 926 } |
| 927 |
| 928 // The chronologically first deferred action in the trace |
| 929 // is used to infer the action needed to restore a register |
| 930 // to its previous state (or not, if it's safe to ignore it). |
| 931 enum DeferredActionUndoType { IGNORE, RESTORE, CLEAR }; |
| 932 DeferredActionUndoType undo_action = IGNORE; |
| 933 |
| 934 intptr_t value = 0; |
| 935 bool absolute = false; |
| 936 bool clear = false; |
| 937 intptr_t store_position = -1; |
| 938 // This is a little tricky because we are scanning the actions in reverse |
| 939 // historical order (newest first). |
| 940 for (DeferredAction* action = actions_; |
| 941 action != NULL; |
| 942 action = action->next()) { |
| 943 if (action->Mentions(reg)) { |
| 944 switch (action->action_type()) { |
| 945 case ActionNode::SET_REGISTER: { |
| 946 Trace::DeferredSetRegister* psr = |
| 947 static_cast<Trace::DeferredSetRegister*>(action); |
| 948 if (!absolute) { |
| 949 value += psr->value(); |
| 950 absolute = true; |
| 951 } |
| 952 // SET_REGISTER is currently only used for newly introduced loop |
| 953 // counters. They can have a significant previous value if they |
| 954 // occour in a loop. TODO(lrn): Propagate this information, so |
| 955 // we can set undo_action to IGNORE if we know there is no value to |
| 956 // restore. |
| 957 undo_action = RESTORE; |
| 958 ASSERT(store_position == -1); |
| 959 ASSERT(!clear); |
| 960 break; |
| 961 } |
| 962 case ActionNode::INCREMENT_REGISTER: |
| 963 if (!absolute) { |
| 964 value++; |
| 965 } |
| 966 ASSERT(store_position == -1); |
| 967 ASSERT(!clear); |
| 968 undo_action = RESTORE; |
| 969 break; |
| 970 case ActionNode::STORE_POSITION: { |
| 971 Trace::DeferredCapture* pc = |
| 972 static_cast<Trace::DeferredCapture*>(action); |
| 973 if (!clear && store_position == -1) { |
| 974 store_position = pc->cp_offset(); |
| 975 } |
| 976 |
| 977 // For captures we know that stores and clears alternate. |
| 978 // Other register, are never cleared, and if the occur |
| 979 // inside a loop, they might be assigned more than once. |
| 980 if (reg <= 1) { |
| 981 // Registers zero and one, aka "capture zero", is |
| 982 // always set correctly if we succeed. There is no |
| 983 // need to undo a setting on backtrack, because we |
| 984 // will set it again or fail. |
| 985 undo_action = IGNORE; |
| 986 } else { |
| 987 undo_action = pc->is_capture() ? CLEAR : RESTORE; |
| 988 } |
| 989 ASSERT(!absolute); |
| 990 ASSERT(value == 0); |
| 991 break; |
| 992 } |
| 993 case ActionNode::CLEAR_CAPTURES: { |
| 994 // Since we're scanning in reverse order, if we've already |
| 995 // set the position we have to ignore historically earlier |
| 996 // clearing operations. |
| 997 if (store_position == -1) { |
| 998 clear = true; |
| 999 } |
| 1000 undo_action = RESTORE; |
| 1001 ASSERT(!absolute); |
| 1002 ASSERT(value == 0); |
| 1003 break; |
| 1004 } |
| 1005 default: |
| 1006 UNREACHABLE(); |
| 1007 break; |
| 1008 } |
| 1009 } |
| 1010 } |
| 1011 // Prepare for the undo-action (e.g., push if it's going to be popped). |
| 1012 if (undo_action == RESTORE) { |
| 1013 assembler->PushRegister(reg); |
| 1014 registers_to_pop->Set(reg, isolate); |
| 1015 } else if (undo_action == CLEAR) { |
| 1016 registers_to_clear->Set(reg, isolate); |
| 1017 } |
| 1018 // Perform the chronologically last action (or accumulated increment) |
| 1019 // for the register. |
| 1020 if (store_position != -1) { |
| 1021 assembler->WriteCurrentPositionToRegister(reg, store_position); |
| 1022 } else if (clear) { |
| 1023 assembler->ClearRegisters(reg, reg); |
| 1024 } else if (absolute) { |
| 1025 assembler->SetRegister(reg, value); |
| 1026 } else if (value != 0) { |
| 1027 assembler->AdvanceRegister(reg, value); |
| 1028 } |
| 1029 } |
| 1030 } |
| 1031 |
| 1032 |
| 1033 void Trace::Flush(RegExpCompiler* compiler, RegExpNode* successor) { |
| 1034 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 1035 |
| 1036 ASSERT(!is_trivial()); |
| 1037 |
| 1038 if (actions_ == NULL && backtrack() == NULL) { |
| 1039 // Here we just have some deferred cp advances to fix and we are back to |
| 1040 // a normal situation. We may also have to forget some information gained |
| 1041 // through a quick check that was already performed. |
| 1042 if (cp_offset_ != 0) assembler->AdvanceCurrentPosition(cp_offset_); |
| 1043 // Create a new trivial state and generate the node with that. |
| 1044 Trace new_state; |
| 1045 successor->Emit(compiler, &new_state); |
| 1046 return; |
| 1047 } |
| 1048 |
| 1049 // Generate deferred actions here along with code to undo them again. |
| 1050 OutSet affected_registers; |
| 1051 |
| 1052 if (backtrack() != NULL) { |
| 1053 // Here we have a concrete backtrack location. These are set up by choice |
| 1054 // nodes and so they indicate that we have a deferred save of the current |
| 1055 // position which we may need to emit here. |
| 1056 assembler->PushCurrentPosition(); |
| 1057 } |
| 1058 |
| 1059 intptr_t max_register = FindAffectedRegisters(&affected_registers, CI); |
| 1060 OutSet registers_to_pop; |
| 1061 OutSet registers_to_clear; |
| 1062 PerformDeferredActions(assembler, |
| 1063 max_register, |
| 1064 affected_registers, |
| 1065 ®isters_to_pop, |
| 1066 ®isters_to_clear, |
| 1067 CI); |
| 1068 if (cp_offset_ != 0) { |
| 1069 assembler->AdvanceCurrentPosition(cp_offset_); |
| 1070 } |
| 1071 |
| 1072 // Create a new trivial state and generate the node with that. |
| 1073 BlockLabel undo; |
| 1074 assembler->PushBacktrack(&undo); |
| 1075 Trace new_state; |
| 1076 successor->Emit(compiler, &new_state); |
| 1077 |
| 1078 // On backtrack we need to restore state. |
| 1079 assembler->BindBlock(&undo); |
| 1080 RestoreAffectedRegisters(assembler, |
| 1081 max_register, |
| 1082 registers_to_pop, |
| 1083 registers_to_clear); |
| 1084 if (backtrack() == NULL) { |
| 1085 assembler->Backtrack(); |
| 1086 } else { |
| 1087 assembler->PopCurrentPosition(); |
| 1088 assembler->GoTo(backtrack()); |
| 1089 } |
| 1090 } |
| 1091 |
| 1092 |
| 1093 void Trace::InvalidateCurrentCharacter() { |
| 1094 characters_preloaded_ = 0; |
| 1095 } |
| 1096 |
| 1097 |
| 1098 void Trace::AdvanceCurrentPositionInTrace(intptr_t by, |
| 1099 RegExpCompiler* compiler) { |
| 1100 ASSERT(by > 0); |
| 1101 // We don't have an instruction for shifting the current character register |
| 1102 // down or for using a shifted value for anything so lets just forget that |
| 1103 // we preloaded any characters into it. |
| 1104 characters_preloaded_ = 0; |
| 1105 // Adjust the offsets of the quick check performed information. This |
| 1106 // information is used to find out what we already determined about the |
| 1107 // characters by means of mask and compare. |
| 1108 quick_check_performed_.Advance(by, compiler->ascii()); |
| 1109 cp_offset_ += by; |
| 1110 if (cp_offset_ > RegExpMacroAssembler::kMaxCPOffset) { |
| 1111 compiler->SetRegExpTooBig(); |
| 1112 cp_offset_ = 0; |
| 1113 } |
| 1114 bound_checked_up_to_ = Utils::Maximum(static_cast<intptr_t>(0), |
| 1115 bound_checked_up_to_ - by); |
| 1116 } |
| 1117 |
| 1118 |
| 1119 void EndNode::Emit(RegExpCompiler* compiler, Trace* trace) { |
| 1120 if (!trace->is_trivial()) { |
| 1121 trace->Flush(compiler, this); |
| 1122 return; |
| 1123 } |
| 1124 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 1125 if (!label()->IsBound()) { |
| 1126 assembler->BindBlock(label()); |
| 1127 } |
| 1128 switch (action_) { |
| 1129 case ACCEPT: |
| 1130 assembler->Succeed(); |
| 1131 return; |
| 1132 case BACKTRACK: |
| 1133 assembler->GoTo(trace->backtrack()); |
| 1134 return; |
| 1135 case NEGATIVE_SUBMATCH_SUCCESS: |
| 1136 // This case is handled in a different virtual method. |
| 1137 UNREACHABLE(); |
| 1138 } |
| 1139 UNIMPLEMENTED(); |
| 1140 } |
| 1141 |
| 1142 |
| 1143 bool QuickCheckDetails::Rationalize(bool asc) { |
| 1144 bool found_useful_op = false; |
| 1145 uint32_t char_mask; |
| 1146 if (asc) { |
| 1147 char_mask = Symbols::kMaxOneCharCodeSymbol; |
| 1148 } else { |
| 1149 char_mask = Utf16::kMaxCodeUnit; |
| 1150 } |
| 1151 mask_ = 0; |
| 1152 value_ = 0; |
| 1153 intptr_t char_shift = 0; |
| 1154 for (intptr_t i = 0; i < characters_; i++) { |
| 1155 Position* pos = &positions_[i]; |
| 1156 if ((pos->mask & Symbols::kMaxOneCharCodeSymbol) != 0) { |
| 1157 found_useful_op = true; |
| 1158 } |
| 1159 mask_ |= (pos->mask & char_mask) << char_shift; |
| 1160 value_ |= (pos->value & char_mask) << char_shift; |
| 1161 char_shift += asc ? 8 : 16; |
| 1162 } |
| 1163 return found_useful_op; |
| 1164 } |
| 1165 |
| 1166 |
| 1167 bool RegExpNode::EmitQuickCheck(RegExpCompiler* compiler, |
| 1168 Trace* trace, |
| 1169 bool preload_has_checked_bounds, |
| 1170 BlockLabel* on_possible_success, |
| 1171 QuickCheckDetails* details, |
| 1172 bool fall_through_on_failure) { |
| 1173 if (details->characters() == 0) return false; |
| 1174 GetQuickCheckDetails( |
| 1175 details, compiler, 0, trace->at_start() == Trace::FALSE_VALUE); |
| 1176 if (details->cannot_match()) return false; |
| 1177 if (!details->Rationalize(compiler->ascii())) return false; |
| 1178 ASSERT(details->characters() == 1 || |
| 1179 compiler->macro_assembler()->CanReadUnaligned()); |
| 1180 uint32_t mask = details->mask(); |
| 1181 uint32_t value = details->value(); |
| 1182 |
| 1183 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 1184 |
| 1185 if (trace->characters_preloaded() != details->characters()) { |
| 1186 assembler->LoadCurrentCharacter(trace->cp_offset(), |
| 1187 trace->backtrack(), |
| 1188 !preload_has_checked_bounds, |
| 1189 details->characters()); |
| 1190 } |
| 1191 |
| 1192 |
| 1193 bool need_mask = true; |
| 1194 |
| 1195 if (details->characters() == 1) { |
| 1196 // If number of characters preloaded is 1 then we used a byte or 16 bit |
| 1197 // load so the value is already masked down. |
| 1198 uint32_t char_mask; |
| 1199 if (compiler->ascii()) { |
| 1200 char_mask = Symbols::kMaxOneCharCodeSymbol; |
| 1201 } else { |
| 1202 char_mask = Utf16::kMaxCodeUnit; |
| 1203 } |
| 1204 if ((mask & char_mask) == char_mask) need_mask = false; |
| 1205 mask &= char_mask; |
| 1206 } else { |
| 1207 // For 2-character preloads in ASCII mode or 1-character preloads in |
| 1208 // TWO_BYTE mode we also use a 16 bit load with zero extend. |
| 1209 if (details->characters() == 2 && compiler->ascii()) { |
| 1210 if ((mask & 0xffff) == 0xffff) need_mask = false; |
| 1211 } else if (details->characters() == 1 && !compiler->ascii()) { |
| 1212 if ((mask & 0xffff) == 0xffff) need_mask = false; |
| 1213 } else { |
| 1214 if (mask == 0xffffffff) need_mask = false; |
| 1215 } |
| 1216 } |
| 1217 |
| 1218 if (fall_through_on_failure) { |
| 1219 if (need_mask) { |
| 1220 assembler->CheckCharacterAfterAnd(value, mask, on_possible_success); |
| 1221 } else { |
| 1222 assembler->CheckCharacter(value, on_possible_success); |
| 1223 } |
| 1224 } else { |
| 1225 if (need_mask) { |
| 1226 assembler->CheckNotCharacterAfterAnd(value, mask, trace->backtrack()); |
| 1227 } else { |
| 1228 assembler->CheckNotCharacter(value, trace->backtrack()); |
| 1229 } |
| 1230 } |
| 1231 return true; |
| 1232 } |
| 1233 |
| 1234 |
| 1235 // Emit the code to check for a ^ in multiline mode (1-character lookbehind |
| 1236 // that matches newline or the start of input). |
| 1237 static void EmitHat(RegExpCompiler* compiler, |
| 1238 RegExpNode* on_success, |
| 1239 Trace* trace) { |
| 1240 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 1241 // We will be loading the previous character into the current character |
| 1242 // register. |
| 1243 Trace new_trace(*trace); |
| 1244 new_trace.InvalidateCurrentCharacter(); |
| 1245 |
| 1246 BlockLabel ok; |
| 1247 if (new_trace.cp_offset() == 0) { |
| 1248 // The start of input counts as a newline in this context, so skip to |
| 1249 // ok if we are at the start. |
| 1250 assembler->CheckAtStart(&ok); |
| 1251 } |
| 1252 // We already checked that we are not at the start of input so it must be |
| 1253 // OK to load the previous character. |
| 1254 assembler->LoadCurrentCharacter(new_trace.cp_offset() -1, |
| 1255 new_trace.backtrack(), |
| 1256 false); |
| 1257 if (!assembler->CheckSpecialCharacterClass('n', |
| 1258 new_trace.backtrack())) { |
| 1259 // Newline means \n, \r, 0x2028 or 0x2029. |
| 1260 if (!compiler->ascii()) { |
| 1261 assembler->CheckCharacterAfterAnd(0x2028, 0xfffe, &ok); |
| 1262 } |
| 1263 assembler->CheckCharacter('\n', &ok); |
| 1264 assembler->CheckNotCharacter('\r', new_trace.backtrack()); |
| 1265 } |
| 1266 assembler->BindBlock(&ok); |
| 1267 on_success->Emit(compiler, &new_trace); |
| 1268 } |
| 1269 |
| 1270 |
| 1271 void AssertionNode::Emit(RegExpCompiler* compiler, Trace* trace) { |
| 1272 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 1273 switch (assertion_type_) { |
| 1274 case AT_END: { |
| 1275 BlockLabel ok; |
| 1276 assembler->CheckPosition(trace->cp_offset(), &ok); |
| 1277 assembler->GoTo(trace->backtrack()); |
| 1278 assembler->BindBlock(&ok); |
| 1279 break; |
| 1280 } |
| 1281 case AT_START: { |
| 1282 if (trace->at_start() == Trace::FALSE_VALUE) { |
| 1283 assembler->GoTo(trace->backtrack()); |
| 1284 return; |
| 1285 } |
| 1286 if (trace->at_start() == Trace::UNKNOWN) { |
| 1287 assembler->CheckNotAtStart(trace->backtrack()); |
| 1288 Trace at_start_trace = *trace; |
| 1289 at_start_trace.set_at_start(true); |
| 1290 on_success()->Emit(compiler, &at_start_trace); |
| 1291 return; |
| 1292 } |
| 1293 } |
| 1294 break; |
| 1295 case AFTER_NEWLINE: |
| 1296 EmitHat(compiler, on_success(), trace); |
| 1297 return; |
| 1298 case AT_BOUNDARY: |
| 1299 case AT_NON_BOUNDARY: { |
| 1300 EmitBoundaryCheck(compiler, trace); |
| 1301 return; |
| 1302 } |
| 1303 } |
| 1304 on_success()->Emit(compiler, trace); |
| 1305 } |
| 1306 |
| 1307 |
| 1308 RegExpNode::LimitResult RegExpNode::LimitVersions(RegExpCompiler* compiler, |
| 1309 Trace* trace) { |
| 1310 // If we are generating a greedy loop then don't stop and don't reuse code. |
| 1311 if (trace->stop_node() != NULL) { |
| 1312 return CONTINUE; |
| 1313 } |
| 1314 |
| 1315 RegExpMacroAssembler* macro_assembler = compiler->macro_assembler(); |
| 1316 if (trace->is_trivial()) { |
| 1317 if (label_.IsBound()) { |
| 1318 // We are being asked to generate a generic version, but that's already |
| 1319 // been done so just go to it. |
| 1320 macro_assembler->GoTo(&label_); |
| 1321 return DONE; |
| 1322 } |
| 1323 if (compiler->recursion_depth() >= RegExpCompiler::kMaxRecursion) { |
| 1324 // To avoid too deep recursion we push the node to the work queue and just |
| 1325 // generate a goto here. |
| 1326 compiler->AddWork(this); |
| 1327 macro_assembler->GoTo(&label_); |
| 1328 return DONE; |
| 1329 } |
| 1330 // Generate generic version of the node and bind the label for later use. |
| 1331 macro_assembler->BindBlock(&label_); |
| 1332 return CONTINUE; |
| 1333 } |
| 1334 |
| 1335 // We are being asked to make a non-generic version. Keep track of how many |
| 1336 // non-generic versions we generate so as not to overdo it. |
| 1337 trace_count_++; |
| 1338 if (kRegexpOptimization && |
| 1339 trace_count_ < kMaxCopiesCodeGenerated && |
| 1340 compiler->recursion_depth() <= RegExpCompiler::kMaxRecursion) { |
| 1341 return CONTINUE; |
| 1342 } |
| 1343 |
| 1344 // If we get here code has been generated for this node too many times or |
| 1345 // recursion is too deep. Time to switch to a generic version. The code for |
| 1346 // generic versions above can handle deep recursion properly. |
| 1347 trace->Flush(compiler, this); |
| 1348 return DONE; |
| 1349 } |
| 1350 |
| 1351 |
| 1352 // This generates the code to match a text node. A text node can contain |
| 1353 // straight character sequences (possibly to be matched in a case-independent |
| 1354 // way) and character classes. For efficiency we do not do this in a single |
| 1355 // pass from left to right. Instead we pass over the text node several times, |
| 1356 // emitting code for some character positions every time. See the comment on |
| 1357 // TextEmitPass for details. |
| 1358 void TextNode::Emit(RegExpCompiler* compiler, Trace* trace) { |
| 1359 LimitResult limit_result = LimitVersions(compiler, trace); |
| 1360 if (limit_result == DONE) return; |
| 1361 ASSERT(limit_result == CONTINUE); |
| 1362 |
| 1363 if (trace->cp_offset() + Length() > RegExpMacroAssembler::kMaxCPOffset) { |
| 1364 compiler->SetRegExpTooBig(); |
| 1365 return; |
| 1366 } |
| 1367 |
| 1368 if (compiler->ascii()) { |
| 1369 intptr_t dummy = 0; |
| 1370 TextEmitPass(compiler, NON_ASCII_MATCH, false, trace, false, &dummy); |
| 1371 } |
| 1372 |
| 1373 bool first_elt_done = false; |
| 1374 intptr_t bound_checked_to = trace->cp_offset() - 1; |
| 1375 bound_checked_to += trace->bound_checked_up_to(); |
| 1376 |
| 1377 // If a character is preloaded into the current character register then |
| 1378 // check that now. |
| 1379 if (trace->characters_preloaded() == 1) { |
| 1380 for (intptr_t pass = kFirstRealPass; pass <= kLastPass; pass++) { |
| 1381 if (!SkipPass(pass, compiler->ignore_case())) { |
| 1382 TextEmitPass(compiler, |
| 1383 static_cast<TextEmitPassType>(pass), |
| 1384 true, |
| 1385 trace, |
| 1386 false, |
| 1387 &bound_checked_to); |
| 1388 } |
| 1389 } |
| 1390 first_elt_done = true; |
| 1391 } |
| 1392 |
| 1393 for (intptr_t pass = kFirstRealPass; pass <= kLastPass; pass++) { |
| 1394 if (!SkipPass(pass, compiler->ignore_case())) { |
| 1395 TextEmitPass(compiler, |
| 1396 static_cast<TextEmitPassType>(pass), |
| 1397 false, |
| 1398 trace, |
| 1399 first_elt_done, |
| 1400 &bound_checked_to); |
| 1401 } |
| 1402 } |
| 1403 |
| 1404 Trace successor_trace(*trace); |
| 1405 successor_trace.set_at_start(false); |
| 1406 successor_trace.AdvanceCurrentPositionInTrace(Length(), compiler); |
| 1407 RecursionCheck rc(compiler); |
| 1408 on_success()->Emit(compiler, &successor_trace); |
| 1409 } |
| 1410 |
| 1411 |
| 1412 void LoopChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) { |
| 1413 RegExpMacroAssembler* macro_assembler = compiler->macro_assembler(); |
| 1414 if (trace->stop_node() == this) { |
| 1415 intptr_t text_length = |
| 1416 GreedyLoopTextLengthForAlternative(&((*alternatives_)[0])); |
| 1417 ASSERT(text_length != kNodeIsTooComplexForGreedyLoops); |
| 1418 // Update the counter-based backtracking info on the stack. This is an |
| 1419 // optimization for greedy loops (see below). |
| 1420 ASSERT(trace->cp_offset() == text_length); |
| 1421 macro_assembler->AdvanceCurrentPosition(text_length); |
| 1422 macro_assembler->GoTo(trace->loop_label()); |
| 1423 return; |
| 1424 } |
| 1425 ASSERT(trace->stop_node() == NULL); |
| 1426 if (!trace->is_trivial()) { |
| 1427 trace->Flush(compiler, this); |
| 1428 return; |
| 1429 } |
| 1430 ChoiceNode::Emit(compiler, trace); |
| 1431 } |
| 1432 |
