regexp-compiler.cc 146 KB
Newer Older
1 2 3 4 5 6
// Copyright 2019 the V8 project authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

#include "src/regexp/regexp-compiler.h"

7
#include "src/base/safe_conversions.h"
8 9 10 11
#include "src/execution/isolate.h"
#include "src/objects/objects-inl.h"
#include "src/regexp/regexp-macro-assembler-arch.h"
#include "src/regexp/regexp-macro-assembler-tracer.h"
12
#include "src/strings/unicode-inl.h"
13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273
#include "src/zone/zone-list-inl.h"

#ifdef V8_INTL_SUPPORT
#include "unicode/locid.h"
#include "unicode/uniset.h"
#include "unicode/utypes.h"
#endif  // V8_INTL_SUPPORT

namespace v8 {
namespace internal {

using namespace regexp_compiler_constants;  // NOLINT(build/namespaces)

// -------------------------------------------------------------------
// Implementation of the Irregexp regular expression engine.
//
// The Irregexp regular expression engine is intended to be a complete
// implementation of ECMAScript regular expressions.  It generates either
// bytecodes or native code.

//   The Irregexp regexp engine is structured in three steps.
//   1) The parser generates an abstract syntax tree.  See ast.cc.
//   2) From the AST a node network is created.  The nodes are all
//      subclasses of RegExpNode.  The nodes represent states when
//      executing a regular expression.  Several optimizations are
//      performed on the node network.
//   3) From the nodes we generate either byte codes or native code
//      that can actually execute the regular expression (perform
//      the search).  The code generation step is described in more
//      detail below.

// Code generation.
//
//   The nodes are divided into four main categories.
//   * Choice nodes
//        These represent places where the regular expression can
//        match in more than one way.  For example on entry to an
//        alternation (foo|bar) or a repetition (*, +, ? or {}).
//   * Action nodes
//        These represent places where some action should be
//        performed.  Examples include recording the current position
//        in the input string to a register (in order to implement
//        captures) or other actions on register for example in order
//        to implement the counters needed for {} repetitions.
//   * Matching nodes
//        These attempt to match some element part of the input string.
//        Examples of elements include character classes, plain strings
//        or back references.
//   * End nodes
//        These are used to implement the actions required on finding
//        a successful match or failing to find a match.
//
//   The code generated (whether as byte codes or native code) maintains
//   some state as it runs.  This consists of the following elements:
//
//   * The capture registers.  Used for string captures.
//   * Other registers.  Used for counters etc.
//   * The current position.
//   * The stack of backtracking information.  Used when a matching node
//     fails to find a match and needs to try an alternative.
//
// Conceptual regular expression execution model:
//
//   There is a simple conceptual model of regular expression execution
//   which will be presented first.  The actual code generated is a more
//   efficient simulation of the simple conceptual model:
//
//   * Choice nodes are implemented as follows:
//     For each choice except the last {
//       push current position
//       push backtrack code location
//       <generate code to test for choice>
//       backtrack code location:
//       pop current position
//     }
//     <generate code to test for last choice>
//
//   * Actions nodes are generated as follows
//     <push affected registers on backtrack stack>
//     <generate code to perform action>
//     push backtrack code location
//     <generate code to test for following nodes>
//     backtrack code location:
//     <pop affected registers to restore their state>
//     <pop backtrack location from stack and go to it>
//
//   * Matching nodes are generated as follows:
//     if input string matches at current position
//       update current position
//       <generate code to test for following nodes>
//     else
//       <pop backtrack location from stack and go to it>
//
//   Thus it can be seen that the current position is saved and restored
//   by the choice nodes, whereas the registers are saved and restored by
//   by the action nodes that manipulate them.
//
//   The other interesting aspect of this model is that nodes are generated
//   at the point where they are needed by a recursive call to Emit().  If
//   the node has already been code generated then the Emit() call will
//   generate a jump to the previously generated code instead.  In order to
//   limit recursion it is possible for the Emit() function to put the node
//   on a work list for later generation and instead generate a jump.  The
//   destination of the jump is resolved later when the code is generated.
//
// Actual regular expression code generation.
//
//   Code generation is actually more complicated than the above.  In order
//   to improve the efficiency of the generated code some optimizations are
//   performed
//
//   * Choice nodes have 1-character lookahead.
//     A choice node looks at the following character and eliminates some of
//     the choices immediately based on that character.  This is not yet
//     implemented.
//   * Simple greedy loops store reduced backtracking information.
//     A quantifier like /.*foo/m will greedily match the whole input.  It will
//     then need to backtrack to a point where it can match "foo".  The naive
//     implementation of this would push each character position onto the
//     backtracking stack, then pop them off one by one.  This would use space
//     proportional to the length of the input string.  However since the "."
//     can only match in one way and always has a constant length (in this case
//     of 1) it suffices to store the current position on the top of the stack
//     once.  Matching now becomes merely incrementing the current position and
//     backtracking becomes decrementing the current position and checking the
//     result against the stored current position.  This is faster and saves
//     space.
//   * The current state is virtualized.
//     This is used to defer expensive operations until it is clear that they
//     are needed and to generate code for a node more than once, allowing
//     specialized an efficient versions of the code to be created. This is
//     explained in the section below.
//
// Execution state virtualization.
//
//   Instead of emitting code, nodes that manipulate the state can record their
//   manipulation in an object called the Trace.  The Trace object can record a
//   current position offset, an optional backtrack code location on the top of
//   the virtualized backtrack stack and some register changes.  When a node is
//   to be emitted it can flush the Trace or update it.  Flushing the Trace
//   will emit code to bring the actual state into line with the virtual state.
//   Avoiding flushing the state can postpone some work (e.g. updates of capture
//   registers).  Postponing work can save time when executing the regular
//   expression since it may be found that the work never has to be done as a
//   failure to match can occur.  In addition it is much faster to jump to a
//   known backtrack code location than it is to pop an unknown backtrack
//   location from the stack and jump there.
//
//   The virtual state found in the Trace affects code generation.  For example
//   the virtual state contains the difference between the actual current
//   position and the virtual current position, and matching code needs to use
//   this offset to attempt a match in the correct location of the input
//   string.  Therefore code generated for a non-trivial trace is specialized
//   to that trace.  The code generator therefore has the ability to generate
//   code for each node several times.  In order to limit the size of the
//   generated code there is an arbitrary limit on how many specialized sets of
//   code may be generated for a given node.  If the limit is reached, the
//   trace is flushed and a generic version of the code for a node is emitted.
//   This is subsequently used for that node.  The code emitted for non-generic
//   trace is not recorded in the node and so it cannot currently be reused in
//   the event that code generation is requested for an identical trace.

void RegExpTree::AppendToText(RegExpText* text, Zone* zone) { UNREACHABLE(); }

void RegExpAtom::AppendToText(RegExpText* text, Zone* zone) {
  text->AddElement(TextElement::Atom(this), zone);
}

void RegExpCharacterClass::AppendToText(RegExpText* text, Zone* zone) {
  text->AddElement(TextElement::CharClass(this), zone);
}

void RegExpText::AppendToText(RegExpText* text, Zone* zone) {
  for (int i = 0; i < elements()->length(); i++)
    text->AddElement(elements()->at(i), zone);
}