| 1433 |
| 1434 // This class is used when generating the alternatives in a choice node. It |
| 1435 // records the way the alternative is being code generated. |
| 1436 struct AlternativeGeneration { |
| 1437 AlternativeGeneration() |
| 1438 : possible_success(), |
| 1439 expects_preload(false), |
| 1440 after(), |
| 1441 quick_check_details() { } |
| 1442 BlockLabel possible_success; |
| 1443 bool expects_preload; |
| 1444 BlockLabel after; |
| 1445 QuickCheckDetails quick_check_details; |
| 1446 }; |
| 1447 |
| 1448 |
| 1449 // Creates a list of AlternativeGenerations. If the list has a reasonable |
| 1450 // size then it is on the stack, otherwise the excess is on the heap. |
| 1451 class AlternativeGenerationList { |
| 1452 public: |
| 1453 explicit AlternativeGenerationList(intptr_t count) |
| 1454 : alt_gens_(count) { |
| 1455 for (intptr_t i = 0; i < count && i < kAFew; i++) { |
| 1456 alt_gens_.Add(a_few_alt_gens_ + i); |
| 1457 } |
| 1458 for (intptr_t i = kAFew; i < count; i++) { |
| 1459 alt_gens_.Add(new AlternativeGeneration()); |
| 1460 } |
| 1461 } |
| 1462 ~AlternativeGenerationList() { |
| 1463 for (intptr_t i = kAFew; i < alt_gens_.length(); i++) { |
| 1464 delete alt_gens_[i]; |
| 1465 alt_gens_[i] = NULL; |
| 1466 } |
| 1467 } |
| 1468 |
| 1469 AlternativeGeneration* at(intptr_t i) { |
| 1470 return alt_gens_[i]; |
| 1471 } |
| 1472 |
| 1473 private: |
| 1474 static const intptr_t kAFew = 10; |
| 1475 GrowableArray<AlternativeGeneration*> alt_gens_; |
| 1476 AlternativeGeneration a_few_alt_gens_[kAFew]; |
| 1477 |
| 1478 DISALLOW_ALLOCATION(); |
| 1479 }; |
| 1480 |
| 1481 |
| 1482 void ChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) { |
| 1483 RegExpMacroAssembler* macro_assembler = compiler->macro_assembler(); |
| 1484 intptr_t choice_count = alternatives_->length(); |
| 1485 |
| 1486 #ifdef DEBUG |
| 1487 for (intptr_t i = 0; i < choice_count - 1; i++) { |
| 1488 GuardedAlternative alternative = alternatives_->At(i); |
| 1489 ZoneGrowableArray<Guard*>* guards = alternative.guards(); |
| 1490 intptr_t guard_count = (guards == NULL) ? 0 : guards->length(); |
| 1491 for (intptr_t j = 0; j < guard_count; j++) { |
| 1492 ASSERT(!trace->mentions_reg(guards->At(j)->reg())); |
| 1493 } |
| 1494 } |
| 1495 #endif |
| 1496 |
| 1497 LimitResult limit_result = LimitVersions(compiler, trace); |
| 1498 if (limit_result == DONE) return; |
| 1499 ASSERT(limit_result == CONTINUE); |
| 1500 |
| 1501 intptr_t new_flush_budget = trace->flush_budget() / choice_count; |
| 1502 if (trace->flush_budget() == 0 && trace->actions() != NULL) { |
| 1503 trace->Flush(compiler, this); |
| 1504 return; |
| 1505 } |
| 1506 |
| 1507 RecursionCheck rc(compiler); |
| 1508 |
| 1509 Trace* current_trace = trace; |
| 1510 |
| 1511 intptr_t text_length = |
| 1512 GreedyLoopTextLengthForAlternative(&((*alternatives_)[0])); |
| 1513 bool greedy_loop = false; |
| 1514 BlockLabel greedy_loop_label; |
| 1515 Trace counter_backtrack_trace; |
| 1516 counter_backtrack_trace.set_backtrack(&greedy_loop_label); |
| 1517 if (not_at_start()) counter_backtrack_trace.set_at_start(false); |
| 1518 |
| 1519 if (choice_count > 1 && text_length != kNodeIsTooComplexForGreedyLoops) { |
| 1520 // Here we have special handling for greedy loops containing only text nodes |
| 1521 // and other simple nodes. These are handled by pushing the current |
| 1522 // position on the stack and then incrementing the current position each |
| 1523 // time around the switch. On backtrack we decrement the current position |
| 1524 // and check it against the pushed value. This avoids pushing backtrack |
| 1525 // information for each iteration of the loop, which could take up a lot of |
| 1526 // space. |
| 1527 greedy_loop = true; |
| 1528 ASSERT(trace->stop_node() == NULL); |
| 1529 macro_assembler->PushCurrentPosition(); |
| 1530 current_trace = &counter_backtrack_trace; |
| 1531 BlockLabel greedy_match_failed; |
| 1532 Trace greedy_match_trace; |
| 1533 if (not_at_start()) greedy_match_trace.set_at_start(false); |
| 1534 greedy_match_trace.set_backtrack(&greedy_match_failed); |
| 1535 BlockLabel loop_label; |
| 1536 macro_assembler->BindBlock(&loop_label); |
| 1537 greedy_match_trace.set_stop_node(this); |
| 1538 greedy_match_trace.set_loop_label(&loop_label); |
| 1539 (*alternatives_)[0].node()->Emit(compiler, &greedy_match_trace); |
| 1540 macro_assembler->BindBlock(&greedy_match_failed); |
| 1541 } |
| 1542 |
| 1543 BlockLabel second_choice; // For use in greedy matches. |
| 1544 macro_assembler->BindBlock(&second_choice); |
| 1545 |
| 1546 intptr_t first_normal_choice = greedy_loop ? 1 : 0; |
| 1547 |
| 1548 bool not_at_start = current_trace->at_start() == Trace::FALSE_VALUE; |
| 1549 const intptr_t kEatsAtLeastNotYetInitialized = -1; |
| 1550 intptr_t eats_at_least = kEatsAtLeastNotYetInitialized; |
| 1551 |
| 1552 bool skip_was_emitted = false; |
| 1553 |
| 1554 if (!greedy_loop && choice_count == 2) { |
| 1555 GuardedAlternative alt1 = (*alternatives_)[1]; |
| 1556 if (alt1.guards() == NULL || alt1.guards()->length() == 0) { |
| 1557 RegExpNode* eats_anything_node = alt1.node(); |
| 1558 if (eats_anything_node->GetSuccessorOfOmnivorousTextNode(compiler) == |
| 1559 this) { |
| 1560 // At this point we know that we are at a non-greedy loop that will eat |
| 1561 // any character one at a time. Any non-anchored regexp has such a |
| 1562 // loop prepended to it in order to find where it starts. We look for |
| 1563 // a pattern of the form ...abc... where we can look 6 characters ahead |
| 1564 // and step forwards 3 if the character is not one of abc. Abc need |
| 1565 // not be atoms, they can be any reasonably limited character class or |
| 1566 // small alternation. |
| 1567 ASSERT(trace->is_trivial()); // This is the case on LoopChoiceNodes. |
| 1568 BoyerMooreLookahead* lookahead = bm_info(not_at_start); |
| 1569 if (lookahead == NULL) { |
| 1570 eats_at_least = Utils::Minimum(kMaxLookaheadForBoyerMoore, |
| 1571 EatsAtLeast(kMaxLookaheadForBoyerMoore, |
| 1572 kRecursionBudget, |
| 1573 not_at_start)); |
| 1574 if (eats_at_least >= 1) { |
| 1575 BoyerMooreLookahead* bm = |
| 1576 new(I) BoyerMooreLookahead(eats_at_least, compiler, I); |
| 1577 GuardedAlternative alt0 = alternatives_->At(0); |
| 1578 alt0.node()->FillInBMInfo(0, kRecursionBudget, bm, not_at_start); |
| 1579 skip_was_emitted = bm->EmitSkipInstructions(macro_assembler); |
| 1580 } |
| 1581 } else { |
| 1582 skip_was_emitted = lookahead->EmitSkipInstructions(macro_assembler); |
| 1583 } |
| 1584 } |
| 1585 } |
| 1586 } |
| 1587 |
| 1588 if (eats_at_least == kEatsAtLeastNotYetInitialized) { |
| 1589 // Save some time by looking at most one machine word ahead. |
| 1590 eats_at_least = |
| 1591 EatsAtLeast(compiler->ascii() ? 4 : 2, kRecursionBudget, not_at_start); |
| 1592 } |
| 1593 intptr_t preload_characters = |
| 1594 CalculatePreloadCharacters(compiler, eats_at_least); |
| 1595 |
| 1596 bool preload_is_current = !skip_was_emitted && |
| 1597 (current_trace->characters_preloaded() == preload_characters); |
| 1598 bool preload_has_checked_bounds = preload_is_current; |
| 1599 |
| 1600 AlternativeGenerationList alt_gens(choice_count); |
| 1601 |
| 1602 // For now we just call all choices one after the other. The idea ultimately |
| 1603 // is to use the Dispatch table to try only the relevant ones. |
| 1604 for (intptr_t i = first_normal_choice; i < choice_count; i++) { |
| 1605 GuardedAlternative alternative = alternatives_->At(i); |
| 1606 AlternativeGeneration* alt_gen = alt_gens.at(i); |
| 1607 alt_gen->quick_check_details.set_characters(preload_characters); |
| 1608 ZoneGrowableArray<Guard*>* guards = alternative.guards(); |
| 1609 intptr_t guard_count = (guards == NULL) ? 0 : guards->length(); |
| 1610 Trace new_trace(*current_trace); |
| 1611 new_trace.set_characters_preloaded(preload_is_current ? |
| 1612 preload_characters : |
| 1613 0); |
| 1614 if (preload_has_checked_bounds) { |
| 1615 new_trace.set_bound_checked_up_to(preload_characters); |
| 1616 } |
| 1617 new_trace.quick_check_performed()->Clear(); |
| 1618 if (not_at_start_) new_trace.set_at_start(Trace::FALSE_VALUE); |
| 1619 alt_gen->expects_preload = preload_is_current; |
| 1620 bool generate_full_check_inline = false; |
| 1621 if (kRegexpOptimization && |
| 1622 try_to_emit_quick_check_for_alternative(i) && |
| 1623 alternative.node()->EmitQuickCheck(compiler, |
| 1624 &new_trace, |
| 1625 preload_has_checked_bounds, |
| 1626 &alt_gen->possible_success, |
| 1627 &alt_gen->quick_check_details, |
| 1628 i < choice_count - 1)) { |
| 1629 // Quick check was generated for this choice. |
| 1630 preload_is_current = true; |
| 1631 preload_has_checked_bounds = true; |
| 1632 // On the last choice in the ChoiceNode we generated the quick |
| 1633 // check to fall through on possible success. So now we need to |
| 1634 // generate the full check inline. |
| 1635 if (i == choice_count - 1) { |
| 1636 macro_assembler->BindBlock(&alt_gen->possible_success); |
| 1637 new_trace.set_quick_check_performed(&alt_gen->quick_check_details); |
| 1638 new_trace.set_characters_preloaded(preload_characters); |
| 1639 new_trace.set_bound_checked_up_to(preload_characters); |
| 1640 generate_full_check_inline = true; |
| 1641 } |
| 1642 } else if (alt_gen->quick_check_details.cannot_match()) { |
| 1643 if (i == choice_count - 1 && !greedy_loop) { |
| 1644 macro_assembler->GoTo(trace->backtrack()); |
| 1645 } |
| 1646 continue; |
| 1647 } else { |
| 1648 // No quick check was generated. Put the full code here. |
| 1649 // If this is not the first choice then there could be slow checks from |
| 1650 // previous cases that go here when they fail. There's no reason to |
| 1651 // insist that they preload characters since the slow check we are about |
| 1652 // to generate probably can't use it. |
| 1653 if (i != first_normal_choice) { |
| 1654 alt_gen->expects_preload = false; |
| 1655 new_trace.InvalidateCurrentCharacter(); |
| 1656 } |
| 1657 if (i < choice_count - 1) { |
| 1658 new_trace.set_backtrack(&alt_gen->after); |
| 1659 } |
| 1660 generate_full_check_inline = true; |
| 1661 } |
| 1662 if (generate_full_check_inline) { |
| 1663 if (new_trace.actions() != NULL) { |
| 1664 new_trace.set_flush_budget(new_flush_budget); |
| 1665 } |
| 1666 for (intptr_t j = 0; j < guard_count; j++) { |
| 1667 GenerateGuard(macro_assembler, guards->At(j), &new_trace); |
| 1668 } |
| 1669 alternative.node()->Emit(compiler, &new_trace); |
| 1670 preload_is_current = false; |
| 1671 } |
| 1672 macro_assembler->BindBlock(&alt_gen->after); |
| 1673 } |
| 1674 if (greedy_loop) { |
| 1675 macro_assembler->BindBlock(&greedy_loop_label); |
| 1676 // If we have unwound to the bottom then backtrack. |
| 1677 macro_assembler->CheckGreedyLoop(trace->backtrack()); |
| 1678 // Otherwise try the second priority at an earlier position. |
| 1679 macro_assembler->AdvanceCurrentPosition(-text_length); |
| 1680 macro_assembler->GoTo(&second_choice); |
| 1681 } |
| 1682 |
| 1683 // At this point we need to generate slow checks for the alternatives where |
| 1684 // the quick check was inlined. We can recognize these because the associated |
| 1685 // label was bound. |
| 1686 for (intptr_t i = first_normal_choice; i < choice_count - 1; i++) { |
| 1687 AlternativeGeneration* alt_gen = alt_gens.at(i); |
| 1688 Trace new_trace(*current_trace); |
| 1689 // If there are actions to be flushed we have to limit how many times |
| 1690 // they are flushed. Take the budget of the parent trace and distribute |
| 1691 // it fairly amongst the children. |
| 1692 if (new_trace.actions() != NULL) { |
| 1693 new_trace.set_flush_budget(new_flush_budget); |
| 1694 } |
| 1695 EmitOutOfLineContinuation(compiler, |
| 1696 &new_trace, |
| 1697 alternatives_->At(i), |
| 1698 alt_gen, |
| 1699 preload_characters, |
| 1700 alt_gens.at(i + 1)->expects_preload); |
| 1701 } |
| 1702 } |
| 1703 |
| 1704 |
| 1705 void ChoiceNode::EmitOutOfLineContinuation(RegExpCompiler* compiler, |
| 1706 Trace* trace, |
| 1707 GuardedAlternative alternative, |
| 1708 AlternativeGeneration* alt_gen, |
| 1709 intptr_t preload_characters, |
| 1710 bool next_expects_preload) { |
| 1711 if (!alt_gen->possible_success.IsLinked()) return; |
| 1712 |
| 1713 RegExpMacroAssembler* macro_assembler = compiler->macro_assembler(); |
| 1714 macro_assembler->BindBlock(&alt_gen->possible_success); |
| 1715 Trace out_of_line_trace(*trace); |
| 1716 out_of_line_trace.set_characters_preloaded(preload_characters); |
| 1717 out_of_line_trace.set_quick_check_performed(&alt_gen->quick_check_details); |
| 1718 if (not_at_start_) out_of_line_trace.set_at_start(Trace::FALSE_VALUE); |
| 1719 ZoneGrowableArray<Guard*>* guards = alternative.guards(); |
| 1720 intptr_t guard_count = (guards == NULL) ? 0 : guards->length(); |
| 1721 if (next_expects_preload) { |
| 1722 BlockLabel reload_current_char; |
| 1723 out_of_line_trace.set_backtrack(&reload_current_char); |
| 1724 for (intptr_t j = 0; j < guard_count; j++) { |
| 1725 GenerateGuard(macro_assembler, guards->At(j), &out_of_line_trace); |
| 1726 } |
| 1727 alternative.node()->Emit(compiler, &out_of_line_trace); |
| 1728 macro_assembler->BindBlock(&reload_current_char); |
| 1729 // Reload the current character, since the next quick check expects that. |
| 1730 // We don't need to check bounds here because we only get into this |
| 1731 // code through a quick check which already did the checked load. |
| 1732 macro_assembler->LoadCurrentCharacter(trace->cp_offset(), |
| 1733 NULL, |
| 1734 false, |
| 1735 preload_characters); |
| 1736 macro_assembler->GoTo(&(alt_gen->after)); |
| 1737 } else { |
| 1738 out_of_line_trace.set_backtrack(&(alt_gen->after)); |
| 1739 for (intptr_t j = 0; j < guard_count; j++) { |
| 1740 GenerateGuard(macro_assembler, guards->At(j), &out_of_line_trace); |
| 1741 } |
| 1742 alternative.node()->Emit(compiler, &out_of_line_trace); |
| 1743 } |
| 1744 } |
| 1745 |
| 1746 |
| 1747 void ActionNode::Emit(RegExpCompiler* compiler, Trace* trace) { |
| 1748 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 1749 LimitResult limit_result = LimitVersions(compiler, trace); |
| 1750 if (limit_result == DONE) return; |
| 1751 ASSERT(limit_result == CONTINUE); |
| 1752 |
| 1753 RecursionCheck rc(compiler); |
| 1754 |
| 1755 switch (action_type_) { |
| 1756 case STORE_POSITION: { |
| 1757 Trace::DeferredCapture |
| 1758 new_capture(data_.u_position_register.reg, |
| 1759 data_.u_position_register.is_capture, |
| 1760 trace); |
| 1761 Trace new_trace = *trace; |
| 1762 new_trace.add_action(&new_capture); |
| 1763 on_success()->Emit(compiler, &new_trace); |
| 1764 break; |
| 1765 } |
| 1766 case INCREMENT_REGISTER: { |
| 1767 Trace::DeferredIncrementRegister |
| 1768 new_increment(data_.u_increment_register.reg); |
| 1769 Trace new_trace = *trace; |
| 1770 new_trace.add_action(&new_increment); |
| 1771 on_success()->Emit(compiler, &new_trace); |
| 1772 break; |
| 1773 } |
| 1774 case SET_REGISTER: { |
| 1775 Trace::DeferredSetRegister |
| 1776 new_set(data_.u_store_register.reg, data_.u_store_register.value); |
| 1777 Trace new_trace = *trace; |
| 1778 new_trace.add_action(&new_set); |
| 1779 on_success()->Emit(compiler, &new_trace); |
| 1780 break; |
| 1781 } |
| 1782 case CLEAR_CAPTURES: { |
| 1783 Trace::DeferredClearCaptures |
| 1784 new_capture(Interval(data_.u_clear_captures.range_from, |
| 1785 data_.u_clear_captures.range_to)); |
| 1786 Trace new_trace = *trace; |
| 1787 new_trace.add_action(&new_capture); |
| 1788 on_success()->Emit(compiler, &new_trace); |
| 1789 break; |
| 1790 } |
| 1791 case BEGIN_SUBMATCH: |
| 1792 if (!trace->is_trivial()) { |
| 1793 trace->Flush(compiler, this); |
| 1794 } else { |
| 1795 assembler->WriteCurrentPositionToRegister( |
| 1796 data_.u_submatch.current_position_register, 0); |
| 1797 assembler->WriteStackPointerToRegister( |
| 1798 data_.u_submatch.stack_pointer_register); |
| 1799 on_success()->Emit(compiler, trace); |
| 1800 } |
| 1801 break; |
| 1802 case EMPTY_MATCH_CHECK: { |
| 1803 intptr_t start_pos_reg = data_.u_empty_match_check.start_register; |
| 1804 intptr_t stored_pos = 0; |
| 1805 intptr_t rep_reg = data_.u_empty_match_check.repetition_register; |
| 1806 bool has_minimum = (rep_reg != RegExpCompiler::kNoRegister); |
| 1807 bool know_dist = trace->GetStoredPosition(start_pos_reg, &stored_pos); |
| 1808 if (know_dist && !has_minimum && stored_pos == trace->cp_offset()) { |
| 1809 // If we know we haven't advanced and there is no minimum we |
| 1810 // can just backtrack immediately. |
| 1811 assembler->GoTo(trace->backtrack()); |
| 1812 } else if (know_dist && stored_pos < trace->cp_offset()) { |
| 1813 // If we know we've advanced we can generate the continuation |
| 1814 // immediately. |
| 1815 on_success()->Emit(compiler, trace); |
| 1816 } else if (!trace->is_trivial()) { |
| 1817 trace->Flush(compiler, this); |
| 1818 } else { |
| 1819 BlockLabel skip_empty_check; |
| 1820 // If we have a minimum number of repetitions we check the current |
| 1821 // number first and skip the empty check if it's not enough. |
| 1822 if (has_minimum) { |
| 1823 intptr_t limit = data_.u_empty_match_check.repetition_limit; |
| 1824 assembler->IfRegisterLT(rep_reg, limit, &skip_empty_check); |
| 1825 } |
| 1826 // If the match is empty we bail out, otherwise we fall through |
| 1827 // to the on-success continuation. |
| 1828 assembler->IfRegisterEqPos(data_.u_empty_match_check.start_register, |
| 1829 trace->backtrack()); |
| 1830 assembler->BindBlock(&skip_empty_check); |
| 1831 on_success()->Emit(compiler, trace); |
| 1832 } |
| 1833 break; |
| 1834 } |
| 1835 case POSITIVE_SUBMATCH_SUCCESS: { |
| 1836 if (!trace->is_trivial()) { |
| 1837 trace->Flush(compiler, this); |
| 1838 return; |
| 1839 } |
| 1840 assembler->ReadCurrentPositionFromRegister( |
| 1841 data_.u_submatch.current_position_register); |
| 1842 assembler->ReadStackPointerFromRegister( |
| 1843 data_.u_submatch.stack_pointer_register); |
| 1844 intptr_t clear_register_count = data_.u_submatch.clear_register_count; |
| 1845 if (clear_register_count == 0) { |
| 1846 on_success()->Emit(compiler, trace); |
| 1847 return; |
| 1848 } |
| 1849 intptr_t clear_registers_from = data_.u_submatch.clear_register_from; |
| 1850 BlockLabel clear_registers_backtrack; |
| 1851 Trace new_trace = *trace; |
| 1852 new_trace.set_backtrack(&clear_registers_backtrack); |
| 1853 on_success()->Emit(compiler, &new_trace); |
| 1854 |
| 1855 assembler->BindBlock(&clear_registers_backtrack); |
| 1856 intptr_t clear_registers_to = |
| 1857 clear_registers_from + clear_register_count - 1; |
| 1858 assembler->ClearRegisters(clear_registers_from, clear_registers_to); |
| 1859 |
| 1860 ASSERT(trace->backtrack() == NULL); |
| 1861 assembler->Backtrack(); |
| 1862 return; |
| 1863 } |
| 1864 default: |
| 1865 UNREACHABLE(); |
| 1866 } |
| 1867 } |
| 1868 |
| 1869 |
| 1870 void BackReferenceNode::Emit(RegExpCompiler* compiler, Trace* trace) { |
| 1871 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 1872 if (!trace->is_trivial()) { |
| 1873 trace->Flush(compiler, this); |
| 1874 return; |
| 1875 } |
| 1876 |
| 1877 LimitResult limit_result = LimitVersions(compiler, trace); |
| 1878 if (limit_result == DONE) return; |
| 1879 ASSERT(limit_result == CONTINUE); |
| 1880 |
| 1881 RecursionCheck rc(compiler); |
| 1882 |
| 1883 ASSERT(start_reg_ + 1 == end_reg_); |
| 1884 if (compiler->ignore_case()) { |
| 1885 assembler->CheckNotBackReferenceIgnoreCase(start_reg_, |
| 1886 trace->backtrack()); |
| 1887 } else { |
| 1888 assembler->CheckNotBackReference(start_reg_, trace->backtrack()); |
| 1889 } |
| 1890 on_success()->Emit(compiler, trace); |
| 1891 } |
| 1892 |
| 1893 |
| 1894 void ActionNode::FillInBMInfo(intptr_t offset, |
| 1895 intptr_t budget, |
| 1896 BoyerMooreLookahead* bm, |
| 1897 bool not_at_start) { |
| 1898 if (action_type_ == BEGIN_SUBMATCH) { |
| 1899 bm->SetRest(offset); |
| 1900 } else if (action_type_ != POSITIVE_SUBMATCH_SUCCESS) { |
| 1901 on_success()->FillInBMInfo(offset, budget - 1, bm, not_at_start); |
| 1902 } |
| 1903 SaveBMInfo(bm, not_at_start, offset); |
| 1904 } |
| 1905 |
| 1906 |
| 1907 void AssertionNode::FillInBMInfo(intptr_t offset, |
| 1908 intptr_t budget, |
| 1909 BoyerMooreLookahead* bm, |
| 1910 bool not_at_start) { |
| 1911 // Match the behaviour of EatsAtLeast on this node. |
| 1912 if (assertion_type() == AT_START && not_at_start) return; |
| 1913 on_success()->FillInBMInfo(offset, budget - 1, bm, not_at_start); |
| 1914 SaveBMInfo(bm, not_at_start, offset); |
| 1915 } |
| 1916 |
| 1917 |
| 1918 void BackReferenceNode::FillInBMInfo(intptr_t offset, |
| 1919 intptr_t budget, |
| 1920 BoyerMooreLookahead* bm, |
| 1921 bool not_at_start) { |
| 1922 // Working out the set of characters that a backreference can match is too |
| 1923 // hard, so we just say that any character can match. |
| 1924 bm->SetRest(offset); |
| 1925 SaveBMInfo(bm, not_at_start, offset); |
| 1926 } |
| 1927 |
| 1928 |
| 1929 // Returns the number of characters in the equivalence class, omitting those |
| 1930 // that cannot occur in the source string because it is ASCII. |
| 1931 static intptr_t GetCaseIndependentLetters(uint16_t character, |
| 1932 bool ascii_subject, |
| 1933 int32_t* letters) { |
| 1934 unibrow::Mapping<unibrow::Ecma262UnCanonicalize> jsregexp_uncanonicalize; |
| 1935 intptr_t length = jsregexp_uncanonicalize.get(character, '\0', letters); |
| 1936 // Unibrow returns 0 or 1 for characters where case independence is |
| 1937 // trivial. |
| 1938 if (length == 0) { |
| 1939 letters[0] = character; |
| 1940 length = 1; |
| 1941 } |
| 1942 if (!ascii_subject || character <= Symbols::kMaxOneCharCodeSymbol) { |
| 1943 return length; |
| 1944 } |
| 1945 // The standard requires that non-ASCII characters cannot have ASCII |
| 1946 // character codes in their equivalence class. |
| 1947 return 0; |
| 1948 } |
| 1949 |
| 1950 |
| 1951 void ChoiceNode::FillInBMInfo(intptr_t offset, |
| 1952 intptr_t budget, |
| 1953 BoyerMooreLookahead* bm, |
| 1954 bool not_at_start) { |
| 1955 ZoneGrowableArray<GuardedAlternative>* alts = alternatives(); |
| 1956 budget = (budget - 1) / alts->length(); |
| 1957 for (intptr_t i = 0; i < alts->length(); i++) { |
| 1958 GuardedAlternative& alt = (*alts)[i]; |
| 1959 if (alt.guards() != NULL && alt.guards()->length() != 0) { |
| 1960 bm->SetRest(offset); // Give up trying to fill in info. |
| 1961 SaveBMInfo(bm, not_at_start, offset); |
| 1962 return; |
| 1963 } |
| 1964 alt.node()->FillInBMInfo(offset, budget, bm, not_at_start); |
| 1965 } |
| 1966 SaveBMInfo(bm, not_at_start, offset); |
| 1967 } |
| 1968 |
| 1969 |
| 1970 void EndNode::FillInBMInfo(intptr_t offset, |
| 1971 intptr_t budget, |
| 1972 BoyerMooreLookahead* bm, |
| 1973 bool not_at_start) { |
| 1974 // Returning 0 from EatsAtLeast should ensure we never get here. |
| 1975 UNREACHABLE(); |
| 1976 } |
| 1977 |
| 1978 |
| 1979 void LoopChoiceNode::FillInBMInfo(intptr_t offset, |
| 1980 intptr_t budget, |
| 1981 BoyerMooreLookahead* bm, |
| 1982 bool not_at_start) { |
| 1983 if (body_can_be_zero_length_ || budget <= 0) { |
| 1984 bm->SetRest(offset); |
| 1985 SaveBMInfo(bm, not_at_start, offset); |
| 1986 return; |
| 1987 } |
| 1988 ChoiceNode::FillInBMInfo(offset, budget - 1, bm, not_at_start); |
| 1989 SaveBMInfo(bm, not_at_start, offset); |
| 1990 } |
| 1991 |
| 1992 |
| 1993 void TextNode::FillInBMInfo(intptr_t initial_offset, |
| 1994 intptr_t budget, |
| 1995 BoyerMooreLookahead* bm, |
| 1996 bool not_at_start) { |
| 1997 if (initial_offset >= bm->length()) return; |
| 1998 intptr_t offset = initial_offset; |
| 1999 intptr_t max_char = bm->max_char(); |
| 2000 for (intptr_t i = 0; i < elements()->length(); i++) { |
| 2001 if (offset >= bm->length()) { |
| 2002 if (initial_offset == 0) set_bm_info(not_at_start, bm); |
| 2003 return; |
| 2004 } |
| 2005 TextElement text = elements()->At(i); |
| 2006 if (text.text_type() == TextElement::ATOM) { |
| 2007 RegExpAtom* atom = text.atom(); |
| 2008 for (intptr_t j = 0; j < atom->length(); j++, offset++) { |
| 2009 if (offset >= bm->length()) { |
| 2010 if (initial_offset == 0) set_bm_info(not_at_start, bm); |
| 2011 return; |
| 2012 } |
| 2013 uint16_t character = atom->data()->At(j); |
| 2014 if (bm->compiler()->ignore_case()) { |
| 2015 int32_t chars[unibrow::Ecma262UnCanonicalize::kMaxWidth]; |
| 2016 intptr_t length = GetCaseIndependentLetters( |
| 2017 character, |
| 2018 bm->max_char() == Symbols::kMaxOneCharCodeSymbol, |
| 2019 chars); |
| 2020 for (intptr_t j = 0; j < length; j++) { |
| 2021 bm->Set(offset, chars[j]); |
| 2022 } |
| 2023 } else { |
| 2024 if (character <= max_char) bm->Set(offset, character); |
| 2025 } |
| 2026 } |
| 2027 } else { |
| 2028 ASSERT(text.text_type() == TextElement::CHAR_CLASS); |
| 2029 RegExpCharacterClass* char_class = text.char_class(); |
| 2030 ZoneGrowableArray<CharacterRange>* ranges = char_class->ranges(); |
| 2031 if (char_class->is_negated()) { |
| 2032 bm->SetAll(offset); |
| 2033 } else { |
| 2034 for (intptr_t k = 0; k < ranges->length(); k++) { |
| 2035 CharacterRange& range = (*ranges)[k]; |
| 2036 if (range.from() > max_char) continue; |
| 2037 intptr_t to = Utils::Minimum(max_char, |
| 2038 static_cast<intptr_t>(range.to())); |
| 2039 bm->SetInterval(offset, Interval(range.from(), to)); |
| 2040 } |
| 2041 } |
| 2042 offset++; |
| 2043 } |
| 2044 } |
| 2045 if (offset >= bm->length()) { |
| 2046 if (initial_offset == 0) set_bm_info(not_at_start, bm); |
| 2047 return; |
| 2048 } |
| 2049 on_success()->FillInBMInfo(offset, |
| 2050 budget - 1, |
| 2051 bm, |
| 2052 true); // Not at start after a text node. |
| 2053 if (initial_offset == 0) set_bm_info(not_at_start, bm); |
| 2054 } |
| 2055 |
| 2056 |
| 2057 // Check for [0-9A-Z_a-z]. |
| 2058 static void EmitWordCheck(RegExpMacroAssembler* assembler, |
| 2059 BlockLabel* word, |
| 2060 BlockLabel* non_word, |
| 2061 bool fall_through_on_word) { |
| 2062 if (assembler->CheckSpecialCharacterClass( |
| 2063 fall_through_on_word ? 'w' : 'W', |
| 2064 fall_through_on_word ? non_word : word)) { |
| 2065 // Optimized implementation available. |
| 2066 return; |
| 2067 } |
| 2068 assembler->CheckCharacterGT('z', non_word); |
| 2069 assembler->CheckCharacterLT('0', non_word); |
| 2070 assembler->CheckCharacterGT('a' - 1, word); |
| 2071 assembler->CheckCharacterLT('9' + 1, word); |
| 2072 assembler->CheckCharacterLT('A', non_word); |
| 2073 assembler->CheckCharacterLT('Z' + 1, word); |
| 2074 if (fall_through_on_word) { |
| 2075 assembler->CheckNotCharacter('_', non_word); |
| 2076 } else { |
| 2077 assembler->CheckCharacter('_', word); |
| 2078 } |
| 2079 } |
| 2080 |
| 2081 |
| 2082 // Emit the code to handle \b and \B (word-boundary or non-word-boundary). |
| 2083 void AssertionNode::EmitBoundaryCheck(RegExpCompiler* compiler, Trace* trace) { |
| 2084 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 2085 Trace::TriBool next_is_word_character = Trace::UNKNOWN; |
| 2086 bool not_at_start = (trace->at_start() == Trace::FALSE_VALUE); |
| 2087 BoyerMooreLookahead* lookahead = bm_info(not_at_start); |
| 2088 if (lookahead == NULL) { |
| 2089 intptr_t eats_at_least = |
| 2090 Utils::Minimum(kMaxLookaheadForBoyerMoore, |
| 2091 EatsAtLeast(kMaxLookaheadForBoyerMoore, |
| 2092 kRecursionBudget, |
| 2093 not_at_start)); |
| 2094 if (eats_at_least >= 1) { |
| 2095 BoyerMooreLookahead* bm = |
| 2096 new(I) BoyerMooreLookahead(eats_at_least, compiler, I); |
| 2097 FillInBMInfo(0, kRecursionBudget, bm, not_at_start); |
| 2098 if (bm->at(0)->is_non_word()) |
| 2099 next_is_word_character = Trace::FALSE_VALUE; |
| 2100 if (bm->at(0)->is_word()) next_is_word_character = Trace::TRUE_VALUE; |
| 2101 } |
| 2102 } else { |
| 2103 if (lookahead->at(0)->is_non_word()) |
| 2104 next_is_word_character = Trace::FALSE_VALUE; |
| 2105 if (lookahead->at(0)->is_word()) |
| 2106 next_is_word_character = Trace::TRUE_VALUE; |
| 2107 } |
| 2108 bool at_boundary = (assertion_type_ == AssertionNode::AT_BOUNDARY); |
| 2109 if (next_is_word_character == Trace::UNKNOWN) { |
| 2110 BlockLabel before_non_word; |
| 2111 BlockLabel before_word; |
| 2112 if (trace->characters_preloaded() != 1) { |
| 2113 assembler->LoadCurrentCharacter(trace->cp_offset(), &before_non_word); |
| 2114 } |
| 2115 // Fall through on non-word. |
| 2116 EmitWordCheck(assembler, &before_word, &before_non_word, false); |
| 2117 // Next character is not a word character. |
| 2118 assembler->BindBlock(&before_non_word); |
| 2119 BlockLabel ok; |
| 2120 // Backtrack on \B (non-boundary check) if previous is a word, |
| 2121 // since we know next *is not* a word and this would be a boundary. |
| 2122 BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord); |
| 2123 |
| 2124 if (!assembler->IsClosed()) { |
| 2125 assembler->GoTo(&ok); |
| 2126 } |
| 2127 |
| 2128 assembler->BindBlock(&before_word); |
| 2129 BacktrackIfPrevious(compiler, trace, at_boundary ? kIsWord : kIsNonWord); |
| 2130 assembler->BindBlock(&ok); |
| 2131 } else if (next_is_word_character == Trace::TRUE_VALUE) { |
| 2132 BacktrackIfPrevious(compiler, trace, at_boundary ? kIsWord : kIsNonWord); |
| 2133 } else { |
| 2134 ASSERT(next_is_word_character == Trace::FALSE_VALUE); |
| 2135 BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord); |
| 2136 } |
| 2137 } |
| 2138 |
| 2139 |
| 2140 void AssertionNode::BacktrackIfPrevious( |
| 2141 RegExpCompiler* compiler, |
| 2142 Trace* trace, |
| 2143 AssertionNode::IfPrevious backtrack_if_previous) { |
| 2144 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 2145 Trace new_trace(*trace); |
| 2146 new_trace.InvalidateCurrentCharacter(); |
| 2147 |
| 2148 BlockLabel fall_through, dummy; |
| 2149 |
| 2150 BlockLabel* non_word = backtrack_if_previous == kIsNonWord ? |
| 2151 new_trace.backtrack() : |
| 2152 &fall_through; |
| 2153 BlockLabel* word = backtrack_if_previous == kIsNonWord ? |
| 2154 &fall_through : |
| 2155 new_trace.backtrack(); |
| 2156 |
| 2157 if (new_trace.cp_offset() == 0) { |
| 2158 // The start of input counts as a non-word character, so the question is |
| 2159 // decided if we are at the start. |
| 2160 assembler->CheckAtStart(non_word); |
| 2161 } |
| 2162 // We already checked that we are not at the start of input so it must be |
| 2163 // OK to load the previous character. |
| 2164 assembler->LoadCurrentCharacter(new_trace.cp_offset() - 1, &dummy, false); |
| 2165 EmitWordCheck(assembler, word, non_word, backtrack_if_previous == kIsNonWord); |
| 2166 |
| 2167 assembler->BindBlock(&fall_through); |
| 2168 on_success()->Emit(compiler, &new_trace); |
| 2169 } |
| 2170 |
| 2171 |
| 2172 static bool DeterminedAlready(QuickCheckDetails* quick_check, intptr_t offset) { |
| 2173 if (quick_check == NULL) return false; |
| 2174 if (offset >= quick_check->characters()) return false; |
| 2175 return quick_check->positions(offset)->determines_perfectly; |
| 2176 } |
| 2177 |
| 2178 |
| 2179 static void UpdateBoundsCheck(intptr_t index, intptr_t* checked_up_to) { |
| 2180 if (index > *checked_up_to) { |
| 2181 *checked_up_to = index; |
| 2182 } |
| 2183 } |
| 2184 |
| 2185 |
| 2186 typedef bool EmitCharacterFunction(Isolate* isolate, |
| 2187 RegExpCompiler* compiler, |
| 2188 uint16_t c, |
| 2189 BlockLabel* on_failure, |
| 2190 intptr_t cp_offset, |
| 2191 bool check, |
| 2192 bool preloaded); |
| 2193 |
| 2194 |
| 2195 static inline bool EmitSimpleCharacter(Isolate* isolate, |
| 2196 RegExpCompiler* compiler, |
| 2197 uint16_t c, |
| 2198 BlockLabel* on_failure, |
| 2199 intptr_t cp_offset, |
| 2200 bool check, |
| 2201 bool preloaded) { |
| 2202 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 2203 bool bound_checked = false; |
| 2204 if (!preloaded) { |
| 2205 assembler->LoadCurrentCharacter( |
| 2206 cp_offset, |
| 2207 on_failure, |
| 2208 check); |
| 2209 bound_checked = true; |
| 2210 } |
| 2211 assembler->CheckNotCharacter(c, on_failure); |
| 2212 return bound_checked; |
| 2213 } |
| 2214 |
| 2215 |
| 2216 static bool ShortCutEmitCharacterPair(RegExpMacroAssembler* macro_assembler, |
| 2217 bool ascii, |
| 2218 uint16_t c1, |
| 2219 uint16_t c2, |
| 2220 BlockLabel* on_failure) { |
| 2221 uint16_t char_mask; |
| 2222 if (ascii) { |
| 2223 char_mask = Symbols::kMaxOneCharCodeSymbol; |
| 2224 } else { |
| 2225 char_mask = Utf16::kMaxCodeUnit; |
| 2226 } |
| 2227 uint16_t exor = c1 ^ c2; |
| 2228 // Check whether exor has only one bit set. |
| 2229 if (((exor - 1) & exor) == 0) { |
| 2230 // If c1 and c2 differ only by one bit. |
| 2231 // Ecma262UnCanonicalize always gives the highest number last. |
| 2232 ASSERT(c2 > c1); |
| 2233 uint16_t mask = char_mask ^ exor; |
| 2234 macro_assembler->CheckNotCharacterAfterAnd(c1, mask, on_failure); |
| 2235 return true; |
| 2236 } |
| 2237 ASSERT(c2 > c1); |
| 2238 uint16_t diff = c2 - c1; |
| 2239 if (((diff - 1) & diff) == 0 && c1 >= diff) { |
| 2240 // If the characters differ by 2^n but don't differ by one bit then |
| 2241 // subtract the difference from the found character, then do the or |
| 2242 // trick. We avoid the theoretical case where negative numbers are |
| 2243 // involved in order to simplify code generation. |
| 2244 uint16_t mask = char_mask ^ diff; |
| 2245 macro_assembler->CheckNotCharacterAfterMinusAnd(c1 - diff, |
| 2246 diff, |
| 2247 mask, |
| 2248 on_failure); |
| 2249 return true; |
| 2250 } |
| 2251 return false; |
| 2252 } |
| 2253 |
| 2254 // Only emits letters (things that have case). Only used for case independent |
| 2255 // matches. |
| 2256 static inline bool EmitAtomLetter(Isolate* isolate, |
| 2257 RegExpCompiler* compiler, |
| 2258 uint16_t c, |
| 2259 BlockLabel* on_failure, |
| 2260 intptr_t cp_offset, |
| 2261 bool check, |
| 2262 bool preloaded) { |
| 2263 RegExpMacroAssembler* macro_assembler = compiler->macro_assembler(); |
| 2264 bool ascii = compiler->ascii(); |
| 2265 int32_t chars[unibrow::Ecma262UnCanonicalize::kMaxWidth]; |
| 2266 intptr_t length = GetCaseIndependentLetters(c, ascii, chars); |
| 2267 if (length <= 1) return false; |
| 2268 // We may not need to check against the end of the input string |
| 2269 // if this character lies before a character that matched. |
| 2270 if (!preloaded) { |
| 2271 macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check); |
| 2272 } |
| 2273 BlockLabel ok; |
| 2274 ASSERT(unibrow::Ecma262UnCanonicalize::kMaxWidth == 4); |
| 2275 switch (length) { |
| 2276 case 2: { |
| 2277 if (ShortCutEmitCharacterPair(macro_assembler, |
| 2278 ascii, |
| 2279 chars[0], |
| 2280 chars[1], |
| 2281 on_failure)) { |
| 2282 } else { |
| 2283 macro_assembler->CheckCharacter(chars[0], &ok); |
| 2284 macro_assembler->CheckNotCharacter(chars[1], on_failure); |
| 2285 macro_assembler->BindBlock(&ok); |
| 2286 } |
| 2287 break; |
| 2288 } |
| 2289 case 4: |
| 2290 macro_assembler->CheckCharacter(chars[3], &ok); |
| 2291 // Fall through! |
| 2292 case 3: |
| 2293 macro_assembler->CheckCharacter(chars[0], &ok); |
| 2294 macro_assembler->CheckCharacter(chars[1], &ok); |
| 2295 macro_assembler->CheckNotCharacter(chars[2], on_failure); |
| 2296 macro_assembler->BindBlock(&ok); |
| 2297 break; |
| 2298 default: |
| 2299 UNREACHABLE(); |
| 2300 break; |
| 2301 } |
| 2302 return true; |
| 2303 } |
| 2304 |
| 2305 |
| 2306 // Only emits non-letters (things that don't have case). Only used for case |
| 2307 // independent matches. |
| 2308 static inline bool EmitAtomNonLetter(Isolate* isolate, |
| 2309 RegExpCompiler* compiler, |
| 2310 uint16_t c, |
| 2311 BlockLabel* on_failure, |
| 2312 intptr_t cp_offset, |
| 2313 bool check, |
| 2314 bool preloaded) { |
| 2315 RegExpMacroAssembler* macro_assembler = compiler->macro_assembler(); |
| 2316 bool ascii = compiler->ascii(); |
| 2317 int32_t chars[unibrow::Ecma262UnCanonicalize::kMaxWidth]; |
| 2318 intptr_t length = GetCaseIndependentLetters(c, ascii, chars); |
| 2319 if (length < 1) { |
| 2320 // This can't match. Must be an ASCII subject and a non-ASCII character. |
| 2321 // We do not need to do anything since the ASCII pass already handled this. |
| 2322 return false; // Bounds not checked. |
| 2323 } |
| 2324 bool checked = false; |
| 2325 // We handle the length > 1 case in a later pass. |
| 2326 if (length == 1) { |
| 2327 if (ascii && c > Symbols::kMaxOneCharCodeSymbol) { |
| 2328 // Can't match - see above. |
| 2329 return false; // Bounds not checked. |
| 2330 } |
| 2331 if (!preloaded) { |
| 2332 macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check); |
| 2333 checked = check; |
| 2334 } |
| 2335 macro_assembler->CheckNotCharacter(c, on_failure); |
| 2336 } |
| 2337 return checked; |
| 2338 } |
| 2339 |
| 2340 |
| 2341 static void EmitBoundaryTest(RegExpMacroAssembler* masm, |
| 2342 intptr_t border, |
| 2343 BlockLabel* fall_through, |
| 2344 BlockLabel* above_or_equal, |
| 2345 BlockLabel* below) { |
| 2346 if (below != fall_through) { |
| 2347 masm->CheckCharacterLT(border, below); |
| 2348 if (above_or_equal != fall_through) masm->GoTo(above_or_equal); |
| 2349 } else { |
| 2350 masm->CheckCharacterGT(border - 1, above_or_equal); |
| 2351 } |
| 2352 } |
| 2353 |
| 2354 |
| 2355 static void EmitDoubleBoundaryTest(RegExpMacroAssembler* masm, |
| 2356 intptr_t first, |
| 2357 intptr_t last, |
| 2358 BlockLabel* fall_through, |
| 2359 BlockLabel* in_range, |
| 2360 BlockLabel* out_of_range) { |
| 2361 if (in_range == fall_through) { |
| 2362 if (first == last) { |
| 2363 masm->CheckNotCharacter(first, out_of_range); |
| 2364 } else { |
| 2365 masm->CheckCharacterNotInRange(first, last, out_of_range); |
| 2366 } |
| 2367 } else { |
| 2368 if (first == last) { |
| 2369 masm->CheckCharacter(first, in_range); |
| 2370 } else { |