TextElement TextElement::Atom(RegExpAtom* atom) {
  return TextElement(ATOM, atom);
}

TextElement TextElement::CharClass(RegExpCharacterClass* char_class) {
  return TextElement(CHAR_CLASS, char_class);
}

int TextElement::length() const {
  switch (text_type()) {
    case ATOM:
      return atom()->length();

    case CHAR_CLASS:
      return 1;
  }
  UNREACHABLE();
}

class RecursionCheck {
 public:
  explicit RecursionCheck(RegExpCompiler* compiler) : compiler_(compiler) {
    compiler->IncrementRecursionDepth();
  }
  ~RecursionCheck() { compiler_->DecrementRecursionDepth(); }

 private:
  RegExpCompiler* compiler_;
};

// Attempts to compile the regexp using an Irregexp code generator.  Returns
// a fixed array or a null handle depending on whether it succeeded.
RegExpCompiler::RegExpCompiler(Isolate* isolate, Zone* zone, int capture_count,
                               bool one_byte)
    : next_register_(2 * (capture_count + 1)),
      unicode_lookaround_stack_register_(kNoRegister),
      unicode_lookaround_position_register_(kNoRegister),
      work_list_(nullptr),
      recursion_depth_(0),
      one_byte_(one_byte),
      reg_exp_too_big_(false),
      limiting_recursion_(false),
      optimize_(FLAG_regexp_optimization),
      read_backward_(false),
      current_expansion_factor_(1),
      frequency_collator_(),
      isolate_(isolate),
      zone_(zone) {
  accept_ = new (zone) EndNode(EndNode::ACCEPT, zone);
  DCHECK_GE(RegExpMacroAssembler::kMaxRegister, next_register_ - 1);
}

RegExpCompiler::CompilationResult RegExpCompiler::Assemble(
    Isolate* isolate, RegExpMacroAssembler* macro_assembler, RegExpNode* start,
    int capture_count, Handle<String> pattern) {
#ifdef DEBUG
  if (FLAG_trace_regexp_assembler)
    macro_assembler_ = new RegExpMacroAssemblerTracer(isolate, macro_assembler);
  else
#endif
    macro_assembler_ = macro_assembler;

  std::vector<RegExpNode*> work_list;
  work_list_ = &work_list;
  Label fail;
  macro_assembler_->PushBacktrack(&fail);
  Trace new_trace;
  start->Emit(this, &new_trace);
  macro_assembler_->Bind(&fail);
  macro_assembler_->Fail();
  while (!work_list.empty()) {
    RegExpNode* node = work_list.back();
    work_list.pop_back();
    node->set_on_work_list(false);
    if (!node->label()->is_bound()) node->Emit(this, &new_trace);
  }
  if (reg_exp_too_big_) {
    macro_assembler_->AbortedCodeGeneration();
    return CompilationResult::RegExpTooBig();
  }

  Handle<HeapObject> code = macro_assembler_->GetCode(pattern);
  isolate->IncreaseTotalRegexpCodeGenerated(code->Size());
  work_list_ = nullptr;
274

275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315
#ifdef DEBUG
  if (FLAG_trace_regexp_assembler) {
    delete macro_assembler_;
  }
#endif
  return {*code, next_register_};
}

bool Trace::DeferredAction::Mentions(int that) {
  if (action_type() == ActionNode::CLEAR_CAPTURES) {
    Interval range = static_cast<DeferredClearCaptures*>(this)->range();
    return range.Contains(that);
  } else {
    return reg() == that;
  }
}

bool Trace::mentions_reg(int reg) {
  for (DeferredAction* action = actions_; action != nullptr;
       action = action->next()) {
    if (action->Mentions(reg)) return true;
  }
  return false;
}

bool Trace::GetStoredPosition(int reg, int* cp_offset) {
  DCHECK_EQ(0, *cp_offset);
  for (DeferredAction* action = actions_; action != nullptr;
       action = action->next()) {
    if (action->Mentions(reg)) {
      if (action->action_type() == ActionNode::STORE_POSITION) {
        *cp_offset = static_cast<DeferredCapture*>(action)->cp_offset();
        return true;
      } else {
        return false;
      }
    }
  }
  return false;
}

316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350
// A (dynamically-sized) set of unsigned integers that behaves especially well
// on small integers (< kFirstLimit). May do zone-allocation.
class DynamicBitSet : public ZoneObject {
 public:
  V8_EXPORT_PRIVATE bool Get(unsigned value) const {
    if (value < kFirstLimit) {
      return (first_ & (1 << value)) != 0;
    } else if (remaining_ == nullptr) {
      return false;
    } else {
      return remaining_->Contains(value);
    }
  }

  // Destructively set a value in this set.
  void Set(unsigned value, Zone* zone) {
    if (value < kFirstLimit) {
      first_ |= (1 << value);
    } else {
      if (remaining_ == nullptr)
        remaining_ = new (zone) ZoneList<unsigned>(1, zone);
      if (remaining_->is_empty() || !remaining_->Contains(value))
        remaining_->Add(value, zone);
    }
  }

 private:
  static constexpr unsigned kFirstLimit = 32;

  uint32_t first_ = 0;
  ZoneList<unsigned>* remaining_ = nullptr;
};

int Trace::FindAffectedRegisters(DynamicBitSet* affected_registers,
                                 Zone* zone) {
351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368
  int max_register = RegExpCompiler::kNoRegister;
  for (DeferredAction* action = actions_; action != nullptr;
       action = action->next()) {
    if (action->action_type() == ActionNode::CLEAR_CAPTURES) {
      Interval range = static_cast<DeferredClearCaptures*>(action)->range();
      for (int i = range.from(); i <= range.to(); i++)
        affected_registers->Set(i, zone);
      if (range.to() > max_register) max_register = range.to();
    } else {
      affected_registers->Set(action->reg(), zone);
      if (action->reg() > max_register) max_register = action->reg();
    }
  }
  return max_register;
}

void Trace::RestoreAffectedRegisters(RegExpMacroAssembler* assembler,
                                     int max_register,
369 370
                                     const DynamicBitSet& registers_to_pop,
                                     const DynamicBitSet& registers_to_clear) {
371 372 373 374 375 376 377 378 379 380 381 382 383 384 385
  for (int reg = max_register; reg >= 0; reg--) {
    if (registers_to_pop.Get(reg)) {
      assembler->PopRegister(reg);
    } else if (registers_to_clear.Get(reg)) {
      int clear_to = reg;
      while (reg > 0 && registers_to_clear.Get(reg - 1)) {
        reg--;
      }
      assembler->ClearRegisters(reg, clear_to);
    }
  }
}

void Trace::PerformDeferredActions(RegExpMacroAssembler* assembler,
                                   int max_register,
386 387 388 389
                                   const DynamicBitSet& affected_registers,
                                   DynamicBitSet* registers_to_pop,
                                   DynamicBitSet* registers_to_clear,
                                   Zone* zone) {
390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417
  // The "+1" is to avoid a push_limit of zero if stack_limit_slack() is 1.
  const int push_limit = (assembler->stack_limit_slack() + 1) / 2;

  // Count pushes performed to force a stack limit check occasionally.
  int pushes = 0;

  for (int reg = 0; reg <= max_register; reg++) {
    if (!affected_registers.Get(reg)) {
      continue;
    }

    // The chronologically first deferred action in the trace
    // is used to infer the action needed to restore a register
    // to its previous state (or not, if it's safe to ignore it).
    enum DeferredActionUndoType { IGNORE, RESTORE, CLEAR };
    DeferredActionUndoType undo_action = IGNORE;

    int value = 0;
    bool absolute = false;
    bool clear = false;
    static const int kNoStore = kMinInt;
    int store_position = kNoStore;
    // This is a little tricky because we are scanning the actions in reverse
    // historical order (newest first).
    for (DeferredAction* action = actions_; action != nullptr;
         action = action->next()) {
      if (action->Mentions(reg)) {
        switch (action->action_type()) {
418 419 420
          case ActionNode::SET_REGISTER_FOR_LOOP: {
            Trace::DeferredSetRegisterForLoop* psr =
                static_cast<Trace::DeferredSetRegisterForLoop*>(action);
421 422 423 424
            if (!absolute) {
              value += psr->value();
              absolute = true;
            }
425
            // SET_REGISTER_FOR_LOOP is only used for newly introduced loop
426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532
            // counters. They can have a significant previous value if they
            // occur in a loop. TODO(lrn): Propagate this information, so
            // we can set undo_action to IGNORE if we know there is no value to
            // restore.
            undo_action = RESTORE;
            DCHECK_EQ(store_position, kNoStore);
            DCHECK(!clear);
            break;
          }
          case ActionNode::INCREMENT_REGISTER:
            if (!absolute) {
              value++;
            }
            DCHECK_EQ(store_position, kNoStore);
            DCHECK(!clear);
            undo_action = RESTORE;
            break;
          case ActionNode::STORE_POSITION: {
            Trace::DeferredCapture* pc =
                static_cast<Trace::DeferredCapture*>(action);
            if (!clear && store_position == kNoStore) {
              store_position = pc->cp_offset();
            }

            // For captures we know that stores and clears alternate.
            // Other register, are never cleared, and if the occur
            // inside a loop, they might be assigned more than once.
            if (reg <= 1) {
              // Registers zero and one, aka "capture zero", is
              // always set correctly if we succeed. There is no
              // need to undo a setting on backtrack, because we
              // will set it again or fail.
              undo_action = IGNORE;
            } else {
              undo_action = pc->is_capture() ? CLEAR : RESTORE;
            }
            DCHECK(!absolute);
            DCHECK_EQ(value, 0);
            break;
          }
          case ActionNode::CLEAR_CAPTURES: {
            // Since we're scanning in reverse order, if we've already
            // set the position we have to ignore historically earlier
            // clearing operations.
            if (store_position == kNoStore) {
              clear = true;
            }
            undo_action = RESTORE;
            DCHECK(!absolute);
            DCHECK_EQ(value, 0);
            break;
          }
          default:
            UNREACHABLE();
            break;
        }
      }
    }
    // Prepare for the undo-action (e.g., push if it's going to be popped).
    if (undo_action == RESTORE) {
      pushes++;
      RegExpMacroAssembler::StackCheckFlag stack_check =
          RegExpMacroAssembler::kNoStackLimitCheck;
      if (pushes == push_limit) {
        stack_check = RegExpMacroAssembler::kCheckStackLimit;
        pushes = 0;
      }

      assembler->PushRegister(reg, stack_check);
      registers_to_pop->Set(reg, zone);
    } else if (undo_action == CLEAR) {
      registers_to_clear->Set(reg, zone);
    }
    // Perform the chronologically last action (or accumulated increment)
    // for the register.
    if (store_position != kNoStore) {
      assembler->WriteCurrentPositionToRegister(reg, store_position);
    } else if (clear) {
      assembler->ClearRegisters(reg, reg);
    } else if (absolute) {
      assembler->SetRegister(reg, value);
    } else if (value != 0) {
      assembler->AdvanceRegister(reg, value);
    }
  }
}

// This is called as we come into a loop choice node and some other tricky
// nodes.  It normalizes the state of the code generator to ensure we can
// generate generic code.
void Trace::Flush(RegExpCompiler* compiler, RegExpNode* successor) {
  RegExpMacroAssembler* assembler = compiler->macro_assembler();

  DCHECK(!is_trivial());

  if (actions_ == nullptr && backtrack() == nullptr) {
    // Here we just have some deferred cp advances to fix and we are back to
    // a normal situation.  We may also have to forget some information gained
    // through a quick check that was already performed.
    if (cp_offset_ != 0) assembler->AdvanceCurrentPosition(cp_offset_);
    // Create a new trivial state and generate the node with that.
    Trace new_state;
    successor->Emit(compiler, &new_state);
    return;
  }

  // Generate deferred actions here along with code to undo them again.
533
  DynamicBitSet affected_registers;
534 535 536 537 538 539 540 541 542 543

  if (backtrack() != nullptr) {
    // Here we have a concrete backtrack location.  These are set up by choice
    // nodes and so they indicate that we have a deferred save of the current
    // position which we may need to emit here.
    assembler->PushCurrentPosition();
  }

  int max_register =
      FindAffectedRegisters(&affected_registers, compiler->zone());
544 545
  DynamicBitSet registers_to_pop;
  DynamicBitSet registers_to_clear;
546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629
  PerformDeferredActions(assembler, max_register, affected_registers,
                         &registers_to_pop, &registers_to_clear,
                         compiler->zone());
  if (cp_offset_ != 0) {
    assembler->AdvanceCurrentPosition(cp_offset_);
  }

  // Create a new trivial state and generate the node with that.
  Label undo;
  assembler->PushBacktrack(&undo);
  if (successor->KeepRecursing(compiler)) {
    Trace new_state;
    successor->Emit(compiler, &new_state);
  } else {
    compiler->AddWork(successor);
    assembler->GoTo(successor->label());
  }

  // On backtrack we need to restore state.
  assembler->Bind(&undo);
  RestoreAffectedRegisters(assembler, max_register, registers_to_pop,
                           registers_to_clear);
  if (backtrack() == nullptr) {
    assembler->Backtrack();
  } else {
    assembler->PopCurrentPosition();
    assembler->GoTo(backtrack());
  }
}

void NegativeSubmatchSuccess::Emit(RegExpCompiler* compiler, Trace* trace) {
  RegExpMacroAssembler* assembler = compiler->macro_assembler();

  // Omit flushing the trace. We discard the entire stack frame anyway.

  if (!label()->is_bound()) {
    // We are completely independent of the trace, since we ignore it,
    // so this code can be used as the generic version.
    assembler->Bind(label());
  }