| 2371 masm->CheckCharacterInRange(first, last, in_range); |
| 2372 } |
| 2373 if (out_of_range != fall_through) masm->GoTo(out_of_range); |
| 2374 } |
| 2375 } |
| 2376 |
| 2377 |
| 2378 static void CutOutRange(RegExpMacroAssembler* masm, |
| 2379 ZoneGrowableArray<int>* ranges, |
| 2380 intptr_t start_index, |
| 2381 intptr_t end_index, |
| 2382 intptr_t cut_index, |
| 2383 BlockLabel* even_label, |
| 2384 BlockLabel* odd_label) { |
| 2385 bool odd = (((cut_index - start_index) & 1) == 1); |
| 2386 BlockLabel* in_range_label = odd ? odd_label : even_label; |
| 2387 BlockLabel dummy; |
| 2388 EmitDoubleBoundaryTest(masm, |
| 2389 ranges->At(cut_index), |
| 2390 ranges->At(cut_index + 1) - 1, |
| 2391 &dummy, |
| 2392 in_range_label, |
| 2393 &dummy); |
| 2394 ASSERT(!dummy.IsLinked()); |
| 2395 // Cut out the single range by rewriting the array. This creates a new |
| 2396 // range that is a merger of the two ranges on either side of the one we |
| 2397 // are cutting out. The oddity of the labels is preserved. |
| 2398 for (intptr_t j = cut_index; j > start_index; j--) { |
| 2399 (*ranges)[j] = ranges->At(j - 1); |
| 2400 } |
| 2401 for (intptr_t j = cut_index + 1; j < end_index; j++) { |
| 2402 (*ranges)[j] = ranges->At(j + 1); |
| 2403 } |
| 2404 } |
| 2405 |
| 2406 |
| 2407 // even_label is for ranges[i] to ranges[i + 1] where i - start_index is even. |
| 2408 // odd_label is for ranges[i] to ranges[i + 1] where i - start_index is odd. |
| 2409 static void EmitUseLookupTable( |
| 2410 RegExpMacroAssembler* masm, |
| 2411 ZoneGrowableArray<int>* ranges, |
| 2412 intptr_t start_index, |
| 2413 intptr_t end_index, |
| 2414 intptr_t min_char, |
| 2415 BlockLabel* fall_through, |
| 2416 BlockLabel* even_label, |
| 2417 BlockLabel* odd_label) { |
| 2418 static const intptr_t kSize = RegExpMacroAssembler::kTableSize; |
| 2419 static const intptr_t kMask = RegExpMacroAssembler::kTableMask; |
| 2420 |
| 2421 intptr_t base = (min_char & ~kMask); |
| 2422 |
| 2423 // Assert that everything is on one kTableSize page. |
| 2424 for (intptr_t i = start_index; i <= end_index; i++) { |
| 2425 ASSERT((ranges->At(i) & ~kMask) == base); |
| 2426 } |
| 2427 ASSERT(start_index == 0 || (ranges->At(start_index - 1) & ~kMask) <= base); |
| 2428 |
| 2429 char templ[kSize]; |
| 2430 BlockLabel* on_bit_set; |
| 2431 BlockLabel* on_bit_clear; |
| 2432 intptr_t bit; |
| 2433 if (even_label == fall_through) { |
| 2434 on_bit_set = odd_label; |
| 2435 on_bit_clear = even_label; |
| 2436 bit = 1; |
| 2437 } else { |
| 2438 on_bit_set = even_label; |
| 2439 on_bit_clear = odd_label; |
| 2440 bit = 0; |
| 2441 } |
| 2442 for (intptr_t i = 0; i < (ranges->At(start_index) & kMask) && i < kSize; |
| 2443 i++) { |
| 2444 templ[i] = bit; |
| 2445 } |
| 2446 intptr_t j = 0; |
| 2447 bit ^= 1; |
| 2448 for (intptr_t i = start_index; i < end_index; i++) { |
| 2449 for (j = (ranges->At(i) & kMask); j < (ranges->At(i + 1) & kMask); j++) { |
| 2450 templ[j] = bit; |
| 2451 } |
| 2452 bit ^= 1; |
| 2453 } |
| 2454 for (intptr_t i = j; i < kSize; i++) { |
| 2455 templ[i] = bit; |
| 2456 } |
| 2457 // TODO(erikcorry): Cache these. |
| 2458 const TypedData& ba = TypedData::ZoneHandle( |
| 2459 masm->isolate(), |
| 2460 TypedData::New(kTypedDataUint8ArrayCid, kSize, Heap::kOld)); |
| 2461 for (intptr_t i = 0; i < kSize; i++) { |
| 2462 ba.SetUint8(i, templ[i]); |
| 2463 } |
| 2464 masm->CheckBitInTable(ba, on_bit_set); |
| 2465 if (on_bit_clear != fall_through) masm->GoTo(on_bit_clear); |
| 2466 } |
| 2467 |
| 2468 |
| 2469 // Unicode case. Split the search space into kSize spaces that are handled |
| 2470 // with recursion. |
| 2471 static void SplitSearchSpace(ZoneGrowableArray<int>* ranges, |
| 2472 intptr_t start_index, |
| 2473 intptr_t end_index, |
| 2474 intptr_t* new_start_index, |
| 2475 intptr_t* new_end_index, |
| 2476 intptr_t* border) { |
| 2477 static const intptr_t kSize = RegExpMacroAssembler::kTableSize; |
| 2478 static const intptr_t kMask = RegExpMacroAssembler::kTableMask; |
| 2479 |
| 2480 intptr_t first = ranges->At(start_index); |
| 2481 intptr_t last = ranges->At(end_index) - 1; |
| 2482 |
| 2483 *new_start_index = start_index; |
| 2484 *border = (ranges->At(start_index) & ~kMask) + kSize; |
| 2485 while (*new_start_index < end_index) { |
| 2486 if (ranges->At(*new_start_index) > *border) break; |
| 2487 (*new_start_index)++; |
| 2488 } |
| 2489 // new_start_index is the index of the first edge that is beyond the |
| 2490 // current kSize space. |
| 2491 |
| 2492 // For very large search spaces we do a binary chop search of the non-ASCII |
| 2493 // space instead of just going to the end of the current kSize space. The |
| 2494 // heuristics are complicated a little by the fact that any 128-character |
| 2495 // encoding space can be quickly tested with a table lookup, so we don't |
| 2496 // wish to do binary chop search at a smaller granularity than that. A |
| 2497 // 128-character space can take up a lot of space in the ranges array if, |
| 2498 // for example, we only want to match every second character (eg. the lower |
| 2499 // case characters on some Unicode pages). |
| 2500 intptr_t binary_chop_index = (end_index + start_index) / 2; |
| 2501 // The first test ensures that we get to the code that handles the ASCII |
| 2502 // range with a single not-taken branch, speeding up this important |
| 2503 // character range (even non-ASCII charset-based text has spaces and |
| 2504 // punctuation). |
| 2505 if (*border - 1 > Symbols::kMaxOneCharCodeSymbol && // ASCII case. |
| 2506 end_index - start_index > (*new_start_index - start_index) * 2 && |
| 2507 last - first > kSize * 2 && |
| 2508 binary_chop_index > *new_start_index && |
| 2509 ranges->At(binary_chop_index) >= first + 2 * kSize) { |
| 2510 intptr_t scan_forward_for_section_border = binary_chop_index;; |
| 2511 intptr_t new_border = (ranges->At(binary_chop_index) | kMask) + 1; |
| 2512 |
| 2513 while (scan_forward_for_section_border < end_index) { |
| 2514 if (ranges->At(scan_forward_for_section_border) > new_border) { |
| 2515 *new_start_index = scan_forward_for_section_border; |
| 2516 *border = new_border; |
| 2517 break; |
| 2518 } |
| 2519 scan_forward_for_section_border++; |
| 2520 } |
| 2521 } |
| 2522 |
| 2523 ASSERT(*new_start_index > start_index); |
| 2524 *new_end_index = *new_start_index - 1; |
| 2525 if (ranges->At(*new_end_index) == *border) { |
| 2526 (*new_end_index)--; |
| 2527 } |
| 2528 if (*border >= ranges->At(end_index)) { |
| 2529 *border = ranges->At(end_index); |
| 2530 *new_start_index = end_index; // Won't be used. |
| 2531 *new_end_index = end_index - 1; |
| 2532 } |
| 2533 } |
| 2534 |
| 2535 |
| 2536 // Gets a series of segment boundaries representing a character class. If the |
| 2537 // character is in the range between an even and an odd boundary (counting from |
| 2538 // start_index) then go to even_label, otherwise go to odd_label. We already |
| 2539 // know that the character is in the range of min_char to max_char inclusive. |
| 2540 // Either label can be NULL indicating backtracking. Either label can also be |
| 2541 // equal to the fall_through label. |
| 2542 static void GenerateBranches(RegExpMacroAssembler* masm, |
| 2543 ZoneGrowableArray<int>* ranges, |
| 2544 intptr_t start_index, |
| 2545 intptr_t end_index, |
| 2546 uint16_t min_char, |
| 2547 uint16_t max_char, |
| 2548 BlockLabel* fall_through, |
| 2549 BlockLabel* even_label, |
| 2550 BlockLabel* odd_label) { |
| 2551 intptr_t first = ranges->At(start_index); |
| 2552 intptr_t last = ranges->At(end_index) - 1; |
| 2553 |
| 2554 ASSERT(min_char < first); |
| 2555 |
| 2556 // Just need to test if the character is before or on-or-after |
| 2557 // a particular character. |
| 2558 if (start_index == end_index) { |
| 2559 EmitBoundaryTest(masm, first, fall_through, even_label, odd_label); |
| 2560 return; |
| 2561 } |
| 2562 |
| 2563 // Another almost trivial case: There is one interval in the middle that is |
| 2564 // different from the end intervals. |
| 2565 if (start_index + 1 == end_index) { |
| 2566 EmitDoubleBoundaryTest( |
| 2567 masm, first, last, fall_through, even_label, odd_label); |
| 2568 return; |
| 2569 } |
| 2570 |
| 2571 // It's not worth using table lookup if there are very few intervals in the |
| 2572 // character class. |
| 2573 if (end_index - start_index <= 6) { |
| 2574 // It is faster to test for individual characters, so we look for those |
| 2575 // first, then try arbitrary ranges in the second round. |
| 2576 static intptr_t kNoCutIndex = -1; |
| 2577 intptr_t cut = kNoCutIndex; |
| 2578 for (intptr_t i = start_index; i < end_index; i++) { |
| 2579 if (ranges->At(i) == ranges->At(i + 1) - 1) { |
| 2580 cut = i; |
| 2581 break; |
| 2582 } |
| 2583 } |
| 2584 if (cut == kNoCutIndex) cut = start_index; |
| 2585 CutOutRange( |
| 2586 masm, ranges, start_index, end_index, cut, even_label, odd_label); |
| 2587 ASSERT(end_index - start_index >= 2); |
| 2588 GenerateBranches(masm, |
| 2589 ranges, |
| 2590 start_index + 1, |
| 2591 end_index - 1, |
| 2592 min_char, |
| 2593 max_char, |
| 2594 fall_through, |
| 2595 even_label, |
| 2596 odd_label); |
| 2597 return; |
| 2598 } |
| 2599 |
| 2600 // If there are a lot of intervals in the regexp, then we will use tables to |
| 2601 // determine whether the character is inside or outside the character class. |
| 2602 static const intptr_t kBits = RegExpMacroAssembler::kTableSizeBits; |
| 2603 |
| 2604 if ((max_char >> kBits) == (min_char >> kBits)) { |
| 2605 EmitUseLookupTable(masm, |
| 2606 ranges, |
| 2607 start_index, |
| 2608 end_index, |
| 2609 min_char, |
| 2610 fall_through, |
| 2611 even_label, |
| 2612 odd_label); |
| 2613 return; |
| 2614 } |
| 2615 |
| 2616 if ((min_char >> kBits) != (first >> kBits)) { |
| 2617 masm->CheckCharacterLT(first, odd_label); |
| 2618 GenerateBranches(masm, |
| 2619 ranges, |
| 2620 start_index + 1, |
| 2621 end_index, |
| 2622 first, |
| 2623 max_char, |
| 2624 fall_through, |
| 2625 odd_label, |
| 2626 even_label); |
| 2627 return; |
| 2628 } |
| 2629 |
| 2630 intptr_t new_start_index = 0; |
| 2631 intptr_t new_end_index = 0; |
| 2632 intptr_t border = 0; |
| 2633 |
| 2634 SplitSearchSpace(ranges, |
| 2635 start_index, |
| 2636 end_index, |
| 2637 &new_start_index, |
| 2638 &new_end_index, |
| 2639 &border); |
| 2640 |
| 2641 BlockLabel handle_rest; |
| 2642 BlockLabel* above = &handle_rest; |
| 2643 if (border == last + 1) { |
| 2644 // We didn't find any section that started after the limit, so everything |
| 2645 // above the border is one of the terminal labels. |
| 2646 above = (end_index & 1) != (start_index & 1) ? odd_label : even_label; |
| 2647 ASSERT(new_end_index == end_index - 1); |
| 2648 } |
| 2649 |
| 2650 ASSERT(start_index <= new_end_index); |
| 2651 ASSERT(new_start_index <= end_index); |
| 2652 ASSERT(start_index < new_start_index); |
| 2653 ASSERT(new_end_index < end_index); |
| 2654 ASSERT(new_end_index + 1 == new_start_index || |
| 2655 (new_end_index + 2 == new_start_index && |
| 2656 border == ranges->At(new_end_index + 1))); |
| 2657 ASSERT(min_char < border - 1); |
| 2658 ASSERT(border < max_char); |
| 2659 ASSERT(ranges->At(new_end_index) < border); |
| 2660 ASSERT(border < ranges->At(new_start_index) || |
| 2661 (border == ranges->At(new_start_index) && |
| 2662 new_start_index == end_index && |
| 2663 new_end_index == end_index - 1 && |
| 2664 border == last + 1)); |
| 2665 ASSERT(new_start_index == 0 || border >= ranges->At(new_start_index - 1)); |
| 2666 |
| 2667 masm->CheckCharacterGT(border - 1, above); |
| 2668 BlockLabel dummy; |
| 2669 GenerateBranches(masm, |
| 2670 ranges, |
| 2671 start_index, |
| 2672 new_end_index, |
| 2673 min_char, |
| 2674 border - 1, |
| 2675 &dummy, |
| 2676 even_label, |
| 2677 odd_label); |
| 2678 |
| 2679 if (handle_rest.IsLinked()) { |
| 2680 masm->BindBlock(&handle_rest); |
| 2681 bool flip = (new_start_index & 1) != (start_index & 1); |
| 2682 GenerateBranches(masm, |
| 2683 ranges, |
| 2684 new_start_index, |
| 2685 end_index, |
| 2686 border, |
| 2687 max_char, |
| 2688 &dummy, |
| 2689 flip ? odd_label : even_label, |
| 2690 flip ? even_label : odd_label); |
| 2691 } |
| 2692 } |
| 2693 |
| 2694 |
| 2695 static void EmitCharClass(RegExpMacroAssembler* macro_assembler, |
| 2696 RegExpCharacterClass* cc, |
| 2697 bool ascii, |
| 2698 BlockLabel* on_failure, |
| 2699 intptr_t cp_offset, |
| 2700 bool check_offset, |
| 2701 bool preloaded, |
| 2702 Isolate* isolate) { |
| 2703 ZoneGrowableArray<CharacterRange>* ranges = cc->ranges(); |
| 2704 if (!CharacterRange::IsCanonical(ranges)) { |
| 2705 CharacterRange::Canonicalize(ranges); |
| 2706 } |
| 2707 |
| 2708 intptr_t max_char; |
| 2709 if (ascii) { |
| 2710 max_char = Symbols::kMaxOneCharCodeSymbol; |
| 2711 } else { |
| 2712 max_char = Utf16::kMaxCodeUnit; |
| 2713 } |
| 2714 |
| 2715 intptr_t range_count = ranges->length(); |
| 2716 |
| 2717 intptr_t last_valid_range = range_count - 1; |
| 2718 while (last_valid_range >= 0) { |
| 2719 CharacterRange& range = (*ranges)[last_valid_range]; |
| 2720 if (range.from() <= max_char) { |
| 2721 break; |
| 2722 } |
| 2723 last_valid_range--; |
| 2724 } |
| 2725 |
| 2726 if (last_valid_range < 0) { |
| 2727 if (!cc->is_negated()) { |
| 2728 macro_assembler->GoTo(on_failure); |
| 2729 } |
| 2730 if (check_offset) { |
| 2731 macro_assembler->CheckPosition(cp_offset, on_failure); |
| 2732 } |
| 2733 return; |
| 2734 } |
| 2735 |
| 2736 if (last_valid_range == 0 && |
| 2737 ranges->At(0).IsEverything(max_char)) { |
| 2738 if (cc->is_negated()) { |
| 2739 macro_assembler->GoTo(on_failure); |
| 2740 } else { |
| 2741 // This is a common case hit by non-anchored expressions. |
| 2742 if (check_offset) { |
| 2743 macro_assembler->CheckPosition(cp_offset, on_failure); |
| 2744 } |
| 2745 } |
| 2746 return; |
| 2747 } |
| 2748 if (last_valid_range == 0 && |
| 2749 !cc->is_negated() && |
| 2750 ranges->At(0).IsEverything(max_char)) { |
| 2751 // This is a common case hit by non-anchored expressions. |
| 2752 if (check_offset) { |
| 2753 macro_assembler->CheckPosition(cp_offset, on_failure); |
| 2754 } |
| 2755 return; |
| 2756 } |
| 2757 |
| 2758 if (!preloaded) { |
| 2759 macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check_offset); |
| 2760 } |
| 2761 |
| 2762 if (cc->is_standard() && |
| 2763 macro_assembler->CheckSpecialCharacterClass(cc->standard_type(), |
| 2764 on_failure)) { |
| 2765 return; |
| 2766 } |
| 2767 |
| 2768 |
| 2769 // A new list with ascending entries. Each entry is a code unit |
| 2770 // where there is a boundary between code units that are part of |
| 2771 // the class and code units that are not. Normally we insert an |
| 2772 // entry at zero which goes to the failure label, but if there |
| 2773 // was already one there we fall through for success on that entry. |
| 2774 // Subsequent entries have alternating meaning (success/failure). |
| 2775 ZoneGrowableArray<int>* range_boundaries = |
| 2776 new(isolate) ZoneGrowableArray<int>(last_valid_range); |
| 2777 |
| 2778 bool zeroth_entry_is_failure = !cc->is_negated(); |
| 2779 |
| 2780 for (intptr_t i = 0; i <= last_valid_range; i++) { |
| 2781 CharacterRange& range = (*ranges)[i]; |
| 2782 if (range.from() == 0) { |
| 2783 ASSERT(i == 0); |
| 2784 zeroth_entry_is_failure = !zeroth_entry_is_failure; |
| 2785 } else { |
| 2786 range_boundaries->Add(range.from()); |
| 2787 } |
| 2788 range_boundaries->Add(range.to() + 1); |
| 2789 } |
| 2790 intptr_t end_index = range_boundaries->length() - 1; |
| 2791 if (range_boundaries->At(end_index) > max_char) { |
| 2792 end_index--; |
| 2793 } |
| 2794 |
| 2795 BlockLabel fall_through; |
| 2796 GenerateBranches(macro_assembler, |
| 2797 range_boundaries, |
| 2798 0, // start_index. |
| 2799 end_index, |
| 2800 0, // min_char. |
| 2801 max_char, |
| 2802 &fall_through, |
| 2803 zeroth_entry_is_failure ? &fall_through : on_failure, |
| 2804 zeroth_entry_is_failure ? on_failure : &fall_through); |
| 2805 macro_assembler->BindBlock(&fall_through); |
| 2806 } |
| 2807 |
| 2808 |
| 2809 // We call this repeatedly to generate code for each pass over the text node. |
| 2810 // The passes are in increasing order of difficulty because we hope one |
| 2811 // of the first passes will fail in which case we are saved the work of the |
| 2812 // later passes. for example for the case independent regexp /%[asdfghjkl]a/ |
| 2813 // we will check the '%' in the first pass, the case independent 'a' in the |
| 2814 // second pass and the character class in the last pass. |
| 2815 // |
| 2816 // The passes are done from right to left, so for example to test for /bar/ |
| 2817 // we will first test for an 'r' with offset 2, then an 'a' with offset 1 |
| 2818 // and then a 'b' with offset 0. This means we can avoid the end-of-input |
| 2819 // bounds check most of the time. In the example we only need to check for |
| 2820 // end-of-input when loading the putative 'r'. |
| 2821 // |
| 2822 // A slight complication involves the fact that the first character may already |
| 2823 // be fetched into a register by the previous node. In this case we want to |
| 2824 // do the test for that character first. We do this in separate passes. The |
| 2825 // 'preloaded' argument indicates that we are doing such a 'pass'. If such a |
| 2826 // pass has been performed then subsequent passes will have true in |
| 2827 // first_element_checked to indicate that that character does not need to be |
| 2828 // checked again. |
| 2829 // |
| 2830 // In addition to all this we are passed a Trace, which can |
| 2831 // contain an AlternativeGeneration object. In this AlternativeGeneration |
| 2832 // object we can see details of any quick check that was already passed in |
| 2833 // order to get to the code we are now generating. The quick check can involve |
| 2834 // loading characters, which means we do not need to recheck the bounds |
| 2835 // up to the limit the quick check already checked. In addition the quick |
| 2836 // check can have involved a mask and compare operation which may simplify |
| 2837 // or obviate the need for further checks at some character positions. |
| 2838 void TextNode::TextEmitPass(RegExpCompiler* compiler, |
| 2839 TextEmitPassType pass, |
| 2840 bool preloaded, |
| 2841 Trace* trace, |
| 2842 bool first_element_checked, |
| 2843 intptr_t* checked_up_to) { |
| 2844 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 2845 bool ascii = compiler->ascii(); |
| 2846 BlockLabel* backtrack = trace->backtrack(); |
| 2847 QuickCheckDetails* quick_check = trace->quick_check_performed(); |
| 2848 intptr_t element_count = elms_->length(); |
| 2849 for (intptr_t i = preloaded ? 0 : element_count - 1; i >= 0; i--) { |
| 2850 TextElement elm = elms_->At(i); |
| 2851 intptr_t cp_offset = trace->cp_offset() + elm.cp_offset(); |