  // Throw away everything on the backtrack stack since the start
  // of the negative submatch and restore the character position.
  assembler->ReadCurrentPositionFromRegister(current_position_register_);
  assembler->ReadStackPointerFromRegister(stack_pointer_register_);
  if (clear_capture_count_ > 0) {
    // Clear any captures that might have been performed during the success
    // of the body of the negative look-ahead.
    int clear_capture_end = clear_capture_start_ + clear_capture_count_ - 1;
    assembler->ClearRegisters(clear_capture_start_, clear_capture_end);
  }
  // Now that we have unwound the stack we find at the top of the stack the
  // backtrack that the BeginSubmatch node got.
  assembler->Backtrack();
}

void EndNode::Emit(RegExpCompiler* compiler, Trace* trace) {
  if (!trace->is_trivial()) {
    trace->Flush(compiler, this);
    return;
  }
  RegExpMacroAssembler* assembler = compiler->macro_assembler();
  if (!label()->is_bound()) {
    assembler->Bind(label());
  }
  switch (action_) {
    case ACCEPT:
      assembler->Succeed();
      return;
    case BACKTRACK:
      assembler->GoTo(trace->backtrack());
      return;
    case NEGATIVE_SUBMATCH_SUCCESS:
      // This case is handled in a different virtual method.
      UNREACHABLE();
  }
  UNIMPLEMENTED();
}

void GuardedAlternative::AddGuard(Guard* guard, Zone* zone) {
  if (guards_ == nullptr) guards_ = new (zone) ZoneList<Guard*>(1, zone);
  guards_->Add(guard, zone);
}

630 631
ActionNode* ActionNode::SetRegisterForLoop(int reg, int val,
                                           RegExpNode* on_success) {
632
  ActionNode* result =
633
      new (on_success->zone()) ActionNode(SET_REGISTER_FOR_LOOP, on_success);
634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727
  result->data_.u_store_register.reg = reg;
  result->data_.u_store_register.value = val;
  return result;
}

ActionNode* ActionNode::IncrementRegister(int reg, RegExpNode* on_success) {
  ActionNode* result =
      new (on_success->zone()) ActionNode(INCREMENT_REGISTER, on_success);
  result->data_.u_increment_register.reg = reg;
  return result;
}

ActionNode* ActionNode::StorePosition(int reg, bool is_capture,
                                      RegExpNode* on_success) {
  ActionNode* result =
      new (on_success->zone()) ActionNode(STORE_POSITION, on_success);
  result->data_.u_position_register.reg = reg;
  result->data_.u_position_register.is_capture = is_capture;
  return result;
}

ActionNode* ActionNode::ClearCaptures(Interval range, RegExpNode* on_success) {
  ActionNode* result =
      new (on_success->zone()) ActionNode(CLEAR_CAPTURES, on_success);
  result->data_.u_clear_captures.range_from = range.from();
  result->data_.u_clear_captures.range_to = range.to();
  return result;
}

ActionNode* ActionNode::BeginSubmatch(int stack_reg, int position_reg,
                                      RegExpNode* on_success) {
  ActionNode* result =
      new (on_success->zone()) ActionNode(BEGIN_SUBMATCH, on_success);
  result->data_.u_submatch.stack_pointer_register = stack_reg;
  result->data_.u_submatch.current_position_register = position_reg;
  return result;
}

ActionNode* ActionNode::PositiveSubmatchSuccess(int stack_reg, int position_reg,
                                                int clear_register_count,
                                                int clear_register_from,
                                                RegExpNode* on_success) {
  ActionNode* result = new (on_success->zone())
      ActionNode(POSITIVE_SUBMATCH_SUCCESS, on_success);
  result->data_.u_submatch.stack_pointer_register = stack_reg;
  result->data_.u_submatch.current_position_register = position_reg;
  result->data_.u_submatch.clear_register_count = clear_register_count;
  result->data_.u_submatch.clear_register_from = clear_register_from;
  return result;
}

ActionNode* ActionNode::EmptyMatchCheck(int start_register,
                                        int repetition_register,
                                        int repetition_limit,
                                        RegExpNode* on_success) {
  ActionNode* result =
      new (on_success->zone()) ActionNode(EMPTY_MATCH_CHECK, on_success);
  result->data_.u_empty_match_check.start_register = start_register;
  result->data_.u_empty_match_check.repetition_register = repetition_register;
  result->data_.u_empty_match_check.repetition_limit = repetition_limit;
  return result;
}

#define DEFINE_ACCEPT(Type) \
  void Type##Node::Accept(NodeVisitor* visitor) { visitor->Visit##Type(this); }
FOR_EACH_NODE_TYPE(DEFINE_ACCEPT)
#undef DEFINE_ACCEPT

// -------------------------------------------------------------------
// Emit code.

void ChoiceNode::GenerateGuard(RegExpMacroAssembler* macro_assembler,
                               Guard* guard, Trace* trace) {
  switch (guard->op()) {
    case Guard::LT:
      DCHECK(!trace->mentions_reg(guard->reg()));
      macro_assembler->IfRegisterGE(guard->reg(), guard->value(),
                                    trace->backtrack());
      break;
    case Guard::GEQ:
      DCHECK(!trace->mentions_reg(guard->reg()));
      macro_assembler->IfRegisterLT(guard->reg(), guard->value(),
                                    trace->backtrack());
      break;
  }
}

// Returns the number of characters in the equivalence class, omitting those
// that cannot occur in the source string because it is Latin1.
static int GetCaseIndependentLetters(Isolate* isolate, uc16 character,
                                     bool one_byte_subject,
                                     unibrow::uchar* letters,
                                     int letter_length) {
#ifdef V8_INTL_SUPPORT
728 729 730 731 732
  // Special case for U+017F which has upper case in ASCII range.
  if (character == 0x017f) {
    letters[0] = character;
    return 1;
  }
733 734 735 736 737 738 739 740 741
  icu::UnicodeSet set;
  set.add(character);
  set = set.closeOver(USET_CASE_INSENSITIVE);
  int32_t range_count = set.getRangeCount();
  int items = 0;
  for (int32_t i = 0; i < range_count; i++) {
    UChar32 start = set.getRangeStart(i);
    UChar32 end = set.getRangeEnd(i);
    CHECK(end - start + items <= letter_length);
742 743 744 745 746 747 748 749 750 751 752
    // Only add to the output if character is not in ASCII range
    // or the case equivalent character is in ASCII range.
    // #sec-runtime-semantics-canonicalize-ch
    // 3.g If the numeric value of ch ≥ 128 and the numeric value of cu < 128,
    //     return ch.
    if (!((start >= 128) && (character < 128))) {
      // No range have start and end span across code point 128.
      DCHECK((start >= 128) == (end >= 128));
      for (UChar32 cu = start; cu <= end; cu++) {
        if (one_byte_subject && cu > String::kMaxOneByteCharCode) break;
        letters[items++] = (unibrow::uchar)(cu);
753
      }
754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333
    }
  }
  return items;
#else
  int length =
      isolate->jsregexp_uncanonicalize()->get(character, '\0', letters);
  // Unibrow returns 0 or 1 for characters where case independence is
  // trivial.
  if (length == 0) {
    letters[0] = character;
    length = 1;
  }

  if (one_byte_subject) {
    int new_length = 0;
    for (int i = 0; i < length; i++) {
      if (letters[i] <= String::kMaxOneByteCharCode) {
        letters[new_length++] = letters[i];
      }
    }
    length = new_length;
  }

  return length;
#endif  // V8_INTL_SUPPORT
}

static inline bool EmitSimpleCharacter(Isolate* isolate,
                                       RegExpCompiler* compiler, uc16 c,
                                       Label* on_failure, int cp_offset,
                                       bool check, bool preloaded) {
  RegExpMacroAssembler* assembler = compiler->macro_assembler();
  bool bound_checked = false;
  if (!preloaded) {
    assembler->LoadCurrentCharacter(cp_offset, on_failure, check);
    bound_checked = true;
  }
  assembler->CheckNotCharacter(c, on_failure);
  return bound_checked;
}

// Only emits non-letters (things that don't have case).  Only used for case
// independent matches.
static inline bool EmitAtomNonLetter(Isolate* isolate, RegExpCompiler* compiler,
                                     uc16 c, Label* on_failure, int cp_offset,
                                     bool check, bool preloaded) {
  RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
  bool one_byte = compiler->one_byte();
  unibrow::uchar chars[4];
  int length = GetCaseIndependentLetters(isolate, c, one_byte, chars, 4);
  if (length < 1) {
    // This can't match.  Must be an one-byte subject and a non-one-byte
    // character.  We do not need to do anything since the one-byte pass
    // already handled this.
    return false;  // Bounds not checked.
  }
  bool checked = false;
  // We handle the length > 1 case in a later pass.
  if (length == 1) {
    if (one_byte && c > String::kMaxOneByteCharCodeU) {
      // Can't match - see above.
      return false;  // Bounds not checked.
    }
    if (!preloaded) {
      macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check);
      checked = check;
    }
    macro_assembler->CheckNotCharacter(c, on_failure);
  }
  return checked;
}

static bool ShortCutEmitCharacterPair(RegExpMacroAssembler* macro_assembler,
                                      bool one_byte, uc16 c1, uc16 c2,
                                      Label* on_failure) {
  uc16 char_mask;
  if (one_byte) {
    char_mask = String::kMaxOneByteCharCode;
  } else {
    char_mask = String::kMaxUtf16CodeUnit;
  }
  uc16 exor = c1 ^ c2;
  // Check whether exor has only one bit set.
  if (((exor - 1) & exor) == 0) {
    // If c1 and c2 differ only by one bit.
    // Ecma262UnCanonicalize always gives the highest number last.
    DCHECK(c2 > c1);
    uc16 mask = char_mask ^ exor;
    macro_assembler->CheckNotCharacterAfterAnd(c1, mask, on_failure);
    return true;
  }
  DCHECK(c2 > c1);
  uc16 diff = c2 - c1;
  if (((diff - 1) & diff) == 0 && c1 >= diff) {
    // If the characters differ by 2^n but don't differ by one bit then
    // subtract the difference from the found character, then do the or
    // trick.  We avoid the theoretical case where negative numbers are
    // involved in order to simplify code generation.
    uc16 mask = char_mask ^ diff;
    macro_assembler->CheckNotCharacterAfterMinusAnd(c1 - diff, diff, mask,
                                                    on_failure);
    return true;
  }
  return false;
}

using EmitCharacterFunction = bool(Isolate* isolate, RegExpCompiler* compiler,
                                   uc16 c, Label* on_failure, int cp_offset,
                                   bool check, bool preloaded);

// Only emits letters (things that have case).  Only used for case independent
// matches.
static inline bool EmitAtomLetter(Isolate* isolate, RegExpCompiler* compiler,
                                  uc16 c, Label* on_failure, int cp_offset,
                                  bool check, bool preloaded) {
  RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
  bool one_byte = compiler->one_byte();
  unibrow::uchar chars[4];
  int length = GetCaseIndependentLetters(isolate, c, one_byte, chars, 4);
  if (length <= 1) return false;
  // We may not need to check against the end of the input string
  // if this character lies before a character that matched.
  if (!preloaded) {
    macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check);
  }
  Label ok;
  switch (length) {
    case 2: {
      if (ShortCutEmitCharacterPair(macro_assembler, one_byte, chars[0],
                                    chars[1], on_failure)) {
      } else {
        macro_assembler->CheckCharacter(chars[0], &ok);
        macro_assembler->CheckNotCharacter(chars[1], on_failure);
        macro_assembler->Bind(&ok);
      }
      break;
    }
    case 4:
      macro_assembler->CheckCharacter(chars[3], &ok);
      V8_FALLTHROUGH;
    case 3:
      macro_assembler->CheckCharacter(chars[0], &ok);
      macro_assembler->CheckCharacter(chars[1], &ok);
      macro_assembler->CheckNotCharacter(chars[2], on_failure);
      macro_assembler->Bind(&ok);
      break;
    default:
      UNREACHABLE();
  }
  return true;
}

static void EmitBoundaryTest(RegExpMacroAssembler* masm, int border,
                             Label* fall_through, Label* above_or_equal,
                             Label* below) {
  if (below != fall_through) {
    masm->CheckCharacterLT(border, below);
    if (above_or_equal != fall_through) masm->GoTo(above_or_equal);
  } else {
    masm->CheckCharacterGT(border - 1, above_or_equal);
  }
}

static void EmitDoubleBoundaryTest(RegExpMacroAssembler* masm, int first,
                                   int last, Label* fall_through,
                                   Label* in_range, Label* out_of_range) {
  if (in_range == fall_through) {
    if (first == last) {
      masm->CheckNotCharacter(first, out_of_range);
    } else {
      masm->CheckCharacterNotInRange(first, last, out_of_range);
    }
  } else {
    if (first == last) {
      masm->CheckCharacter(first, in_range);
    } else {
      masm->CheckCharacterInRange(first, last, in_range);
    }
    if (out_of_range != fall_through) masm->GoTo(out_of_range);
  }
}

// even_label is for ranges[i] to ranges[i + 1] where i - start_index is even.
// odd_label is for ranges[i] to ranges[i + 1] where i - start_index is odd.
static void EmitUseLookupTable(RegExpMacroAssembler* masm,
                               ZoneList<int>* ranges, int start_index,
                               int end_index, int min_char, Label* fall_through,
                               Label* even_label, Label* odd_label) {
  static const int kSize = RegExpMacroAssembler::kTableSize;
  static const int kMask = RegExpMacroAssembler::kTableMask;

  int base = (min_char & ~kMask);
  USE(base);

  // Assert that everything is on one kTableSize page.
  for (int i = start_index; i <= end_index; i++) {
    DCHECK_EQ(ranges->at(i) & ~kMask, base);
  }
  DCHECK(start_index == 0 || (ranges->at(start_index - 1) & ~kMask) <= base);

  char templ[kSize];
  Label* on_bit_set;
  Label* on_bit_clear;
  int bit;
  if (even_label == fall_through) {
    on_bit_set = odd_label;
    on_bit_clear = even_label;
    bit = 1;
  } else {
    on_bit_set = even_label;
    on_bit_clear = odd_label;
    bit = 0;
  }
  for (int i = 0; i < (ranges->at(start_index) & kMask) && i < kSize; i++) {
    templ[i] = bit;
  }
  int j = 0;
  bit ^= 1;
  for (int i = start_index; i < end_index; i++) {
    for (j = (ranges->at(i) & kMask); j < (ranges->at(i + 1) & kMask); j++) {
      templ[j] = bit;
    }
    bit ^= 1;
  }
  for (int i = j; i < kSize; i++) {
    templ[i] = bit;
  }
  Factory* factory = masm->isolate()->factory();
  // TODO(erikcorry): Cache these.
  Handle<ByteArray> ba = factory->NewByteArray(kSize, AllocationType::kOld);
  for (int i = 0; i < kSize; i++) {
    ba->set(i, templ[i]);
  }
  masm->CheckBitInTable(ba, on_bit_set);
  if (on_bit_clear != fall_through) masm->GoTo(on_bit_clear);
}