| 2852 if (elm.text_type() == TextElement::ATOM) { |
| 2853 ZoneGrowableArray<uint16_t>* quarks = elm.atom()->data(); |
| 2854 for (intptr_t j = preloaded ? 0 : quarks->length() - 1; j >= 0; j--) { |
| 2855 if (first_element_checked && i == 0 && j == 0) continue; |
| 2856 if (DeterminedAlready(quick_check, elm.cp_offset() + j)) continue; |
| 2857 EmitCharacterFunction* emit_function = NULL; |
| 2858 switch (pass) { |
| 2859 case NON_ASCII_MATCH: |
| 2860 ASSERT(ascii); |
| 2861 if (quarks->At(j) > Symbols::kMaxOneCharCodeSymbol) { |
| 2862 assembler->GoTo(backtrack); |
| 2863 return; |
| 2864 } |
| 2865 break; |
| 2866 case NON_LETTER_CHARACTER_MATCH: |
| 2867 emit_function = &EmitAtomNonLetter; |
| 2868 break; |
| 2869 case SIMPLE_CHARACTER_MATCH: |
| 2870 emit_function = &EmitSimpleCharacter; |
| 2871 break; |
| 2872 case CASE_CHARACTER_MATCH: |
| 2873 emit_function = &EmitAtomLetter; |
| 2874 break; |
| 2875 default: |
| 2876 break; |
| 2877 } |
| 2878 if (emit_function != NULL) { |
| 2879 bool bound_checked = emit_function(I, |
| 2880 compiler, |
| 2881 quarks->At(j), |
| 2882 backtrack, |
| 2883 cp_offset + j, |
| 2884 *checked_up_to < cp_offset + j, |
| 2885 preloaded); |
| 2886 if (bound_checked) UpdateBoundsCheck(cp_offset + j, checked_up_to); |
| 2887 } |
| 2888 } |
| 2889 } else { |
| 2890 ASSERT(elm.text_type() == TextElement::CHAR_CLASS); |
| 2891 if (pass == CHARACTER_CLASS_MATCH) { |
| 2892 if (first_element_checked && i == 0) continue; |
| 2893 if (DeterminedAlready(quick_check, elm.cp_offset())) continue; |
| 2894 RegExpCharacterClass* cc = elm.char_class(); |
| 2895 EmitCharClass(assembler, |
| 2896 cc, |
| 2897 ascii, |
| 2898 backtrack, |
| 2899 cp_offset, |
| 2900 *checked_up_to < cp_offset, |
| 2901 preloaded, |
| 2902 I); |
| 2903 UpdateBoundsCheck(cp_offset, checked_up_to); |
| 2904 } |
| 2905 } |
| 2906 } |
| 2907 } |
| 2908 |
| 2909 |
| 2910 intptr_t ChoiceNode::CalculatePreloadCharacters(RegExpCompiler* compiler, |
| 2911 intptr_t eats_at_least) { |
| 2912 intptr_t preload_characters = Utils::Minimum(static_cast<intptr_t>(4), |
| 2913 eats_at_least); |
| 2914 if (compiler->macro_assembler()->CanReadUnaligned()) { |
| 2915 bool ascii = compiler->ascii(); |
| 2916 if (ascii) { |
| 2917 if (preload_characters > 4) preload_characters = 4; |
| 2918 // We can't preload 3 characters because there is no machine instruction |
| 2919 // to do that. We can't just load 4 because we could be reading |
| 2920 // beyond the end of the string, which could cause a memory fault. |
| 2921 if (preload_characters == 3) preload_characters = 2; |
| 2922 } else { |
| 2923 if (preload_characters > 2) preload_characters = 2; |
| 2924 } |
| 2925 } else { |
| 2926 if (preload_characters > 1) preload_characters = 1; |
| 2927 } |
| 2928 return preload_characters; |
| 2929 } |
| 2930 |
| 2931 |
| 2932 intptr_t TextNode::EatsAtLeast(intptr_t still_to_find, |
| 2933 intptr_t budget, |
| 2934 bool not_at_start) { |
| 2935 intptr_t answer = Length(); |
| 2936 if (answer >= still_to_find) return answer; |
| 2937 if (budget <= 0) return answer; |
| 2938 // We are not at start after this node so we set the last argument to 'true'. |
| 2939 return answer + on_success()->EatsAtLeast(still_to_find - answer, |
| 2940 budget - 1, |
| 2941 true); |
| 2942 } |
| 2943 |
| 2944 |
| 2945 intptr_t ActionNode::EatsAtLeast(intptr_t still_to_find, |
| 2946 intptr_t budget, |
| 2947 bool not_at_start) { |
| 2948 if (budget <= 0) return 0; |
| 2949 if (action_type_ == POSITIVE_SUBMATCH_SUCCESS) return 0; // Rewinds input! |
| 2950 return on_success()->EatsAtLeast(still_to_find, |
| 2951 budget - 1, |
| 2952 not_at_start); |
| 2953 } |
| 2954 |
| 2955 |
| 2956 intptr_t AssertionNode::EatsAtLeast(intptr_t still_to_find, |
| 2957 intptr_t budget, |
| 2958 bool not_at_start) { |
| 2959 if (budget <= 0) return 0; |
| 2960 // If we know we are not at the start and we are asked "how many characters |
| 2961 // will you match if you succeed?" then we can answer anything since false |
| 2962 // implies false. So lets just return the max answer (still_to_find) since |
| 2963 // that won't prevent us from preloading a lot of characters for the other |
| 2964 // branches in the node graph. |
| 2965 if (assertion_type() == AT_START && not_at_start) return still_to_find; |
| 2966 return on_success()->EatsAtLeast(still_to_find, |
| 2967 budget - 1, |
| 2968 not_at_start); |
| 2969 } |
| 2970 |
| 2971 |
| 2972 intptr_t BackReferenceNode::EatsAtLeast(intptr_t still_to_find, |
| 2973 intptr_t budget, |
| 2974 bool not_at_start) { |
| 2975 if (budget <= 0) return 0; |
| 2976 return on_success()->EatsAtLeast(still_to_find, |
| 2977 budget - 1, |
| 2978 not_at_start); |
| 2979 } |
| 2980 |
| 2981 |
| 2982 intptr_t ChoiceNode::EatsAtLeastHelper(intptr_t still_to_find, |
| 2983 intptr_t budget, |
| 2984 RegExpNode* ignore_this_node, |
| 2985 bool not_at_start) { |
| 2986 if (budget <= 0) return 0; |
| 2987 intptr_t min = 100; |
| 2988 intptr_t choice_count = alternatives_->length(); |
| 2989 budget = (budget - 1) / choice_count; |
| 2990 for (intptr_t i = 0; i < choice_count; i++) { |
| 2991 RegExpNode* node = (*alternatives_)[i].node(); |
| 2992 if (node == ignore_this_node) continue; |
| 2993 intptr_t node_eats_at_least = |
| 2994 node->EatsAtLeast(still_to_find, budget, not_at_start); |
| 2995 if (node_eats_at_least < min) min = node_eats_at_least; |
| 2996 if (min == 0) return 0; |
| 2997 } |
| 2998 return min; |
| 2999 } |
| 3000 |
| 3001 |
| 3002 intptr_t ChoiceNode::EatsAtLeast(intptr_t still_to_find, |
| 3003 intptr_t budget, |
| 3004 bool not_at_start) { |
| 3005 return EatsAtLeastHelper(still_to_find, |
| 3006 budget, |
| 3007 NULL, |
| 3008 not_at_start); |
| 3009 } |
| 3010 |
| 3011 |
| 3012 intptr_t NegativeLookaheadChoiceNode::EatsAtLeast(intptr_t still_to_find, |
| 3013 intptr_t budget, |
| 3014 bool not_at_start) { |
| 3015 if (budget <= 0) return 0; |
| 3016 // Alternative 0 is the negative lookahead, alternative 1 is what comes |
| 3017 // afterwards. |
| 3018 RegExpNode* node = (*alternatives_)[1].node(); |
| 3019 return node->EatsAtLeast(still_to_find, budget - 1, not_at_start); |
| 3020 } |
| 3021 |
| 3022 |
| 3023 // Takes the left-most 1-bit and smears it out, setting all bits to its right. |
| 3024 static inline uint32_t SmearBitsRight(uint32_t v) { |
| 3025 v |= v >> 1; |
| 3026 v |= v >> 2; |
| 3027 v |= v >> 4; |
| 3028 v |= v >> 8; |
| 3029 v |= v >> 16; |
| 3030 return v; |
| 3031 } |
| 3032 |
| 3033 |
| 3034 // Here is the meat of GetQuickCheckDetails (see also the comment on the |
| 3035 // super-class in the .h file). |
| 3036 // |
| 3037 // We iterate along the text object, building up for each character a |
| 3038 // mask and value that can be used to test for a quick failure to match. |
| 3039 // The masks and values for the positions will be combined into a single |
| 3040 // machine word for the current character width in order to be used in |
| 3041 // generating a quick check. |
| 3042 void TextNode::GetQuickCheckDetails(QuickCheckDetails* details, |
| 3043 RegExpCompiler* compiler, |
| 3044 intptr_t characters_filled_in, |
| 3045 bool not_at_start) { |
| 3046 ASSERT(characters_filled_in < details->characters()); |
| 3047 intptr_t characters = details->characters(); |
| 3048 intptr_t char_mask; |
| 3049 if (compiler->ascii()) { |
| 3050 char_mask = Symbols::kMaxOneCharCodeSymbol; |
| 3051 } else { |
| 3052 char_mask = Utf16::kMaxCodeUnit; |
| 3053 } |
| 3054 for (intptr_t k = 0; k < elms_->length(); k++) { |
| 3055 TextElement elm = elms_->At(k); |
| 3056 if (elm.text_type() == TextElement::ATOM) { |
| 3057 ZoneGrowableArray<uint16_t>* quarks = elm.atom()->data(); |
| 3058 for (intptr_t i = 0; i < characters && i < quarks->length(); i++) { |
| 3059 QuickCheckDetails::Position* pos = |
| 3060 details->positions(characters_filled_in); |
| 3061 uint16_t c = quarks->At(i); |
| 3062 if (c > char_mask) { |
| 3063 // If we expect a non-ASCII character from an ASCII string, |
| 3064 // there is no way we can match. Not even case independent |
| 3065 // matching can turn an ASCII character into non-ASCII or |
| 3066 // vice versa. |
| 3067 details->set_cannot_match(); |
| 3068 pos->determines_perfectly = false; |
| 3069 return; |
| 3070 } |
| 3071 if (compiler->ignore_case()) { |
| 3072 int32_t chars[unibrow::Ecma262UnCanonicalize::kMaxWidth]; |
| 3073 intptr_t length = |
| 3074 GetCaseIndependentLetters(c, compiler->ascii(), chars); |
| 3075 ASSERT(length != 0); // Can only happen if c > char_mask (see above). |
| 3076 if (length == 1) { |
| 3077 // This letter has no case equivalents, so it's nice and simple |
| 3078 // and the mask-compare will determine definitely whether we have |
| 3079 // a match at this character position. |
| 3080 pos->mask = char_mask; |
| 3081 pos->value = c; |
| 3082 pos->determines_perfectly = true; |
| 3083 } else { |
| 3084 uint32_t common_bits = char_mask; |
| 3085 uint32_t bits = chars[0]; |
| 3086 for (intptr_t j = 1; j < length; j++) { |
| 3087 uint32_t differing_bits = ((chars[j] & common_bits) ^ bits); |
| 3088 common_bits ^= differing_bits; |
| 3089 bits &= common_bits; |
| 3090 } |
| 3091 // If length is 2 and common bits has only one zero in it then |
| 3092 // our mask and compare instruction will determine definitely |
| 3093 // whether we have a match at this character position. Otherwise |
| 3094 // it can only be an approximate check. |
| 3095 uint32_t one_zero = (common_bits | ~char_mask); |
| 3096 if (length == 2 && ((~one_zero) & ((~one_zero) - 1)) == 0) { |
| 3097 pos->determines_perfectly = true; |
| 3098 } |
| 3099 pos->mask = common_bits; |
| 3100 pos->value = bits; |
| 3101 } |
| 3102 } else { |
| 3103 // Don't ignore case. Nice simple case where the mask-compare will |
| 3104 // determine definitely whether we have a match at this character |
| 3105 // position. |
| 3106 pos->mask = char_mask; |
| 3107 pos->value = c; |
| 3108 pos->determines_perfectly = true; |
| 3109 } |
| 3110 characters_filled_in++; |
| 3111 ASSERT(characters_filled_in <= details->characters()); |
| 3112 if (characters_filled_in == details->characters()) { |
| 3113 return; |
| 3114 } |
| 3115 } |
| 3116 } else { |
| 3117 QuickCheckDetails::Position* pos = |
| 3118 details->positions(characters_filled_in); |
| 3119 RegExpCharacterClass* tree = elm.char_class(); |
| 3120 ZoneGrowableArray<CharacterRange>* ranges = tree->ranges(); |
| 3121 if (tree->is_negated()) { |
| 3122 // A quick check uses multi-character mask and compare. There is no |
| 3123 // useful way to incorporate a negative char class into this scheme |
| 3124 // so we just conservatively create a mask and value that will always |
| 3125 // succeed. |
| 3126 pos->mask = 0; |
| 3127 pos->value = 0; |
| 3128 } else { |
| 3129 intptr_t first_range = 0; |
| 3130 while (ranges->At(first_range).from() > char_mask) { |
| 3131 first_range++; |
| 3132 if (first_range == ranges->length()) { |
| 3133 details->set_cannot_match(); |
| 3134 pos->determines_perfectly = false; |
| 3135 return; |
| 3136 } |
| 3137 } |
| 3138 CharacterRange range = ranges->At(first_range); |
| 3139 uint16_t from = range.from(); |
| 3140 uint16_t to = range.to(); |
| 3141 if (to > char_mask) { |
| 3142 to = char_mask; |
| 3143 } |
| 3144 uint32_t differing_bits = (from ^ to); |
| 3145 // A mask and compare is only perfect if the differing bits form a |
| 3146 // number like 00011111 with one single block of trailing 1s. |
| 3147 if ((differing_bits & (differing_bits + 1)) == 0 && |
| 3148 from + differing_bits == to) { |
| 3149 pos->determines_perfectly = true; |
| 3150 } |
| 3151 uint32_t common_bits = ~SmearBitsRight(differing_bits); |
| 3152 uint32_t bits = (from & common_bits); |
| 3153 for (intptr_t i = first_range + 1; i < ranges->length(); i++) { |
| 3154 CharacterRange range = ranges->At(i); |
| 3155 uint16_t from = range.from(); |
| 3156 uint16_t to = range.to(); |
| 3157 if (from > char_mask) continue; |
| 3158 if (to > char_mask) to = char_mask; |
| 3159 // Here we are combining more ranges into the mask and compare |
| 3160 // value. With each new range the mask becomes more sparse and |
| 3161 // so the chances of a false positive rise. A character class |
| 3162 // with multiple ranges is assumed never to be equivalent to a |
| 3163 // mask and compare operation. |
| 3164 pos->determines_perfectly = false; |
| 3165 uint32_t new_common_bits = (from ^ to); |
| 3166 new_common_bits = ~SmearBitsRight(new_common_bits); |
| 3167 common_bits &= new_common_bits; |
| 3168 bits &= new_common_bits; |
| 3169 uint32_t differing_bits = (from & common_bits) ^ bits; |
| 3170 common_bits ^= differing_bits; |
| 3171 bits &= common_bits; |
| 3172 } |
| 3173 pos->mask = common_bits; |
| 3174 pos->value = bits; |
| 3175 } |
| 3176 characters_filled_in++; |
| 3177 ASSERT(characters_filled_in <= details->characters()); |
| 3178 if (characters_filled_in == details->characters()) { |
| 3179 return; |
| 3180 } |
| 3181 } |
| 3182 } |
| 3183 ASSERT(characters_filled_in != details->characters()); |
| 3184 if (!details->cannot_match()) { |
| 3185 on_success()-> GetQuickCheckDetails(details, |
| 3186 compiler, |
| 3187 characters_filled_in, |
| 3188 true); |
| 3189 } |
| 3190 } |
| 3191 |
| 3192 |
| 3193 void LoopChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details, |
| 3194 RegExpCompiler* compiler, |
| 3195 intptr_t characters_filled_in, |
| 3196 bool not_at_start) { |
| 3197 if (body_can_be_zero_length_ || info()->visited) return; |
| 3198 VisitMarker marker(info()); |
| 3199 return ChoiceNode::GetQuickCheckDetails(details, |
| 3200 compiler, |
| 3201 characters_filled_in, |
| 3202 not_at_start); |
| 3203 } |
| 3204 |
| 3205 |
| 3206 intptr_t TextNode::Length() { |
| 3207 TextElement elm = elms_->Last(); |
| 3208 ASSERT(elm.cp_offset() >= 0); |
| 3209 return elm.cp_offset() + elm.length(); |
| 3210 } |
| 3211 |
| 3212 |
| 3213 bool TextNode::SkipPass(intptr_t intptr_t_pass, bool ignore_case) { |
| 3214 TextEmitPassType pass = static_cast<TextEmitPassType>(intptr_t_pass); |
| 3215 if (ignore_case) { |
| 3216 return pass == SIMPLE_CHARACTER_MATCH; |
| 3217 } else { |
| 3218 return pass == NON_LETTER_CHARACTER_MATCH || pass == CASE_CHARACTER_MATCH; |
| 3219 } |
| 3220 } |
| 3221 |
| 3222 |
| 3223 static bool CompareInverseRanges(ZoneGrowableArray<CharacterRange>* ranges, |
| 3224 const intptr_t* special_class, |
| 3225 intptr_t length) { |
| 3226 length--; // Remove final 0x10000. |
| 3227 ASSERT(special_class[length] == 0x10000); |
| 3228 ASSERT(ranges->length() != 0); |
| 3229 ASSERT(length != 0); |
| 3230 ASSERT(special_class[0] != 0); |
| 3231 if (ranges->length() != (length >> 1) + 1) { |
| 3232 return false; |
| 3233 } |
| 3234 CharacterRange range = ranges->At(0); |
| 3235 if (range.from() != 0) { |
| 3236 return false; |
| 3237 } |
| 3238 for (intptr_t i = 0; i < length; i += 2) { |
| 3239 if (special_class[i] != (range.to() + 1)) { |
| 3240 return false; |
| 3241 } |
| 3242 range = ranges->At((i >> 1) + 1); |
| 3243 if (special_class[i+1] != range.from()) { |
| 3244 return false; |
| 3245 } |
| 3246 } |
| 3247 if (range.to() != 0xffff) { |
| 3248 return false; |
| 3249 } |
| 3250 return true; |
| 3251 } |
| 3252 |
| 3253 |
| 3254 static bool CompareRanges(ZoneGrowableArray<CharacterRange>* ranges, |
| 3255 const intptr_t* special_class, |
| 3256 intptr_t length) { |
| 3257 length--; // Remove final 0x10000. |
| 3258 ASSERT(special_class[length] == 0x10000); |
| 3259 if (ranges->length() * 2 != length) { |
| 3260 return false; |
| 3261 } |
| 3262 for (intptr_t i = 0; i < length; i += 2) { |
| 3263 CharacterRange range = ranges->At(i >> 1); |
| 3264 if (range.from() != special_class[i] || |
| 3265 range.to() != special_class[i + 1] - 1) { |
| 3266 return false; |
| 3267 } |
| 3268 } |
| 3269 return true; |
| 3270 } |
| 3271 |
| 3272 |
| 3273 bool RegExpCharacterClass::is_standard() { |
| 3274 // TODO(lrn): Remove need for this function, by not throwing away information |
| 3275 // along the way. |
| 3276 if (is_negated_) { |
| 3277 return false; |
| 3278 } |
| 3279 if (set_.is_standard()) { |
| 3280 return true; |
| 3281 } |
| 3282 if (CompareRanges(set_.ranges(), kSpaceRanges, kSpaceRangeCount)) { |
| 3283 set_.set_standard_set_type('s'); |
| 3284 return true; |
| 3285 } |
| 3286 if (CompareInverseRanges(set_.ranges(), kSpaceRanges, kSpaceRangeCount)) { |
| 3287 set_.set_standard_set_type('S'); |
| 3288 return true; |
| 3289 } |
| 3290 if (CompareInverseRanges(set_.ranges(), |
| 3291 kLineTerminatorRanges, |
| 3292 kLineTerminatorRangeCount)) { |
| 3293 set_.set_standard_set_type('.'); |
| 3294 return true; |
| 3295 } |
| 3296 if (CompareRanges(set_.ranges(), |
| 3297 kLineTerminatorRanges, |
| 3298 kLineTerminatorRangeCount)) { |
| 3299 set_.set_standard_set_type('n'); |
| 3300 return true; |
| 3301 } |
| 3302 if (CompareRanges(set_.ranges(), kWordRanges, kWordRangeCount)) { |
| 3303 set_.set_standard_set_type('w'); |
| 3304 return true; |
| 3305 } |
| 3306 if (CompareInverseRanges(set_.ranges(), kWordRanges, kWordRangeCount)) { |
| 3307 set_.set_standard_set_type('W'); |
| 3308 return true; |
| 3309 } |
| 3310 return false; |
| 3311 } |
| 3312 |
| 3313 |
| 3314 void TextNode::MakeCaseIndependent(bool is_ascii) { |
| 3315 intptr_t element_count = elms_->length(); |
| 3316 for (intptr_t i = 0; i < element_count; i++) { |
| 3317 TextElement elm = elms_->At(i); |
| 3318 if (elm.text_type() == TextElement::CHAR_CLASS) { |
| 3319 RegExpCharacterClass* cc = elm.char_class(); |
| 3320 // None of the standard character classes is different in the case |
| 3321 // independent case and it slows us down if we don't know that. |
| 3322 if (cc->is_standard()) continue; |
| 3323 ZoneGrowableArray<CharacterRange>* ranges = cc->ranges(); |
| 3324 intptr_t range_count = ranges->length(); |
| 3325 for (intptr_t j = 0; j < range_count; j++) { |
| 3326 (*ranges)[j].AddCaseEquivalents(ranges, is_ascii, I); |
| 3327 } |
| 3328 } |
| 3329 } |
| 3330 } |
| 3331 |
| 3332 |
| 3333 intptr_t TextNode::GreedyLoopTextLength() { |
| 3334 TextElement elm = elms_->At(elms_->length() - 1); |
| 3335 return elm.cp_offset() + elm.length(); |
| 3336 } |
| 3337 |
| 3338 |
| 3339 RegExpNode* TextNode::GetSuccessorOfOmnivorousTextNode( |
| 3340 RegExpCompiler* compiler) { |
| 3341 if (elms_->length() != 1) return NULL; |
| 3342 TextElement elm = elms_->At(0); |
| 3343 if (elm.text_type() != TextElement::CHAR_CLASS) return NULL; |
| 3344 RegExpCharacterClass* node = elm.char_class(); |
| 3345 ZoneGrowableArray<CharacterRange>* ranges = node->ranges(); |
| 3346 if (!CharacterRange::IsCanonical(ranges)) { |
| 3347 CharacterRange::Canonicalize(ranges); |
| 3348 } |
| 3349 if (node->is_negated()) { |
| 3350 return ranges->length() == 0 ? on_success() : NULL; |
| 3351 } |
| 3352 if (ranges->length() != 1) return NULL; |
| 3353 uint32_t max_char; |