static void CutOutRange(RegExpMacroAssembler* masm, ZoneList<int>* ranges,
                        int start_index, int end_index, int cut_index,
                        Label* even_label, Label* odd_label) {
  bool odd = (((cut_index - start_index) & 1) == 1);
  Label* in_range_label = odd ? odd_label : even_label;
  Label dummy;
  EmitDoubleBoundaryTest(masm, ranges->at(cut_index),
                         ranges->at(cut_index + 1) - 1, &dummy, in_range_label,
                         &dummy);
  DCHECK(!dummy.is_linked());
  // Cut out the single range by rewriting the array.  This creates a new
  // range that is a merger of the two ranges on either side of the one we
  // are cutting out.  The oddity of the labels is preserved.
  for (int j = cut_index; j > start_index; j--) {
    ranges->at(j) = ranges->at(j - 1);
  }
  for (int j = cut_index + 1; j < end_index; j++) {
    ranges->at(j) = ranges->at(j + 1);
  }
}

// Unicode case.  Split the search space into kSize spaces that are handled
// with recursion.
static void SplitSearchSpace(ZoneList<int>* ranges, int start_index,
                             int end_index, int* new_start_index,
                             int* new_end_index, int* border) {
  static const int kSize = RegExpMacroAssembler::kTableSize;
  static const int kMask = RegExpMacroAssembler::kTableMask;

  int first = ranges->at(start_index);
  int last = ranges->at(end_index) - 1;

  *new_start_index = start_index;
  *border = (ranges->at(start_index) & ~kMask) + kSize;
  while (*new_start_index < end_index) {
    if (ranges->at(*new_start_index) > *border) break;
    (*new_start_index)++;
  }
  // new_start_index is the index of the first edge that is beyond the
  // current kSize space.

  // For very large search spaces we do a binary chop search of the non-Latin1
  // space instead of just going to the end of the current kSize space.  The
  // heuristics are complicated a little by the fact that any 128-character
  // encoding space can be quickly tested with a table lookup, so we don't
  // wish to do binary chop search at a smaller granularity than that.  A
  // 128-character space can take up a lot of space in the ranges array if,
  // for example, we only want to match every second character (eg. the lower
  // case characters on some Unicode pages).
  int binary_chop_index = (end_index + start_index) / 2;
  // The first test ensures that we get to the code that handles the Latin1
  // range with a single not-taken branch, speeding up this important
  // character range (even non-Latin1 charset-based text has spaces and
  // punctuation).
  if (*border - 1 > String::kMaxOneByteCharCode &&  // Latin1 case.
      end_index - start_index > (*new_start_index - start_index) * 2 &&
      last - first > kSize * 2 && binary_chop_index > *new_start_index &&
      ranges->at(binary_chop_index) >= first + 2 * kSize) {
    int scan_forward_for_section_border = binary_chop_index;
    int new_border = (ranges->at(binary_chop_index) | kMask) + 1;

    while (scan_forward_for_section_border < end_index) {
      if (ranges->at(scan_forward_for_section_border) > new_border) {
        *new_start_index = scan_forward_for_section_border;
        *border = new_border;
        break;
      }
      scan_forward_for_section_border++;
    }
  }

  DCHECK(*new_start_index > start_index);
  *new_end_index = *new_start_index - 1;
  if (ranges->at(*new_end_index) == *border) {
    (*new_end_index)--;
  }
  if (*border >= ranges->at(end_index)) {
    *border = ranges->at(end_index);
    *new_start_index = end_index;  // Won't be used.
    *new_end_index = end_index - 1;
  }
}

// Gets a series of segment boundaries representing a character class.  If the
// character is in the range between an even and an odd boundary (counting from
// start_index) then go to even_label, otherwise go to odd_label.  We already
// know that the character is in the range of min_char to max_char inclusive.
// Either label can be nullptr indicating backtracking.  Either label can also
// be equal to the fall_through label.
static void GenerateBranches(RegExpMacroAssembler* masm, ZoneList<int>* ranges,
                             int start_index, int end_index, uc32 min_char,
                             uc32 max_char, Label* fall_through,
                             Label* even_label, Label* odd_label) {
  DCHECK_LE(min_char, String::kMaxUtf16CodeUnit);
  DCHECK_LE(max_char, String::kMaxUtf16CodeUnit);

  int first = ranges->at(start_index);
  int last = ranges->at(end_index) - 1;

  DCHECK_LT(min_char, first);

  // Just need to test if the character is before or on-or-after
  // a particular character.
  if (start_index == end_index) {
    EmitBoundaryTest(masm, first, fall_through, even_label, odd_label);
    return;
  }

  // Another almost trivial case:  There is one interval in the middle that is
  // different from the end intervals.
  if (start_index + 1 == end_index) {
    EmitDoubleBoundaryTest(masm, first, last, fall_through, even_label,
                           odd_label);
    return;
  }

  // It's not worth using table lookup if there are very few intervals in the
  // character class.
  if (end_index - start_index <= 6) {
    // It is faster to test for individual characters, so we look for those
    // first, then try arbitrary ranges in the second round.
    static int kNoCutIndex = -1;
    int cut = kNoCutIndex;
    for (int i = start_index; i < end_index; i++) {
      if (ranges->at(i) == ranges->at(i + 1) - 1) {
        cut = i;
        break;
      }
    }
    if (cut == kNoCutIndex) cut = start_index;
    CutOutRange(masm, ranges, start_index, end_index, cut, even_label,
                odd_label);
    DCHECK_GE(end_index - start_index, 2);
    GenerateBranches(masm, ranges, start_index + 1, end_index - 1, min_char,
                     max_char, fall_through, even_label, odd_label);
    return;
  }

  // If there are a lot of intervals in the regexp, then we will use tables to
  // determine whether the character is inside or outside the character class.
  static const int kBits = RegExpMacroAssembler::kTableSizeBits;

  if ((max_char >> kBits) == (min_char >> kBits)) {
    EmitUseLookupTable(masm, ranges, start_index, end_index, min_char,
                       fall_through, even_label, odd_label);
    return;
  }

  if ((min_char >> kBits) != (first >> kBits)) {
    masm->CheckCharacterLT(first, odd_label);
    GenerateBranches(masm, ranges, start_index + 1, end_index, first, max_char,
                     fall_through, odd_label, even_label);
    return;
  }

  int new_start_index = 0;
  int new_end_index = 0;
  int border = 0;

  SplitSearchSpace(ranges, start_index, end_index, &new_start_index,
                   &new_end_index, &border);

  Label handle_rest;
  Label* above = &handle_rest;
  if (border == last + 1) {
    // We didn't find any section that started after the limit, so everything
    // above the border is one of the terminal labels.
    above = (end_index & 1) != (start_index & 1) ? odd_label : even_label;
    DCHECK(new_end_index == end_index - 1);
  }

  DCHECK_LE(start_index, new_end_index);
  DCHECK_LE(new_start_index, end_index);
  DCHECK_LT(start_index, new_start_index);
  DCHECK_LT(new_end_index, end_index);
  DCHECK(new_end_index + 1 == new_start_index ||
         (new_end_index + 2 == new_start_index &&
          border == ranges->at(new_end_index + 1)));
  DCHECK_LT(min_char, border - 1);
  DCHECK_LT(border, max_char);
  DCHECK_LT(ranges->at(new_end_index), border);
  DCHECK(border < ranges->at(new_start_index) ||
         (border == ranges->at(new_start_index) &&
          new_start_index == end_index && new_end_index == end_index - 1 &&
          border == last + 1));
  DCHECK(new_start_index == 0 || border >= ranges->at(new_start_index - 1));

  masm->CheckCharacterGT(border - 1, above);
  Label dummy;
  GenerateBranches(masm, ranges, start_index, new_end_index, min_char,
                   border - 1, &dummy, even_label, odd_label);
  if (handle_rest.is_linked()) {
    masm->Bind(&handle_rest);
    bool flip = (new_start_index & 1) != (start_index & 1);
    GenerateBranches(masm, ranges, new_start_index, end_index, border, max_char,
                     &dummy, flip ? odd_label : even_label,
                     flip ? even_label : odd_label);
  }
}

static void EmitCharClass(RegExpMacroAssembler* macro_assembler,
                          RegExpCharacterClass* cc, bool one_byte,
                          Label* on_failure, int cp_offset, bool check_offset,
                          bool preloaded, Zone* zone) {
  ZoneList<CharacterRange>* ranges = cc->ranges(zone);
  CharacterRange::Canonicalize(ranges);

  int max_char;
  if (one_byte) {
    max_char = String::kMaxOneByteCharCode;
  } else {
    max_char = String::kMaxUtf16CodeUnit;
  }

  int range_count = ranges->length();

  int last_valid_range = range_count - 1;
  while (last_valid_range >= 0) {
    CharacterRange& range = ranges->at(last_valid_range);
    if (range.from() <= max_char) {
      break;
    }
    last_valid_range--;
  }

  if (last_valid_range < 0) {
    if (!cc->is_negated()) {
      macro_assembler->GoTo(on_failure);
    }
    if (check_offset) {
      macro_assembler->CheckPosition(cp_offset, on_failure);
    }
    return;
  }

  if (last_valid_range == 0 && ranges->at(0).IsEverything(max_char)) {
    if (cc->is_negated()) {
      macro_assembler->GoTo(on_failure);
    } else {
      // This is a common case hit by non-anchored expressions.
      if (check_offset) {
        macro_assembler->CheckPosition(cp_offset, on_failure);
      }
    }
    return;
  }

  if (!preloaded) {
    macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check_offset);
  }

  if (cc->is_standard(zone) && macro_assembler->CheckSpecialCharacterClass(
                                   cc->standard_type(), on_failure)) {
    return;
  }

  // A new list with ascending entries.  Each entry is a code unit
  // where there is a boundary between code units that are part of
  // the class and code units that are not.  Normally we insert an
  // entry at zero which goes to the failure label, but if there
  // was already one there we fall through for success on that entry.
  // Subsequent entries have alternating meaning (success/failure).
  ZoneList<int>* range_boundaries =
      new (zone) ZoneList<int>(last_valid_range, zone);

  bool zeroth_entry_is_failure = !cc->is_negated();

  for (int i = 0; i <= last_valid_range; i++) {
    CharacterRange& range = ranges->at(i);
    if (range.from() == 0) {
      DCHECK_EQ(i, 0);
      zeroth_entry_is_failure = !zeroth_entry_is_failure;
    } else {
      range_boundaries->Add(range.from(), zone);
    }
    range_boundaries->Add(range.to() + 1, zone);
  }
  int end_index = range_boundaries->length() - 1;
  if (range_boundaries->at(end_index) > max_char) {
    end_index--;
  }

  Label fall_through;
  GenerateBranches(macro_assembler, range_boundaries,
                   0,  // start_index.
                   end_index,
                   0,  // min_char.
                   max_char, &fall_through,
                   zeroth_entry_is_failure ? &fall_through : on_failure,
                   zeroth_entry_is_failure ? on_failure : &fall_through);
  macro_assembler->Bind(&fall_through);
}

RegExpNode::~RegExpNode() = default;

RegExpNode::LimitResult RegExpNode::LimitVersions(RegExpCompiler* compiler,
                                                  Trace* trace) {
  // If we are generating a greedy loop then don't stop and don't reuse code.
  if (trace->stop_node() != nullptr) {
    return CONTINUE;
  }

  RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
  if (trace->is_trivial()) {
    if (label_.is_bound() || on_work_list() || !KeepRecursing(compiler)) {
      // If a generic version is already scheduled to be generated or we have
      // recursed too deeply then just generate a jump to that code.
      macro_assembler->GoTo(&label_);
      // This will queue it up for generation of a generic version if it hasn't
      // already been queued.
      compiler->AddWork(this);
      return DONE;
    }
    // Generate generic version of the node and bind the label for later use.
    macro_assembler->Bind(&label_);
    return CONTINUE;
  }

  // We are being asked to make a non-generic version.  Keep track of how many
  // non-generic versions we generate so as not to overdo it.
  trace_count_++;
  if (KeepRecursing(compiler) && compiler->optimize() &&
      trace_count_ < kMaxCopiesCodeGenerated) {
    return CONTINUE;
  }

  // If we get here code has been generated for this node too many times or
  // recursion is too deep.  Time to switch to a generic version.  The code for
  // generic versions above can handle deep recursion properly.
  bool was_limiting = compiler->limiting_recursion();
  compiler->set_limiting_recursion(true);
  trace->Flush(compiler, this);
  compiler->set_limiting_recursion(was_limiting);
  return DONE;
}

bool RegExpNode::KeepRecursing(RegExpCompiler* compiler) {
  return !compiler->limiting_recursion() &&
         compiler->recursion_depth() <= RegExpCompiler::kMaxRecursion;
}

void ActionNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
                              BoyerMooreLookahead* bm, bool not_at_start) {
1334 1335 1336 1337 1338
  if (action_type_ == POSITIVE_SUBMATCH_SUCCESS) {
    // Anything may follow a positive submatch success, thus we need to accept
    // all characters from this position onwards.
    bm->SetRest(offset);
  } else {
1339 1340 1341 1342 1343
    on_success()->FillInBMInfo(isolate, offset, budget - 1, bm, not_at_start);
  }
  SaveBMInfo(bm, not_at_start, offset);
}

1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355
void ActionNode::GetQuickCheckDetails(QuickCheckDetails* details,
                                      RegExpCompiler* compiler, int filled_in,
                                      bool not_at_start) {
  if (action_type_ == SET_REGISTER_FOR_LOOP) {
    on_success()->GetQuickCheckDetailsFromLoopEntry(details, compiler,
                                                    filled_in, not_at_start);
  } else {
    on_success()->GetQuickCheckDetails(details, compiler, filled_in,
                                       not_at_start);
  }
}