| 3354 if (compiler->ascii()) { |
| 3355 max_char = Symbols::kMaxOneCharCodeSymbol; |
| 3356 } else { |
| 3357 max_char = Utf16::kMaxCodeUnit; |
| 3358 } |
| 3359 return ranges->At(0).IsEverything(max_char) ? on_success() : NULL; |
| 3360 } |
| 3361 |
| 3362 |
| 3363 ActionNode* ActionNode::SetRegister(intptr_t reg, |
| 3364 intptr_t val, |
| 3365 RegExpNode* on_success) { |
| 3366 ActionNode* result = |
| 3367 new(on_success->isolate()) ActionNode(SET_REGISTER, on_success); |
| 3368 result->data_.u_store_register.reg = reg; |
| 3369 result->data_.u_store_register.value = val; |
| 3370 return result; |
| 3371 } |
| 3372 |
| 3373 |
| 3374 ActionNode* ActionNode::IncrementRegister(intptr_t reg, |
| 3375 RegExpNode* on_success) { |
| 3376 ActionNode* result = |
| 3377 new(on_success->isolate()) ActionNode(INCREMENT_REGISTER, on_success); |
| 3378 result->data_.u_increment_register.reg = reg; |
| 3379 return result; |
| 3380 } |
| 3381 |
| 3382 |
| 3383 ActionNode* ActionNode::StorePosition(intptr_t reg, |
| 3384 bool is_capture, |
| 3385 RegExpNode* on_success) { |
| 3386 ActionNode* result = |
| 3387 new(on_success->isolate()) ActionNode(STORE_POSITION, on_success); |
| 3388 result->data_.u_position_register.reg = reg; |
| 3389 result->data_.u_position_register.is_capture = is_capture; |
| 3390 return result; |
| 3391 } |
| 3392 |
| 3393 |
| 3394 ActionNode* ActionNode::ClearCaptures(Interval range, |
| 3395 RegExpNode* on_success) { |
| 3396 ActionNode* result = |
| 3397 new(on_success->isolate()) ActionNode(CLEAR_CAPTURES, on_success); |
| 3398 result->data_.u_clear_captures.range_from = range.from(); |
| 3399 result->data_.u_clear_captures.range_to = range.to(); |
| 3400 return result; |
| 3401 } |
| 3402 |
| 3403 |
| 3404 ActionNode* ActionNode::BeginSubmatch(intptr_t stack_reg, |
| 3405 intptr_t position_reg, |
| 3406 RegExpNode* on_success) { |
| 3407 ActionNode* result = |
| 3408 new(on_success->isolate()) ActionNode(BEGIN_SUBMATCH, on_success); |
| 3409 result->data_.u_submatch.stack_pointer_register = stack_reg; |
| 3410 result->data_.u_submatch.current_position_register = position_reg; |
| 3411 return result; |
| 3412 } |
| 3413 |
| 3414 |
| 3415 ActionNode* ActionNode::PositiveSubmatchSuccess(intptr_t stack_reg, |
| 3416 intptr_t position_reg, |
| 3417 intptr_t clear_register_count, |
| 3418 intptr_t clear_register_from, |
| 3419 RegExpNode* on_success) { |
| 3420 ActionNode* result = |
| 3421 new(on_success->isolate()) ActionNode(POSITIVE_SUBMATCH_SUCCESS, |
| 3422 on_success); |
| 3423 result->data_.u_submatch.stack_pointer_register = stack_reg; |
| 3424 result->data_.u_submatch.current_position_register = position_reg; |
| 3425 result->data_.u_submatch.clear_register_count = clear_register_count; |
| 3426 result->data_.u_submatch.clear_register_from = clear_register_from; |
| 3427 return result; |
| 3428 } |
| 3429 |
| 3430 |
| 3431 ActionNode* ActionNode::EmptyMatchCheck(intptr_t start_register, |
| 3432 intptr_t repetition_register, |
| 3433 intptr_t repetition_limit, |
| 3434 RegExpNode* on_success) { |
| 3435 ActionNode* result = |
| 3436 new(on_success->isolate()) ActionNode(EMPTY_MATCH_CHECK, on_success); |
| 3437 result->data_.u_empty_match_check.start_register = start_register; |
| 3438 result->data_.u_empty_match_check.repetition_register = repetition_register; |
| 3439 result->data_.u_empty_match_check.repetition_limit = repetition_limit; |
| 3440 return result; |
| 3441 } |
| 3442 |
| 3443 |
| 3444 TextElement TextElement::Atom(RegExpAtom* atom) { |
| 3445 return TextElement(ATOM, atom); |
| 3446 } |
| 3447 |
| 3448 |
| 3449 TextElement TextElement::CharClass(RegExpCharacterClass* char_class) { |
| 3450 return TextElement(CHAR_CLASS, char_class); |
| 3451 } |
| 3452 |
| 3453 |
| 3454 intptr_t TextElement::length() const { |
| 3455 switch (text_type()) { |
| 3456 case ATOM: |
| 3457 return atom()->length(); |
| 3458 |
| 3459 case CHAR_CLASS: |
| 3460 return 1; |
| 3461 } |
| 3462 UNREACHABLE(); |
| 3463 return 0; |
| 3464 } |
| 3465 |
| 3466 |
| 3467 static void AddClass(const intptr_t* elmv, |
| 3468 intptr_t elmc, |
| 3469 ZoneGrowableArray<CharacterRange>* ranges) { |
| 3470 elmc--; |
| 3471 ASSERT(elmv[elmc] == 0x10000); |
| 3472 for (intptr_t i = 0; i < elmc; i += 2) { |
| 3473 ASSERT(elmv[i] < elmv[i + 1]); |
| 3474 ranges->Add(CharacterRange(elmv[i], elmv[i + 1] - 1)); |
| 3475 } |
| 3476 } |
| 3477 |
| 3478 |
| 3479 static void AddClassNegated(const intptr_t *elmv, |
| 3480 intptr_t elmc, |
| 3481 ZoneGrowableArray<CharacterRange>* ranges) { |
| 3482 elmc--; |
| 3483 ASSERT(elmv[elmc] == 0x10000); |
| 3484 ASSERT(elmv[0] != 0x0000); |
| 3485 ASSERT(elmv[elmc-1] != Utf16::kMaxCodeUnit); |
| 3486 uint16_t last = 0x0000; |
| 3487 for (intptr_t i = 0; i < elmc; i += 2) { |
| 3488 ASSERT(last <= elmv[i] - 1); |
| 3489 ASSERT(elmv[i] < elmv[i + 1]); |
| 3490 ranges->Add(CharacterRange(last, elmv[i] - 1)); |
| 3491 last = elmv[i + 1]; |
| 3492 } |
| 3493 ranges->Add(CharacterRange(last, Utf16::kMaxCodeUnit)); |
| 3494 } |
| 3495 |
| 3496 |
| 3497 void CharacterRange::AddClassEscape(uint16_t type, |
| 3498 ZoneGrowableArray<CharacterRange>* ranges) { |
| 3499 switch (type) { |
| 3500 case 's': |
| 3501 AddClass(kSpaceRanges, kSpaceRangeCount, ranges); |
| 3502 break; |
| 3503 case 'S': |
| 3504 AddClassNegated(kSpaceRanges, kSpaceRangeCount, ranges); |
| 3505 break; |
| 3506 case 'w': |
| 3507 AddClass(kWordRanges, kWordRangeCount, ranges); |
| 3508 break; |
| 3509 case 'W': |
| 3510 AddClassNegated(kWordRanges, kWordRangeCount, ranges); |
| 3511 break; |
| 3512 case 'd': |
| 3513 AddClass(kDigitRanges, kDigitRangeCount, ranges); |
| 3514 break; |
| 3515 case 'D': |
| 3516 AddClassNegated(kDigitRanges, kDigitRangeCount, ranges); |
| 3517 break; |
| 3518 case '.': |
| 3519 AddClassNegated(kLineTerminatorRanges, |
| 3520 kLineTerminatorRangeCount, |
| 3521 ranges); |
| 3522 break; |
| 3523 // This is not a character range as defined by the spec but a |
| 3524 // convenient shorthand for a character class that matches any |
| 3525 // character. |
| 3526 case '*': |
| 3527 ranges->Add(CharacterRange::Everything()); |
| 3528 break; |
| 3529 // This is the set of characters matched by the $ and ^ symbols |
| 3530 // in multiline mode. |
| 3531 case 'n': |
| 3532 AddClass(kLineTerminatorRanges, |
| 3533 kLineTerminatorRangeCount, |
| 3534 ranges); |
| 3535 break; |
| 3536 default: |
| 3537 UNREACHABLE(); |
| 3538 } |
| 3539 } |
| 3540 |
| 3541 |
| 3542 // Move a number of elements in a zonelist to another position |
| 3543 // in the same list. Handles overlapping source and target areas. |
| 3544 static void MoveRanges(ZoneGrowableArray<CharacterRange>* list, |
| 3545 intptr_t from, |
| 3546 intptr_t to, |
| 3547 intptr_t count) { |
| 3548 // Ranges are potentially overlapping. |
| 3549 if (from < to) { |
| 3550 for (intptr_t i = count - 1; i >= 0; i--) { |
| 3551 (*list)[to + i] = list->At(from + i); |
| 3552 } |
| 3553 } else { |
| 3554 for (intptr_t i = 0; i < count; i++) { |
| 3555 (*list)[to + i] = list->At(from + i); |
| 3556 } |
| 3557 } |
| 3558 } |
| 3559 |
| 3560 |
| 3561 static intptr_t InsertRangeInCanonicalList( |
| 3562 ZoneGrowableArray<CharacterRange>* list, |
| 3563 intptr_t count, |
| 3564 CharacterRange insert) { |
| 3565 // Inserts a range into list[0..count[, which must be sorted |
| 3566 // by from value and non-overlapping and non-adjacent, using at most |
| 3567 // list[0..count] for the result. Returns the number of resulting |
| 3568 // canonicalized ranges. Inserting a range may collapse existing ranges into |
| 3569 // fewer ranges, so the return value can be anything in the range 1..count+1. |
| 3570 uint16_t from = insert.from(); |
| 3571 uint16_t to = insert.to(); |
| 3572 intptr_t start_pos = 0; |
| 3573 intptr_t end_pos = count; |
| 3574 for (intptr_t i = count - 1; i >= 0; i--) { |
| 3575 CharacterRange current = list->At(i); |
| 3576 if (current.from() > to + 1) { |
| 3577 end_pos = i; |
| 3578 } else if (current.to() + 1 < from) { |
| 3579 start_pos = i + 1; |
| 3580 break; |
| 3581 } |
| 3582 } |
| 3583 |
| 3584 // Inserted range overlaps, or is adjacent to, ranges at positions |
| 3585 // [start_pos..end_pos[. Ranges before start_pos or at or after end_pos are |
| 3586 // not affected by the insertion. |
| 3587 // If start_pos == end_pos, the range must be inserted before start_pos. |
| 3588 // if start_pos < end_pos, the entire range from start_pos to end_pos |
| 3589 // must be merged with the insert range. |
| 3590 |
| 3591 if (start_pos == end_pos) { |
| 3592 // Insert between existing ranges at position start_pos. |
| 3593 if (start_pos < count) { |
| 3594 MoveRanges(list, start_pos, start_pos + 1, count - start_pos); |
| 3595 } |
| 3596 (*list)[start_pos] = insert; |
| 3597 return count + 1; |
| 3598 } |
| 3599 if (start_pos + 1 == end_pos) { |
| 3600 // Replace single existing range at position start_pos. |
| 3601 CharacterRange to_replace = list->At(start_pos); |
| 3602 intptr_t new_from = Utils::Minimum(to_replace.from(), from); |
| 3603 intptr_t new_to = Utils::Maximum(to_replace.to(), to); |
| 3604 (*list)[start_pos] = CharacterRange(new_from, new_to); |
| 3605 return count; |
| 3606 } |
| 3607 // Replace a number of existing ranges from start_pos to end_pos - 1. |
| 3608 // Move the remaining ranges down. |
| 3609 |
| 3610 intptr_t new_from = Utils::Minimum(list->At(start_pos).from(), from); |
| 3611 intptr_t new_to = Utils::Maximum(list->At(end_pos - 1).to(), to); |
| 3612 if (end_pos < count) { |
| 3613 MoveRanges(list, end_pos, start_pos + 1, count - end_pos); |
| 3614 } |
| 3615 (*list)[start_pos] = CharacterRange(new_from, new_to); |
| 3616 return count - (end_pos - start_pos) + 1; |
| 3617 } |
| 3618 |
| 3619 |
| 3620 bool CharacterRange::IsCanonical(ZoneGrowableArray<CharacterRange>* ranges) { |
| 3621 ASSERT(ranges != NULL); |
| 3622 intptr_t n = ranges->length(); |
| 3623 if (n <= 1) return true; |
| 3624 intptr_t max = ranges->At(0).to(); |
| 3625 for (intptr_t i = 1; i < n; i++) { |
| 3626 CharacterRange next_range = ranges->At(i); |
| 3627 if (next_range.from() <= max + 1) return false; |
| 3628 max = next_range.to(); |
| 3629 } |
| 3630 return true; |
| 3631 } |
| 3632 |
| 3633 |
| 3634 void CharacterRange::Canonicalize( |
| 3635 ZoneGrowableArray<CharacterRange>* character_ranges) { |
| 3636 if (character_ranges->length() <= 1) return; |
| 3637 // Check whether ranges are already canonical (increasing, non-overlapping, |
| 3638 // non-adjacent). |
| 3639 intptr_t n = character_ranges->length(); |
| 3640 intptr_t max = character_ranges->At(0).to(); |
| 3641 intptr_t i = 1; |
| 3642 while (i < n) { |
| 3643 CharacterRange current = character_ranges->At(i); |
| 3644 if (current.from() <= max + 1) { |
| 3645 break; |
| 3646 } |
| 3647 max = current.to(); |
| 3648 i++; |
| 3649 } |
| 3650 // Canonical until the i'th range. If that's all of them, we are done. |
| 3651 if (i == n) return; |
| 3652 |
| 3653 // The ranges at index i and forward are not canonicalized. Make them so by |
| 3654 // doing the equivalent of insertion sort (inserting each into the previous |
| 3655 // list, in order). |
| 3656 // Notice that inserting a range can reduce the number of ranges in the |
| 3657 // result due to combining of adjacent and overlapping ranges. |
| 3658 intptr_t read = i; // Range to insert. |
| 3659 intptr_t num_canonical = i; // Length of canonicalized part of list. |
| 3660 do { |
| 3661 num_canonical = InsertRangeInCanonicalList(character_ranges, |
| 3662 num_canonical, |
| 3663 character_ranges->At(read)); |
| 3664 read++; |
| 3665 } while (read < n); |
| 3666 character_ranges->TruncateTo(num_canonical); |
| 3667 |
| 3668 ASSERT(CharacterRange::IsCanonical(character_ranges)); |
| 3669 } |
| 3670 |
| 3671 |
| 3672 void CharacterSet::Canonicalize() { |
| 3673 // Special/default classes are always considered canonical. The result |
| 3674 // of calling ranges() will be sorted. |
| 3675 if (ranges_ == NULL) return; |
| 3676 CharacterRange::Canonicalize(ranges_); |
| 3677 } |
| 3678 |
| 3679 |
| 3680 ZoneGrowableArray<CharacterRange>* CharacterSet::ranges() { |
| 3681 if (ranges_ == NULL) { |
| 3682 ranges_ = new ZoneGrowableArray<CharacterRange>(2); |
| 3683 CharacterRange::AddClassEscape(standard_set_type_, ranges_); |
| 3684 } |
| 3685 return ranges_; |
| 3686 } |
| 3687 |
| 3688 |
| 3689 void CharacterRange::Negate(ZoneGrowableArray<CharacterRange>* ranges, |
| 3690 ZoneGrowableArray<CharacterRange>* negated_ranges) { |
| 3691 ASSERT(CharacterRange::IsCanonical(ranges)); |
| 3692 ASSERT(negated_ranges->length() == 0); |
| 3693 intptr_t range_count = ranges->length(); |
| 3694 uint16_t from = 0; |
| 3695 intptr_t i = 0; |
| 3696 if (range_count > 0 && ranges->At(0).from() == 0) { |
| 3697 from = ranges->At(0).to(); |
| 3698 i = 1; |
| 3699 } |
| 3700 while (i < range_count) { |
| 3701 CharacterRange range = ranges->At(i); |
| 3702 negated_ranges->Add(CharacterRange(from + 1, range.from() - 1)); |
| 3703 from = range.to(); |
| 3704 i++; |
| 3705 } |
| 3706 if (from < Utf16::kMaxCodeUnit) { |
| 3707 negated_ranges->Add(CharacterRange(from + 1, Utf16::kMaxCodeUnit)); |
| 3708 } |
| 3709 } |
| 3710 |
| 3711 |
| 3712 void CharacterRange::AddCaseEquivalents( |
| 3713 ZoneGrowableArray<CharacterRange>* ranges, |
| 3714 bool is_ascii, |
| 3715 Isolate* isolate) { |
| 3716 uint16_t bottom = from(); |
| 3717 uint16_t top = to(); |
| 3718 if (is_ascii && !RangeContainsLatin1Equivalents(*this)) { |
| 3719 if (bottom > Symbols::kMaxOneCharCodeSymbol) return; |
| 3720 if (top > Symbols::kMaxOneCharCodeSymbol) { |
| 3721 top = Symbols::kMaxOneCharCodeSymbol; |
| 3722 } |
| 3723 } |
| 3724 |
| 3725 unibrow::Mapping<unibrow::Ecma262UnCanonicalize> jsregexp_uncanonicalize; |
| 3726 unibrow::Mapping<unibrow::CanonicalizationRange> jsregexp_canonrange; |
| 3727 int32_t chars[unibrow::Ecma262UnCanonicalize::kMaxWidth]; |
| 3728 if (top == bottom) { |
| 3729 // If this is a singleton we just expand the one character. |
| 3730 intptr_t length = jsregexp_uncanonicalize.get(bottom, '\0', chars); // NOLIN
T |
| 3731 for (intptr_t i = 0; i < length; i++) { |
| 3732 uint32_t chr = chars[i]; |
| 3733 if (chr != bottom) { |
| 3734 ranges->Add(CharacterRange::Singleton(chars[i])); |
| 3735 } |
| 3736 } |
| 3737 } else { |
| 3738 // If this is a range we expand the characters block by block, |
| 3739 // expanding contiguous subranges (blocks) one at a time. |
| 3740 // The approach is as follows. For a given start character we |
| 3741 // look up the remainder of the block that contains it (represented |
| 3742 // by the end point), for instance we find 'z' if the character |
| 3743 // is 'c'. A block is characterized by the property |
| 3744 // that all characters uncanonicalize in the same way, except that |
| 3745 // each entry in the result is incremented by the distance from the first |
| 3746 // element. So a-z is a block because 'a' uncanonicalizes to ['a', 'A'] and
// NOLINT |
| 3747 // the k'th letter uncanonicalizes to ['a' + k, 'A' + k]. |
| 3748 // Once we've found the end point we look up its uncanonicalization |
| 3749 // and produce a range for each element. For instance for [c-f] |
| 3750 // we look up ['z', 'Z'] and produce [c-f] and [C-F]. We then only |
| 3751 // add a range if it is not already contained in the input, so [c-f] |
| 3752 // will be skipped but [C-F] will be added. If this range is not |
| 3753 // completely contained in a block we do this for all the blocks |
| 3754 // covered by the range (handling characters that is not in a block |
| 3755 // as a "singleton block"). |
| 3756 int32_t range[unibrow::Ecma262UnCanonicalize::kMaxWidth]; |
| 3757 intptr_t pos = bottom; |
| 3758 while (pos <= top) { |
| 3759 intptr_t length = jsregexp_canonrange.get(pos, '\0', range); |
| 3760 uint16_t block_end; |
| 3761 if (length == 0) { |
| 3762 block_end = pos; |
| 3763 } else { |
| 3764 ASSERT(length == 1); |
| 3765 block_end = range[0]; |
| 3766 } |
| 3767 intptr_t end = (block_end > top) ? top : block_end; |
| 3768 length = jsregexp_uncanonicalize.get(block_end, '\0', range); // NOLINT |
| 3769 for (intptr_t i = 0; i < length; i++) { |
| 3770 uint32_t c = range[i]; |
| 3771 uint16_t range_from = c - (block_end - pos); |
| 3772 uint16_t range_to = c - (block_end - end); |
| 3773 if (!(bottom <= range_from && range_to <= top)) { |
| 3774 ranges->Add(CharacterRange(range_from, range_to)); |
| 3775 } |
| 3776 } |
| 3777 pos = end + 1; |
| 3778 } |
| 3779 } |
| 3780 } |
| 3781 |
| 3782 |
| 3783 // ------------------------------------------------------------------- |
| 3784 // Splay tree |
| 3785 |
| 3786 |
| 3787 // Workaround for the fact that ZoneGrowableArray does not have contains(). |
| 3788 static bool ArrayContains(ZoneGrowableArray<unsigned>* array, |
| 3789 unsigned value) { |
| 3790 for (intptr_t i = 0; i < array->length(); i++) { |
| 3791 if (array->At(i) == value) { |
| 3792 return true; |
| 3793 } |
| 3794 } |
| 3795 return false; |
| 3796 } |
| 3797 |
| 3798 |
| 3799 void OutSet::Set(unsigned value, Isolate* isolate) { |
| 3800 if (value < kFirstLimit) { |
| 3801 first_ |= (1 << value); |
| 3802 } else { |
| 3803 if (remaining_ == NULL) |
| 3804 remaining_ = new(isolate) ZoneGrowableArray<unsigned>(1); |
| 3805 |
| 3806 bool remaining_contains_value = ArrayContains(remaining_, value); |
| 3807 if (remaining_->is_empty() || !remaining_contains_value) { |
| 3808 remaining_->Add(value); |
| 3809 } |
| 3810 } |
| 3811 } |
| 3812 |
| 3813 |
| 3814 bool OutSet::Get(unsigned value) const { |
| 3815 if (value < kFirstLimit) { |
| 3816 return (first_ & (1 << value)) != 0; |
| 3817 } else if (remaining_ == NULL) { |
| 3818 return false; |
| 3819 } else { |
| 3820 return ArrayContains(remaining_, value); |
| 3821 } |
| 3822 } |
| 3823 |
| 3824 void AssertionNode::GetQuickCheckDetails(QuickCheckDetails* details, |
| 3825 RegExpCompiler* compiler, |
| 3826 intptr_t filled_in, |
| 3827 bool not_at_start) { |
| 3828 if (assertion_type_ == AT_START && not_at_start) { |
| 3829 details->set_cannot_match(); |
| 3830 return; |
| 3831 } |
| 3832 return on_success()->GetQuickCheckDetails(details, |
| 3833 compiler, |
| 3834 filled_in, |
| 3835 not_at_start); |
| 3836 } |
| 3837 |
| 3838 |
| 3839 void GuardedAlternative::AddGuard(Guard* guard, Isolate* isolate) { |
| 3840 if (guards_ == NULL) |
| 3841 guards_ = new(isolate) ZoneGrowableArray<Guard*>(1); |
| 3842 guards_->Add(guard); |
| 3843 } |
| 3844 |
| 3845 |
| 3846 void LoopChoiceNode::AddLoopAlternative(GuardedAlternative alt) { |
| 3847 ASSERT(loop_node_ == NULL); |
| 3848 AddAlternative(alt); |
| 3849 loop_node_ = alt.node(); |
| 3850 } |
| 3851 |
| 3852 |
| 3853 void LoopChoiceNode::AddContinueAlternative(GuardedAlternative alt) { |
| 3854 ASSERT(continue_node_ == NULL); |
| 3855 AddAlternative(alt); |
| 3856 continue_node_ = alt.node(); |
| 3857 } |
| 3858 |
| 3859 |
| 3860 void LoopChoiceNode::Accept(NodeVisitor* visitor) { |
| 3861 visitor->VisitLoopChoice(this); |
| 3862 } |
| 3863 |
| 3864 |
| 3865 intptr_t LoopChoiceNode::EatsAtLeast(intptr_t still_to_find, |
| 3866 intptr_t budget, |
| 3867 bool not_at_start) { |
| 3868 return EatsAtLeastHelper(still_to_find, |
| 3869 budget - 1, |
| 3870 loop_node_, |
| 3871 not_at_start); |
| 3872 } |
| 3873 |
| 3874 |
| 3875 void ChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details, |
| 3876 RegExpCompiler* compiler, |
| 3877 intptr_t characters_filled_in, |
| 3878 bool not_at_start) { |
| 3879 not_at_start = (not_at_start || not_at_start_); |
| 3880 intptr_t choice_count = alternatives_->length(); |
| 3881 ASSERT(choice_count > 0); |
| 3882 (*alternatives_)[0].node()->GetQuickCheckDetails(details, |
| 3883 compiler, |
| 3884 characters_filled_in, |
| 3885 not_at_start); |
| 3886 for (intptr_t i = 1; i < choice_count; i++) { |
| 3887 QuickCheckDetails new_details(details->characters()); |
| 3888 RegExpNode* node = (*alternatives_)[i].node(); |
| 3889 node->GetQuickCheckDetails(&new_details, compiler, |
| 3890 characters_filled_in, |
| 3891 not_at_start); |
| 3892 // Here we merge the quick match details of the two branches. |
| 3893 details->Merge(&new_details, characters_filled_in); |
| 3894 } |
| 3895 } |
| 3896 |
| 3897 |
| 3898 void QuickCheckDetails::Clear() { |
| 3899 for (intptr_t i = 0; i < characters_; i++) { |
| 3900 positions_[i].mask = 0; |
| 3901 positions_[i].value = 0; |
| 3902 positions_[i].determines_perfectly = false; |
| 3903 } |
| 3904 characters_ = 0; |
| 3905 } |
| 3906 |
| 3907 |
| 3908 void QuickCheckDetails::Advance(intptr_t by, bool ascii) { |
| 3909 ASSERT(by >= 0); |
| 3910 if (by >= characters_) { |
| 3911 Clear(); |
| 3912 return; |
| 3913 } |
| 3914 for (intptr_t i = 0; i < characters_ - by; i++) { |
| 3915 positions_[i] = positions_[by + i]; |
| 3916 } |
| 3917 for (intptr_t i = characters_ - by; i < characters_; i++) { |
| 3918 positions_[i].mask = 0; |
| 3919 positions_[i].value = 0; |
| 3920 positions_[i].determines_perfectly = false; |
| 3921 } |
| 3922 characters_ -= by; |
| 3923 // We could change mask_ and value_ here but we would never advance unless |
| 3924 // they had already been used in a check and they won't be used again because |
| 3925 // it would gain us nothing. So there's no point. |
| 3926 } |
| 3927 |
| 3928 |
| 3929 void QuickCheckDetails::Merge(QuickCheckDetails* other, intptr_t from_index) { |
| 3930 ASSERT(characters_ == other->characters_); |
| 3931 if (other->cannot_match_) { |
| 3932 return; |
| 3933 } |
| 3934 if (cannot_match_) { |
| 3935 *this = *other; |
| 3936 return; |
| 3937 } |
| 3938 for (intptr_t i = from_index; i < characters_; i++) { |
| 3939 QuickCheckDetails::Position* pos = positions(i); |
| 3940 QuickCheckDetails::Position* other_pos = other->positions(i); |
| 3941 if (pos->mask != other_pos->mask || |
| 3942 pos->value != other_pos->value || |
| 3943 !other_pos->determines_perfectly) { |
| 3944 // Our mask-compare operation will be approximate unless we have the |
| 3945 // exact same operation on both sides of the alternation. |
| 3946 pos->determines_perfectly = false; |
| 3947 } |
| 3948 pos->mask &= other_pos->mask; |
| 3949 pos->value &= pos->mask; |
| 3950 other_pos->value &= pos->mask; |
| 3951 uint16_t differing_bits = (pos->value ^ other_pos->value); |
| 3952 pos->mask &= ~differing_bits; |
| 3953 pos->value &= pos->mask; |
| 3954 } |
| 3955 } |
| 3956 |
| 3957 |
| 3958 // Finds the fixed match length of a sequence of nodes that goes from |
| 3959 // this alternative and back to this choice node. If there are variable |
| 3960 // length nodes or other complications in the way then return a sentinel |
| 3961 // value indicating that a greedy loop cannot be constructed. |
| 3962 intptr_t ChoiceNode::GreedyLoopTextLengthForAlternative( |
| 3963 GuardedAlternative* alternative) { |
| 3964 intptr_t length = 0; |
| 3965 RegExpNode* node = alternative->node(); |
| 3966 // Later we will generate code for all these text nodes using recursion |
| 3967 // so we have to limit the max number. |
| 3968 intptr_t recursion_depth = 0; |
| 3969 while (node != this) { |
| 3970 if (recursion_depth++ > RegExpCompiler::kMaxRecursion) { |
| 3971 return kNodeIsTooComplexForGreedyLoops; |
| 3972 } |
| 3973 intptr_t node_length = node->GreedyLoopTextLength(); |
| 3974 if (node_length == kNodeIsTooComplexForGreedyLoops) { |
| 3975 return kNodeIsTooComplexForGreedyLoops; |
| 3976 } |
| 3977 length += node_length; |
| 3978 SeqRegExpNode* seq_node = static_cast<SeqRegExpNode*>(node); |
| 3979 node = seq_node->on_success(); |
| 3980 } |
| 3981 return length; |
| 3982 } |
| 3983 |
| 3984 |
| 3985 void NegativeLookaheadChoiceNode::GetQuickCheckDetails( |
| 3986 QuickCheckDetails* details, |
| 3987 RegExpCompiler* compiler, |
| 3988 intptr_t filled_in, |
| 3989 bool not_at_start) { |
| 3990 // Alternative 0 is the negative lookahead, alternative 1 is what comes |
| 3991 // afterwards. |
| 3992 RegExpNode* node = (*alternatives_)[1].node(); |
| 3993 return node->GetQuickCheckDetails(details, compiler, filled_in, not_at_start); |
| 3994 } |
| 3995 |
| 3996 |
| 3997 static RegExpEngine::CompilationResult IrregexpRegExpTooBig() { |
| 3998 return RegExpEngine::CompilationResult("RegExp too big"); |
| 3999 } |
| 4000 |
| 4001 |
| 4002 // Attempts to compile the regexp using an Irregexp code generator. Returns |
| 4003 // a fixed array or a null handle depending on whether it succeeded. |
| 4004 RegExpCompiler::RegExpCompiler(intptr_t capture_count, bool ignore_case, |
| 4005 intptr_t specialization_cid) |
| 4006 : next_register_(2 * (capture_count + 1)), |
| 4007 work_list_(NULL), |
| 4008 recursion_depth_(0), |
| 4009 ignore_case_(ignore_case), |
| 4010 specialization_cid_(specialization_cid), |
| 4011 reg_exp_too_big_(false), |
| 4012 current_expansion_factor_(1), |
| 4013 isolate_(Isolate::Current()) { |
| 4014 accept_ = new(I) EndNode(EndNode::ACCEPT, I); |
| 4015 } |
| 4016 |
| 4017 |
| 4018 RegExpEngine::CompilationResult RegExpCompiler::Assemble( |
| 4019 IRRegExpMacroAssembler* macro_assembler, |
| 4020 RegExpNode* start, |
| 4021 intptr_t capture_count, |
| 4022 const String& pattern) { |
| 4023 static const bool use_slow_safe_regexp_compiler = false; |
| 4024 |
| 4025 macro_assembler->set_slow_safe(use_slow_safe_regexp_compiler); |
| 4026 macro_assembler_ = macro_assembler; |
| 4027 |
| 4028 ZoneGrowableArray<RegExpNode*> work_list(0); |
| 4029 work_list_ = &work_list; |
| 4030 BlockLabel fail; |
| 4031 macro_assembler_->PushBacktrack(&fail); |
| 4032 Trace new_trace; |
| 4033 start->Emit(this, &new_trace); |
| 4034 macro_assembler_->BindBlock(&fail); |
| 4035 macro_assembler_->Fail(); |
| 4036 while (!work_list.is_empty()) { |
| 4037 work_list.RemoveLast()->Emit(this, &new_trace); |
| 4038 } |
| 4039 if (reg_exp_too_big_) return IrregexpRegExpTooBig(); |
| 4040 |
| 4041 macro_assembler->FinalizeIndirectGotos(); |
| 4042 |
| 4043 return RegExpEngine::CompilationResult(macro_assembler, |
| 4044 macro_assembler->graph_entry(), |
| 4045 macro_assembler->num_blocks(), |
| 4046 macro_assembler->num_stack_locals()); |
| 4047 } |
| 4048 |
| 4049 |
| 4050 RegExpEngine::CompilationResult RegExpEngine::Compile( |
| 4051 RegExpCompileData* data, |
| 4052 const ParsedFunction* parsed_function, |
| 4053 const ZoneGrowableArray<const ICData*>& ic_data_array) { |
| 4054 Isolate* isolate = Isolate::Current(); |
| 4055 |
| 4056 const Function& function = parsed_function->function(); |
| 4057 const intptr_t specialization_cid = function.regexp_cid(); |
| 4058 const bool is_ascii = (specialization_cid == kOneByteStringCid || |
| 4059 specialization_cid == kExternalOneByteStringCid); |
| 4060 JSRegExp& regexp = JSRegExp::Handle(isolate, function.regexp()); |
| 4061 const String& pattern = String::Handle(isolate, regexp.pattern()); |
| 4062 |
| 4063 ASSERT(!regexp.IsNull()); |
| 4064 ASSERT(!pattern.IsNull()); |
| 4065 |
| 4066 const bool ignore_case = regexp.is_ignore_case(); |
| 4067 const bool is_global = regexp.is_global(); |
| 4068 |
| 4069 RegExpCompiler compiler(data->capture_count, ignore_case, specialization_cid); |
| 4070 |
| 4071 // TODO(jgruber): Frequency sampling is currently disabled because of several |
| 4072 // issues. We do not want to store subject strings in the regexp object since |
| 4073 // they might be long and we should not prevent their garbage collection. |
| 4074 // Passing them to this function explicitly does not help, since we must |
| 4075 // generate exactly the same IR for both the unoptimizing and optimizing |
| 4076 // pipelines (otherwise it gets confused when i.e. deopt id's differ). |
| 4077 // An option would be to store sampling results in the regexp object, but |
| 4078 // I'm not sure the performance gains are relevant enough. |
| 4079 |
| 4080 // Wrap the body of the regexp in capture #0. |
| 4081 RegExpNode* captured_body = RegExpCapture::ToNode(data->tree, |
| 4082 0, |
| 4083 &compiler, |
| 4084 compiler.accept()); |
| 4085 |
| 4086 RegExpNode* node = captured_body; |
| 4087 bool is_start_anchored = data->tree->IsAnchoredAtStart(); |
| 4088 bool is_end_anchored = data->tree->IsAnchoredAtEnd(); |
| 4089 intptr_t max_length = data->tree->max_match(); |
| 4090 if (!is_start_anchored) { |
| 4091 // Add a .*? at the beginning, outside the body capture, unless |
| 4092 // this expression is anchored at the beginning. |
| 4093 RegExpNode* loop_node = |
| 4094 RegExpQuantifier::ToNode(0, |
| 4095 RegExpTree::kInfinity, |
| 4096 false, |
| 4097 new(isolate) RegExpCharacterClass('*'), |
| 4098 &compiler, |
| 4099 captured_body, |
| 4100 data->contains_anchor); |
| 4101 |
| 4102 if (data->contains_anchor) { |
| 4103 // Unroll loop once, to take care of the case that might start |
| 4104 // at the start of input. |
| 4105 ChoiceNode* first_step_node = new(isolate) ChoiceNode(2, isolate); |
| 4106 first_step_node->AddAlternative(GuardedAlternative(captured_body)); |
| 4107 first_step_node->AddAlternative(GuardedAlternative( |
| 4108 new(isolate) TextNode( |
| 4109 new(isolate) RegExpCharacterClass('*'), loop_node))); |
| 4110 node = first_step_node; |
| 4111 } else { |
| 4112 node = loop_node; |
| 4113 } |
| 4114 } |
| 4115 if (is_ascii) { |
| 4116 node = node->FilterASCII(RegExpCompiler::kMaxRecursion, ignore_case); |
| 4117 // Do it again to propagate the new nodes to places where they were not |
| 4118 // put because they had not been calculated yet. |
| 4119 if (node != NULL) { |
| 4120 node = node->FilterASCII(RegExpCompiler::kMaxRecursion, ignore_case); |
| 4121 } |
| 4122 } |
| 4123 |
| 4124 if (node == NULL) node = new(isolate) EndNode(EndNode::BACKTRACK, isolate); |
| 4125 data->node = node; |
| 4126 Analysis analysis(ignore_case, is_ascii); |
| 4127 analysis.EnsureAnalyzed(node); |
| 4128 if (analysis.has_failed()) { |
| 4129 const char* error_message = analysis.error_message(); |
| 4130 return CompilationResult(error_message); |
| 4131 } |
| 4132 |
| 4133 // Native regexp implementation. |
| 4134 |
| 4135 IRRegExpMacroAssembler* macro_assembler = |
| 4136 new(isolate) IRRegExpMacroAssembler(specialization_cid, |
| 4137 data->capture_count, |
| 4138 parsed_function, |
| 4139 ic_data_array, |
| 4140 isolate); |
| 4141 |
| 4142 // Inserted here, instead of in Assembler, because it depends on information |
| 4143 // in the AST that isn't replicated in the Node structure. |
| 4144 static const intptr_t kMaxBacksearchLimit = 1024; |
| 4145 if (is_end_anchored && |
| 4146 !is_start_anchored && |
| 4147 max_length < kMaxBacksearchLimit) { |
| 4148 macro_assembler->SetCurrentPositionFromEnd(max_length); |
| 4149 } |
| 4150 |
| 4151 if (is_global) { |
| 4152 macro_assembler->set_global_mode( |
| 4153 (data->tree->min_match() > 0) |
| 4154 ? RegExpMacroAssembler::GLOBAL_NO_ZERO_LENGTH_CHECK |
| 4155 : RegExpMacroAssembler::GLOBAL); |
| 4156 } |
| 4157 |
| 4158 RegExpEngine::CompilationResult result = |
| 4159 compiler.Assemble(macro_assembler, |
| 4160 node, |
| 4161 data->capture_count, |
| 4162 pattern); |
| 4163 |
| 4164 if (FLAG_trace_irregexp) { |
| 4165 macro_assembler->PrintBlocks(); |
| 4166 } |
| 4167 |
| 4168 return result; |
| 4169 } |
| 4170 |
| 4171 |
| 4172 static void CreateSpecializedFunction(Isolate* isolate, |
| 4173 const JSRegExp& regexp, |
| 4174 intptr_t specialization_cid, |
| 4175 const Object& owner) { |
| 4176 const intptr_t kParamCount = RegExpMacroAssembler::kParamCount; |
| 4177 |
| 4178 Function& fn = Function::Handle(isolate, |
| 4179 Function::New(String::Handle(isolate, Symbols::New("RegExp")), |
| 4180 RawFunction::kIrregexpFunction, |
| 4181 true, // Static. |
| 4182 false, // Not const. |
| 4183 false, // Not abstract. |
| 4184 false, // Not external. |
| 4185 false, // Not native. |
| 4186 owner, |
| 4187 0)); // Requires a non-negative token position. |
| 4188 |
| 4189 // TODO(jgruber): Share these arrays between all irregexp functions. |
| 4190 fn.set_num_fixed_parameters(kParamCount); |
| 4191 fn.set_parameter_types(Array::Handle(isolate, Array::New(kParamCount, |
| 4192 Heap::kOld))); |
| 4193 fn.set_parameter_names(Array::Handle(isolate, Array::New(kParamCount, |
| 4194 Heap::kOld))); |
| 4195 fn.SetParameterTypeAt(0, Type::Handle(isolate, Type::DynamicType())); |
| 4196 fn.SetParameterNameAt(0, Symbols::string_param_()); |
| 4197 fn.SetParameterTypeAt(1, Type::Handle(isolate, Type::DynamicType())); |
| 4198 fn.SetParameterNameAt(1, Symbols::start_index_param_()); |
| 4199 fn.set_result_type(Type::Handle(isolate, Type::ArrayType())); |
| 4200 |
| 4201 // Cache the result. |
| 4202 regexp.set_function(specialization_cid, fn); |
| 4203 |
| 4204 fn.set_regexp(regexp); |
| 4205 fn.set_regexp_cid(specialization_cid); |
| 4206 |
| 4207 // The function is compiled lazily during the first call. |
| 4208 } |
| 4209 |
| 4210 |
| 4211 RawJSRegExp* RegExpEngine::CreateJSRegExp(Isolate* isolate, |
| 4212 const String& pattern, |
| 4213 bool multi_line, |
| 4214 bool ignore_case) { |
| 4215 const JSRegExp& regexp = JSRegExp::Handle(JSRegExp::New(0)); |
| 4216 |
| 4217 regexp.set_pattern(pattern); |
| 4218 |
| 4219 if (multi_line) { |
| 4220 regexp.set_is_multi_line(); |
| 4221 } |
| 4222 if (ignore_case) { |
| 4223 regexp.set_is_ignore_case(); |
| 4224 } |
| 4225 |
| 4226 // TODO(jgruber): We might want to use normal string searching algorithms |
| 4227 // for simple patterns. |
| 4228 regexp.set_is_complex(); |
| 4229 regexp.set_is_global(); // All dart regexps are global. |
| 4230 |
| 4231 const Library& lib = Library::Handle(isolate, Library::CoreLibrary()); |
| 4232 const Class& owner = Class::Handle(isolate, |
| 4233 lib.LookupClass(String::Handle(isolate, Symbols::New("RegExp")))); |
| 4234 |
| 4235 CreateSpecializedFunction(isolate, regexp, kOneByteStringCid, owner); |
| 4236 CreateSpecializedFunction(isolate, regexp, kTwoByteStringCid, owner); |
| 4237 CreateSpecializedFunction(isolate, regexp, kExternalOneByteStringCid, owner); |
| 4238 CreateSpecializedFunction(isolate, regexp, kExternalTwoByteStringCid, owner); |
| 4239 |
| 4240 return regexp.raw(); |
| 4241 } |
| 4242 |
| 4243 |
| 4244 void BoyerMoorePositionInfo::Set(intptr_t character) { |
| 4245 SetInterval(Interval(character, character)); |
| 4246 } |
| 4247 |
| 4248 |
| 4249 ContainedInLattice AddRange(ContainedInLattice containment, |
| 4250 const intptr_t* ranges, |
| 4251 intptr_t ranges_length, |
| 4252 Interval new_range) { |
| 4253 ASSERT((ranges_length & 1) == 1); |
| 4254 ASSERT(ranges[ranges_length - 1] == Utf16::kMaxCodeUnit + 1); |
| 4255 if (containment == kLatticeUnknown) return containment; |
| 4256 bool inside = false; |
| 4257 intptr_t last = 0; |
| 4258 for (intptr_t i = 0; i < ranges_length; |
| 4259 inside = !inside, last = ranges[i], i++) { |
| 4260 // Consider the range from last to ranges[i]. |
| 4261 // We haven't got to the new range yet. |
| 4262 if (ranges[i] <= new_range.from()) continue; |
| 4263 // New range is wholly inside last-ranges[i]. Note that new_range.to() is |
| 4264 // inclusive, but the values in ranges are not. |
| 4265 if (last <= new_range.from() && new_range.to() < ranges[i]) { |
| 4266 return Combine(containment, inside ? kLatticeIn : kLatticeOut); |
| 4267 } |
| 4268 return kLatticeUnknown; |
| 4269 } |
| 4270 return containment; |
| 4271 } |
| 4272 |
| 4273 |
| 4274 void BoyerMoorePositionInfo::SetInterval(const Interval& interval) { |
| 4275 s_ = AddRange(s_, kSpaceRanges, kSpaceRangeCount, interval); |
| 4276 w_ = AddRange(w_, kWordRanges, kWordRangeCount, interval); |
| 4277 d_ = AddRange(d_, kDigitRanges, kDigitRangeCount, interval); |
| 4278 surrogate_ = |
| 4279 AddRange(surrogate_, kSurrogateRanges, kSurrogateRangeCount, interval); |
| 4280 if (interval.to() - interval.from() >= kMapSize - 1) { |
| 4281 if (map_count_ != kMapSize) { |
| 4282 map_count_ = kMapSize; |
| 4283 for (intptr_t i = 0; i < kMapSize; i++) (*map_)[i] = true; |
| 4284 } |
| 4285 return; |
| 4286 } |
| 4287 for (intptr_t i = interval.from(); i <= interval.to(); i++) { |
| 4288 intptr_t mod_character = (i & kMask); |
| 4289 if (!map_->At(mod_character)) { |
| 4290 map_count_++; |
| 4291 (*map_)[mod_character] = true; |
| 4292 } |
| 4293 if (map_count_ == kMapSize) return; |
| 4294 } |
| 4295 } |
| 4296 |
| 4297 |
| 4298 void BoyerMoorePositionInfo::SetAll() { |
| 4299 s_ = w_ = d_ = kLatticeUnknown; |
| 4300 if (map_count_ != kMapSize) { |
| 4301 map_count_ = kMapSize; |
| 4302 for (intptr_t i = 0; i < kMapSize; i++) (*map_)[i] = true; |
| 4303 } |
| 4304 } |
| 4305 |
| 4306 |
| 4307 BoyerMooreLookahead::BoyerMooreLookahead( |
| 4308 intptr_t length, RegExpCompiler* compiler, Isolate* isolate) |
| 4309 : length_(length), |
| 4310 compiler_(compiler) { |
| 4311 if (compiler->ascii()) { |
| 4312 max_char_ = Symbols::kMaxOneCharCodeSymbol; |
| 4313 } else { |
| 4314 max_char_ = Utf16::kMaxCodeUnit; |
| 4315 } |
| 4316 bitmaps_ = new(isolate) ZoneGrowableArray<BoyerMoorePositionInfo*>(length); |
| 4317 for (intptr_t i = 0; i < length; i++) { |
| 4318 bitmaps_->Add(new(isolate) BoyerMoorePositionInfo(isolate)); |
| 4319 } |
| 4320 } |
| 4321 |
| 4322 |
| 4323 // Take all the characters that will not prevent a successful match if they |
| 4324 // occur in the subject string in the range between min_lookahead and |
| 4325 // max_lookahead (inclusive) measured from the current position. If the |
| 4326 // character at max_lookahead offset is not one of these characters, then we |
| 4327 // can safely skip forwards by the number of characters in the range. |
| 4328 intptr_t BoyerMooreLookahead::GetSkipTable( |
| 4329 intptr_t min_lookahead, |
| 4330 intptr_t max_lookahead, |
| 4331 const TypedData& boolean_skip_table) { |
| 4332 const intptr_t kSize = RegExpMacroAssembler::kTableSize; |
| 4333 |
| 4334 const intptr_t kSkipArrayEntry = 0; |
| 4335 const intptr_t kDontSkipArrayEntry = 1; |
| 4336 |
| 4337 for (intptr_t i = 0; i < kSize; i++) { |
| 4338 boolean_skip_table.SetUint8(i, kSkipArrayEntry); |
| 4339 } |
| 4340 intptr_t skip = max_lookahead + 1 - min_lookahead; |
| 4341 |
| 4342 for (intptr_t i = max_lookahead; i >= min_lookahead; i--) { |
| 4343 BoyerMoorePositionInfo* map = bitmaps_->At(i); |
| 4344 for (intptr_t j = 0; j < kSize; j++) { |
| 4345 if (map->at(j)) { |
| 4346 boolean_skip_table.SetUint8(j, kDontSkipArrayEntry); |
| 4347 } |
| 4348 } |
| 4349 } |
| 4350 |
| 4351 return skip; |
| 4352 } |
| 4353 |
| 4354 |
| 4355 // Find the longest range of lookahead that has the fewest number of different |
| 4356 // characters that can occur at a given position. Since we are optimizing two |
| 4357 // different parameters at once this is a tradeoff. |
| 4358 bool BoyerMooreLookahead::FindWorthwhileInterval(intptr_t* from, intptr_t* to) { |
| 4359 intptr_t biggest_points = 0; |
| 4360 // If more than 32 characters out of 128 can occur it is unlikely that we can |
| 4361 // be lucky enough to step forwards much of the time. |
| 4362 const intptr_t kMaxMax = 32; |
| 4363 for (intptr_t max_number_of_chars = 4; |
| 4364 max_number_of_chars < kMaxMax; |
| 4365 max_number_of_chars *= 2) { |
| 4366 biggest_points = |
| 4367 FindBestInterval(max_number_of_chars, biggest_points, from, to); |
| 4368 } |
| 4369 if (biggest_points == 0) return false; |
| 4370 return true; |
| 4371 } |
| 4372 |
| 4373 |
| 4374 // Find the highest-points range between 0 and length_ where the character |
| 4375 // information is not too vague. 'Too vague' means that there are more than |
| 4376 // max_number_of_chars that can occur at this position. Calculates the number |
| 4377 // of points as the product of width-of-the-range and |
| 4378 // probability-of-finding-one-of-the-characters, where the probability is |
| 4379 // calculated using the frequency distribution of the sample subject string. |
| 4380 intptr_t BoyerMooreLookahead::FindBestInterval( |
| 4381 intptr_t max_number_of_chars, |
| 4382 intptr_t old_biggest_points, |
| 4383 intptr_t* from, |
| 4384 intptr_t* to) { |
| 4385 intptr_t biggest_points = old_biggest_points; |
| 4386 static const intptr_t kSize = RegExpMacroAssembler::kTableSize; |
| 4387 for (intptr_t i = 0; i < length_; ) { |
| 4388 while (i < length_ && Count(i) > max_number_of_chars) i++; |
| 4389 if (i == length_) break; |
| 4390 intptr_t remembered_from = i; |
| 4391 bool union_map[kSize]; |
| 4392 for (intptr_t j = 0; j < kSize; j++) union_map[j] = false; |
| 4393 while (i < length_ && Count(i) <= max_number_of_chars) { |
| 4394 BoyerMoorePositionInfo* map = bitmaps_->At(i); |
| 4395 for (intptr_t j = 0; j < kSize; j++) union_map[j] |= map->at(j); |
| 4396 i++; |
| 4397 } |
| 4398 intptr_t frequency = 0; |
| 4399 for (intptr_t j = 0; j < kSize; j++) { |
| 4400 if (union_map[j]) { |
| 4401 // Add 1 to the frequency to give a small per-character boost for |
| 4402 // the cases where our sampling is not good enough and many |
| 4403 // characters have a frequency of zero. This means the frequency |
| 4404 // can theoretically be up to 2*kSize though we treat it mostly as |
| 4405 // a fraction of kSize. |
| 4406 frequency += compiler_->frequency_collator()->Frequency(j) + 1; |
| 4407 } |
| 4408 } |
| 4409 // We use the probability of skipping times the distance we are skipping to |
| 4410 // judge the effectiveness of this. Actually we have a cut-off: By |
| 4411 // dividing by 2 we switch off the skipping if the probability of skipping |
| 4412 // is less than 50%. This is because the multibyte mask-and-compare |
| 4413 // skipping in quickcheck is more likely to do well on this case. |
| 4414 bool in_quickcheck_range = ((i - remembered_from < 4) || |
| 4415 (compiler_->ascii() ? remembered_from <= 4 : remembered_from <= 2)); |
| 4416 // Called 'probability' but it is only a rough estimate and can actually |
| 4417 // be outside the 0-kSize range. |
| 4418 intptr_t probability = |
| 4419 (in_quickcheck_range ? kSize / 2 : kSize) - frequency; |
| 4420 intptr_t points = (i - remembered_from) * probability; |
| 4421 if (points > biggest_points) { |
| 4422 *from = remembered_from; |
| 4423 *to = i - 1; |
| 4424 biggest_points = points; |
| 4425 } |
| 4426 } |
| 4427 return biggest_points; |
| 4428 } |
| 4429 |
| 4430 |
| 4431 // See comment above on the implementation of GetSkipTable. |
| 4432 bool BoyerMooreLookahead::EmitSkipInstructions(RegExpMacroAssembler* masm) { |
| 4433 const intptr_t kSize = RegExpMacroAssembler::kTableSize; |
| 4434 |
| 4435 intptr_t min_lookahead = 0; |
| 4436 intptr_t max_lookahead = 0; |
| 4437 |
| 4438 if (!FindWorthwhileInterval(&min_lookahead, &max_lookahead)) return false; |
| 4439 |
| 4440 bool found_single_character = false; |
| 4441 intptr_t single_character = 0; |
| 4442 for (intptr_t i = max_lookahead; i >= min_lookahead; i--) { |
| 4443 BoyerMoorePositionInfo* map = bitmaps_->At(i); |
| 4444 if (map->map_count() > 1 || |
| 4445 (found_single_character && map->map_count() != 0)) { |
| 4446 found_single_character = false; |
| 4447 break; |
| 4448 } |
| 4449 for (intptr_t j = 0; j < kSize; j++) { |
| 4450 if (map->at(j)) { |
| 4451 found_single_character = true; |
| 4452 single_character = j; |
| 4453 break; |
| 4454 } |
| 4455 } |
| 4456 } |
| 4457 |
| 4458 intptr_t lookahead_width = max_lookahead + 1 - min_lookahead; |
| 4459 |
| 4460 if (found_single_character && lookahead_width == 1 && max_lookahead < 3) { |
| 4461 // The mask-compare can probably handle this better. |
| 4462 return false; |
| 4463 } |
| 4464 |
| 4465 if (found_single_character) { |
| 4466 BlockLabel cont, again; |
| 4467 masm->BindBlock(&again); |
| 4468 masm->LoadCurrentCharacter(max_lookahead, &cont, true); |
| 4469 if (max_char_ > kSize) { |
| 4470 masm->CheckCharacterAfterAnd(single_character, |
| 4471 RegExpMacroAssembler::kTableMask, |
| 4472 &cont); |
| 4473 } else { |
| 4474 masm->CheckCharacter(single_character, &cont); |
| 4475 } |
| 4476 masm->AdvanceCurrentPosition(lookahead_width); |
| 4477 masm->GoTo(&again); |
| 4478 masm->BindBlock(&cont); |
| 4479 return true; |
| 4480 } |
| 4481 |
| 4482 const TypedData& boolean_skip_table = TypedData::ZoneHandle( |
| 4483 compiler_->isolate(), |
| 4484 TypedData::New(kTypedDataUint8ArrayCid, kSize, Heap::kOld)); |
| 4485 intptr_t skip_distance = GetSkipTable( |
| 4486 min_lookahead, max_lookahead, boolean_skip_table); |
| 4487 ASSERT(skip_distance != 0); |
| 4488 |
| 4489 BlockLabel cont, again; |
| 4490 |
| 4491 masm->BindBlock(&again); |
| 4492 masm->LoadCurrentCharacter(max_lookahead, &cont, true); |
| 4493 masm->CheckBitInTable(boolean_skip_table, &cont); |
| 4494 masm->AdvanceCurrentPosition(skip_distance); |
| 4495 masm->GoTo(&again); |
| 4496 masm->BindBlock(&cont); |
| 4497 |
| 4498 return true; |
| 4499 } |
| 4500 |
| 4501 |
| 4502 // ------------------------------------------------------------------- |
| 4503 // Analysis |
| 4504 |
| 4505 |
| 4506 void Analysis::EnsureAnalyzed(RegExpNode* that) { |
| 4507 if (that->info()->been_analyzed || that->info()->being_analyzed) |
| 4508 return; |
| 4509 that->info()->being_analyzed = true; |
| 4510 that->Accept(this); |
| 4511 that->info()->being_analyzed = false; |
| 4512 that->info()->been_analyzed = true; |
| 4513 } |
| 4514 |
| 4515 |
| 4516 void Analysis::VisitEnd(EndNode* that) { |
| 4517 // nothing to do |
| 4518 } |
| 4519 |
| 4520 |
| 4521 void TextNode::CalculateOffsets() { |
| 4522 intptr_t element_count = elements()->length(); |
| 4523 // Set up the offsets of the elements relative to the start. This is a fixed |
| 4524 // quantity since a TextNode can only contain fixed-width things. |
| 4525 intptr_t cp_offset = 0; |
| 4526 for (intptr_t i = 0; i < element_count; i++) { |
| 4527 TextElement& elm = (*elements())[i]; |
| 4528 elm.set_cp_offset(cp_offset); |
| 4529 cp_offset += elm.length(); |
| 4530 } |
| 4531 } |
| 4532 |
| 4533 |
| 4534 void Analysis::VisitText(TextNode* that) { |
| 4535 if (ignore_case_) { |
| 4536 that->MakeCaseIndependent(is_ascii_); |
| 4537 } |
| 4538 EnsureAnalyzed(that->on_success()); |
| 4539 if (!has_failed()) { |
| 4540 that->CalculateOffsets(); |
| 4541 } |
| 4542 } |
| 4543 |
| 4544 |
| 4545 void Analysis::VisitAction(ActionNode* that) { |
| 4546 RegExpNode* target = that->on_success(); |
| 4547 EnsureAnalyzed(target); |
| 4548 if (!has_failed()) { |
| 4549 // If the next node is interested in what it follows then this node |
| 4550 // has to be interested too so it can pass the information on. |
| 4551 that->info()->AddFromFollowing(target->info()); |
| 4552 } |
| 4553 } |
| 4554 |
| 4555 |
| 4556 void Analysis::VisitChoice(ChoiceNode* that) { |
| 4557 NodeInfo* info = that->info(); |
| 4558 for (intptr_t i = 0; i < that->alternatives()->length(); i++) { |
| 4559 RegExpNode* node = (*that->alternatives())[i].node(); |
| 4560 EnsureAnalyzed(node); |
| 4561 if (has_failed()) return; |
| 4562 // Anything the following nodes need to know has to be known by |
| 4563 // this node also, so it can pass it on. |
| 4564 info->AddFromFollowing(node->info()); |
| 4565 } |
| 4566 } |
| 4567 |
| 4568 |
| 4569 void Analysis::VisitLoopChoice(LoopChoiceNode* that) { |
| 4570 NodeInfo* info = that->info(); |
| 4571 for (intptr_t i = 0; i < that->alternatives()->length(); i++) { |
| 4572 RegExpNode* node = (*that->alternatives())[i].node(); |
| 4573 if (node != that->loop_node()) { |
| 4574 EnsureAnalyzed(node); |
| 4575 if (has_failed()) return; |
| 4576 info->AddFromFollowing(node->info()); |
| 4577 } |
| 4578 } |
| 4579 // Check the loop last since it may need the value of this node |
| 4580 // to get a correct result. |
| 4581 EnsureAnalyzed(that->loop_node()); |
| 4582 if (!has_failed()) { |
| 4583 info->AddFromFollowing(that->loop_node()->info()); |
| 4584 } |
| 4585 } |
| 4586 |
| 4587 |
| 4588 void Analysis::VisitBackReference(BackReferenceNode* that) { |
| 4589 EnsureAnalyzed(that->on_success()); |
| 4590 } |
| 4591 |
| 4592 |
| 4593 void Analysis::VisitAssertion(AssertionNode* that) { |
| 4594 EnsureAnalyzed(that->on_success()); |
| 4595 } |
| 4596 |
| 4597 |
| 4598 // ------------------------------------------------------------------- |
| 4599 // Dot/dotty output |
| 4600 |
| 4601 |
| 4602 #ifdef DEBUG |
| 4603 |
| 4604 |
| 4605 class DotPrinter: public NodeVisitor { |
| 4606 public: |
| 4607 explicit DotPrinter(bool ignore_case) |
| 4608 : ignore_case_(ignore_case) {} |
| 4609 void PrintNode(const char* label, RegExpNode* node); |
| 4610 void Visit(RegExpNode* node); |
| 4611 void PrintAttributes(RegExpNode* from); |
| 4612 void PrintOnFailure(RegExpNode* from, RegExpNode* to); |
| 4613 #define DECLARE_VISIT(Type) \ |
| 4614 virtual void Visit##Type(Type##Node* that); |
| 4615 FOR_EACH_NODE_TYPE(DECLARE_VISIT) |
| 4616 #undef DECLARE_VISIT |
| 4617 private: |
| 4618 bool ignore_case_; |
| 4619 }; |
| 4620 |
| 4621 |
| 4622 void DotPrinter::PrintNode(const char* label, RegExpNode* node) { |
| 4623 OS::Print("digraph G {\n graph [label=\""); |
| 4624 for (intptr_t i = 0; label[i]; i++) { |
| 4625 switch (label[i]) { |
| 4626 case '\\': |
| 4627 OS::Print("\\\\"); |
| 4628 break; |
| 4629 case '"': |
| 4630 OS::Print("\""); |
| 4631 break; |
| 4632 default: |
| 4633 OS::Print("%c", label[i]); |
| 4634 break; |
| 4635 } |
| 4636 } |
| 4637 OS::Print("\"];\n"); |
| 4638 Visit(node); |
| 4639 OS::Print("}\n"); |
| 4640 } |
| 4641 |
| 4642 |
| 4643 void DotPrinter::Visit(RegExpNode* node) { |
| 4644 if (node->info()->visited) return; |
| 4645 node->info()->visited = true; |
| 4646 node->Accept(this); |
| 4647 } |
| 4648 |
| 4649 |
| 4650 void DotPrinter::PrintOnFailure(RegExpNode* from, RegExpNode* on_failure) { |
| 4651 OS::Print(" n%p -> n%p [style=dotted];\n", from, on_failure); |
| 4652 Visit(on_failure); |
| 4653 } |
| 4654 |
| 4655 |
| 4656 class AttributePrinter : public ValueObject { |
| 4657 public: |
| 4658 AttributePrinter() : first_(true) {} |
| 4659 void PrintSeparator() { |
| 4660 if (first_) { |
| 4661 first_ = false; |
| 4662 } else { |
| 4663 OS::Print("|"); |
| 4664 } |
| 4665 } |
| 4666 void PrintBit(const char* name, bool value) { |
| 4667 if (!value) return; |
| 4668 PrintSeparator(); |
| 4669 OS::Print("{%s}", name); |
| 4670 } |
| 4671 void PrintPositive(const char* name, intptr_t value) { |
| 4672 if (value < 0) return; |
| 4673 PrintSeparator(); |
| 4674 OS::Print("{%s|%" Pd "}", name, value); |
| 4675 } |
| 4676 |
| 4677 private: |
| 4678 bool first_; |
| 4679 }; |
| 4680 |
| 4681 |
| 4682 void DotPrinter::PrintAttributes(RegExpNode* that) { |
| 4683 OS::Print(" a%p [shape=Mrecord, color=grey, fontcolor=grey, " |
| 4684 "margin=0.1, fontsize=10, label=\"{", that); |
| 4685 AttributePrinter printer; |
| 4686 NodeInfo* info = that->info(); |
| 4687 printer.PrintBit("NI", info->follows_newline_interest); |
| 4688 printer.PrintBit("WI", info->follows_word_interest); |
| 4689 printer.PrintBit("SI", info->follows_start_interest); |
| 4690 BlockLabel* label = that->label(); |
| 4691 if (label->IsBound()) |
| 4692 printer.PrintPositive("@", label->Position()); |
| 4693 OS::Print("}\"];\n" |
| 4694 " a%p -> n%p [style=dashed, color=grey, arrowhead=none];\n", |
| 4695 that, that); |
| 4696 } |
| 4697 |
| 4698 |
| 4699 static const bool kPrintDispatchTable = false; |
| 4700 void DotPrinter::VisitChoice(ChoiceNode* that) { |
| 4701 if (kPrintDispatchTable) { |
| 4702 OS::Print(" n%p [shape=Mrecord, label=\"", that); |
| 4703 UNIMPLEMENTED(); |
| 4704 } else { |
| 4705 OS::Print(" n%p [shape=Mrecord, label=\"?\"];\n", that); |
| 4706 for (intptr_t i = 0; i < that->alternatives()->length(); i++) { |
| 4707 GuardedAlternative alt = that->alternatives()->At(i); |
| 4708 OS::Print(" n%p -> n%p", that, alt.node()); |
| 4709 } |
| 4710 } |
| 4711 for (intptr_t i = 0; i < that->alternatives()->length(); i++) { |
| 4712 GuardedAlternative alt = that->alternatives()->At(i); |
| 4713 alt.node()->Accept(this); |
| 4714 } |
| 4715 } |
| 4716 |
| 4717 |
| 4718 void DotPrinter::VisitText(TextNode* that) { |
| 4719 OS::Print(" n%p [label=\"", that); |
| 4720 for (intptr_t i = 0; i < that->elements()->length(); i++) { |
| 4721 if (i > 0) OS::Print(" "); |
| 4722 TextElement elm = that->elements()->At(i); |
| 4723 switch (elm.text_type()) { |
| 4724 case TextElement::ATOM: { |
| 4725 ZoneGrowableArray<uint16_t>* data = elm.atom()->data(); |
| 4726 for (intptr_t i = 0; i < data->length(); i++) { |
| 4727 OS::Print("%c", static_cast<char>(data->At(i))); |
| 4728 } |
| 4729 break; |
| 4730 } |
| 4731 case TextElement::CHAR_CLASS: { |
| 4732 RegExpCharacterClass* node = elm.char_class(); |
| 4733 OS::Print("["); |
| 4734 if (node->is_negated()) OS::Print("^"); |
| 4735 for (intptr_t j = 0; j < node->ranges()->length(); j++) { |
| 4736 CharacterRange range = node->ranges()->At(j); |
| 4737 PrintUtf16(range.from()); |
| 4738 OS::Print("-"); |
| 4739 PrintUtf16(range.to()); |
| 4740 } |
| 4741 OS::Print("]"); |
| 4742 break; |
| 4743 } |
| 4744 default: |
| 4745 UNREACHABLE(); |
| 4746 } |
| 4747 } |
| 4748 OS::Print("\", shape=box, peripheries=2];\n"); |
| 4749 PrintAttributes(that); |
| 4750 OS::Print(" n%p -> n%p;\n", that, that->on_success()); |
| 4751 Visit(that->on_success()); |
| 4752 } |
| 4753 |
| 4754 |
| 4755 void DotPrinter::VisitBackReference(BackReferenceNode* that) { |
| 4756 OS::Print(" n%p [label=\"$%" Pd "..$%" Pd "\", shape=doubleoctagon];\n", |
| 4757 that, that->start_register(), that->end_register()); |
| 4758 PrintAttributes(that); |
| 4759 OS::Print(" n%p -> n%p;\n", that, that->on_success()); |
| 4760 Visit(that->on_success()); |
| 4761 } |
| 4762 |
| 4763 |
| 4764 void DotPrinter::VisitEnd(EndNode* that) { |
| 4765 OS::Print(" n%p [style=bold, shape=point];\n", that); |
| 4766 PrintAttributes(that); |
| 4767 } |
| 4768 |
| 4769 |
| 4770 void DotPrinter::VisitAssertion(AssertionNode* that) { |
| 4771 OS::Print(" n%p [", that); |
| 4772 switch (that->assertion_type()) { |
| 4773 case AssertionNode::AT_END: |
| 4774 OS::Print("label=\"$\", shape=septagon"); |
| 4775 break; |
| 4776 case AssertionNode::AT_START: |
| 4777 OS::Print("label=\"^\", shape=septagon"); |
| 4778 break; |
| 4779 case AssertionNode::AT_BOUNDARY: |
| 4780 OS::Print("label=\"\\b\", shape=septagon"); |
| 4781 break; |
| 4782 case AssertionNode::AT_NON_BOUNDARY: |
| 4783 OS::Print("label=\"\\B\", shape=septagon"); |
| 4784 break; |
| 4785 case AssertionNode::AFTER_NEWLINE: |
| 4786 OS::Print("label=\"(?<=\\n)\", shape=septagon"); |
| 4787 break; |
| 4788 } |
| 4789 OS::Print("];\n"); |
| 4790 PrintAttributes(that); |
| 4791 RegExpNode* successor = that->on_success(); |
| 4792 OS::Print(" n%p -> n%p;\n", that, successor); |
| 4793 Visit(successor); |
| 4794 } |
| 4795 |
| 4796 |
| 4797 void DotPrinter::VisitAction(ActionNode* that) { |
| 4798 OS::Print(" n%p [", that); |
| 4799 switch (that->action_type_) { |
| 4800 case ActionNode::SET_REGISTER: |
| 4801 OS::Print("label=\"$%" Pd ":=%" Pd "\", shape=octagon", |
| 4802 that->data_.u_store_register.reg, |
| 4803 that->data_.u_store_register.value); |
| 4804 break; |
| 4805 case ActionNode::INCREMENT_REGISTER: |
| 4806 OS::Print("label=\"$%" Pd "++\", shape=octagon", |
| 4807 that->data_.u_increment_register.reg); |
| 4808 break; |
| 4809 case ActionNode::STORE_POSITION: |
| 4810 OS::Print("label=\"$%" Pd ":=$pos\", shape=octagon", |
| 4811 that->data_.u_position_register.reg); |
| 4812 break; |
| 4813 case ActionNode::BEGIN_SUBMATCH: |
| 4814 OS::Print("label=\"$%" Pd ":=$pos,begin\", shape=septagon", |
| 4815 that->data_.u_submatch.current_position_register); |
| 4816 break; |
| 4817 case ActionNode::POSITIVE_SUBMATCH_SUCCESS: |
| 4818 OS::Print("label=\"escape\", shape=septagon"); |
| 4819 break; |
| 4820 case ActionNode::EMPTY_MATCH_CHECK: |
| 4821 OS::Print("label=\"$%" Pd "=$pos?,$%" Pd "<%" Pd "?\", shape=septagon", |
| 4822 that->data_.u_empty_match_check.start_register, |
| 4823 that->data_.u_empty_match_check.repetition_register, |
| 4824 that->data_.u_empty_match_check.repetition_limit); |
| 4825 break; |
| 4826 case ActionNode::CLEAR_CAPTURES: { |
| 4827 OS::Print("label=\"clear $%" Pd " to $%" Pd "\", shape=septagon", |
| 4828 that->data_.u_clear_captures.range_from, |
| 4829 that->data_.u_clear_captures.range_to); |
| 4830 break; |
| 4831 } |
| 4832 } |
| 4833 OS::Print("];\n"); |
| 4834 PrintAttributes(that); |
| 4835 RegExpNode* successor = that->on_success(); |
| 4836 OS::Print(" n%p -> n%p;\n", that, successor); |
| 4837 Visit(successor); |
| 4838 } |
| 4839 |
| 4840 |
| 4841 void RegExpEngine::DotPrint(const char* label, |
| 4842 RegExpNode* node, |
| 4843 bool ignore_case) { |
| 4844 DotPrinter printer(ignore_case); |
| 4845 printer.PrintNode(label, node); |
| 4846 } |
| 4847 |
| 4848 |
| 4849 #endif // DEBUG |
| 4850 |
| 4851 } // namespace dart |
OLD | NEW |