1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366
void AssertionNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
                                 BoyerMooreLookahead* bm, bool not_at_start) {
  // Match the behaviour of EatsAtLeast on this node.
  if (assertion_type() == AT_START && not_at_start) return;
  on_success()->FillInBMInfo(isolate, offset, budget - 1, bm, not_at_start);
  SaveBMInfo(bm, not_at_start, offset);
}

void NegativeLookaroundChoiceNode::GetQuickCheckDetails(
    QuickCheckDetails* details, RegExpCompiler* compiler, int filled_in,
    bool not_at_start) {
1367
  RegExpNode* node = continue_node();
1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403
  return node->GetQuickCheckDetails(details, compiler, filled_in, not_at_start);
}

// Takes the left-most 1-bit and smears it out, setting all bits to its right.
static inline uint32_t SmearBitsRight(uint32_t v) {
  v |= v >> 1;
  v |= v >> 2;
  v |= v >> 4;
  v |= v >> 8;
  v |= v >> 16;
  return v;
}

bool QuickCheckDetails::Rationalize(bool asc) {
  bool found_useful_op = false;
  uint32_t char_mask;
  if (asc) {
    char_mask = String::kMaxOneByteCharCode;
  } else {
    char_mask = String::kMaxUtf16CodeUnit;
  }
  mask_ = 0;
  value_ = 0;
  int char_shift = 0;
  for (int i = 0; i < characters_; i++) {
    Position* pos = &positions_[i];
    if ((pos->mask & String::kMaxOneByteCharCode) != 0) {
      found_useful_op = true;
    }
    mask_ |= (pos->mask & char_mask) << char_shift;
    value_ |= (pos->value & char_mask) << char_shift;
    char_shift += asc ? 8 : 16;
  }
  return found_useful_op;
}

1404 1405 1406 1407 1408
int RegExpNode::EatsAtLeast(bool not_at_start) {
  return not_at_start ? eats_at_least_.eats_at_least_from_not_start
                      : eats_at_least_.eats_at_least_from_possibly_start;
}

1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467
EatsAtLeastInfo RegExpNode::EatsAtLeastFromLoopEntry() {
  // SET_REGISTER_FOR_LOOP is only used to initialize loop counters, and it
  // implies that the following node must be a LoopChoiceNode. If we need to
  // set registers to constant values for other reasons, we could introduce a
  // new action type SET_REGISTER that doesn't imply anything about its
  // successor.
  UNREACHABLE();
}

void RegExpNode::GetQuickCheckDetailsFromLoopEntry(QuickCheckDetails* details,
                                                   RegExpCompiler* compiler,
                                                   int characters_filled_in,
                                                   bool not_at_start) {
  // See comment in RegExpNode::EatsAtLeastFromLoopEntry.
  UNREACHABLE();
}

EatsAtLeastInfo LoopChoiceNode::EatsAtLeastFromLoopEntry() {
  DCHECK_EQ(alternatives_->length(), 2);  // There's just loop and continue.

  if (read_backward()) {
    // Can't do anything special for a backward loop, so return the basic values
    // that we got during analysis.
    return *eats_at_least_info();
  }

  // Figure out how much the loop body itself eats, not including anything in
  // the continuation case. In general, the nodes in the loop body should report
  // that they eat at least the number eaten by the continuation node, since any
  // successful match in the loop body must also include the continuation node.
  // However, in some cases involving positive lookaround, the loop body under-
  // reports its appetite, so use saturated math here to avoid negative numbers.
  uint8_t loop_body_from_not_start = base::saturated_cast<uint8_t>(
      loop_node_->EatsAtLeast(true) - continue_node_->EatsAtLeast(true));
  uint8_t loop_body_from_possibly_start = base::saturated_cast<uint8_t>(
      loop_node_->EatsAtLeast(false) - continue_node_->EatsAtLeast(true));

  // Limit the number of loop iterations to avoid overflow in subsequent steps.
  int loop_iterations = base::saturated_cast<uint8_t>(min_loop_iterations());

  EatsAtLeastInfo result;
  result.eats_at_least_from_not_start =
      base::saturated_cast<uint8_t>(loop_iterations * loop_body_from_not_start +
                                    continue_node_->EatsAtLeast(true));
  if (loop_iterations > 0 && loop_body_from_possibly_start > 0) {
    // First loop iteration eats at least one, so all subsequent iterations
    // and the after-loop chunk are guaranteed to not be at the start.
    result.eats_at_least_from_possibly_start = base::saturated_cast<uint8_t>(
        loop_body_from_possibly_start +
        (loop_iterations - 1) * loop_body_from_not_start +
        continue_node_->EatsAtLeast(true));
  } else {
    // Loop body might eat nothing, so only continue node contributes.
    result.eats_at_least_from_possibly_start =
        continue_node_->EatsAtLeast(false);
  }
  return result;
}

1468 1469 1470 1471 1472
bool RegExpNode::EmitQuickCheck(RegExpCompiler* compiler,
                                Trace* bounds_check_trace, Trace* trace,
                                bool preload_has_checked_bounds,
                                Label* on_possible_success,
                                QuickCheckDetails* details,
1473 1474 1475
                                bool fall_through_on_failure,
                                ChoiceNode* predecessor) {
  DCHECK_NOT_NULL(predecessor);
1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489
  if (details->characters() == 0) return false;
  GetQuickCheckDetails(details, compiler, 0,
                       trace->at_start() == Trace::FALSE_VALUE);
  if (details->cannot_match()) return false;
  if (!details->Rationalize(compiler->one_byte())) return false;
  DCHECK(details->characters() == 1 ||
         compiler->macro_assembler()->CanReadUnaligned());
  uint32_t mask = details->mask();
  uint32_t value = details->value();

  RegExpMacroAssembler* assembler = compiler->macro_assembler();

  if (trace->characters_preloaded() != details->characters()) {
    DCHECK(trace->cp_offset() == bounds_check_trace->cp_offset());
1490
    // The bounds check is performed using the minimum number of characters
1491 1492
    // any choice would eat, so if the bounds check fails, then none of the
    // choices can succeed, so we can just immediately backtrack, rather
1493 1494 1495 1496 1497
    // than go to the next choice. The number of characters preloaded may be
    // less than the number used for the bounds check.
    int eats_at_least = predecessor->EatsAtLeast(
        bounds_check_trace->at_start() == Trace::FALSE_VALUE);
    DCHECK_GE(eats_at_least, details->characters());
1498 1499
    assembler->LoadCurrentCharacter(
        trace->cp_offset(), bounds_check_trace->backtrack(),
1500
        !preload_has_checked_bounds, details->characters(), eats_at_least);
1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593
  }

  bool need_mask = true;

  if (details->characters() == 1) {
    // If number of characters preloaded is 1 then we used a byte or 16 bit
    // load so the value is already masked down.
    uint32_t char_mask;
    if (compiler->one_byte()) {
      char_mask = String::kMaxOneByteCharCode;
    } else {
      char_mask = String::kMaxUtf16CodeUnit;
    }
    if ((mask & char_mask) == char_mask) need_mask = false;
    mask &= char_mask;
  } else {
    // For 2-character preloads in one-byte mode or 1-character preloads in
    // two-byte mode we also use a 16 bit load with zero extend.
    static const uint32_t kTwoByteMask = 0xFFFF;
    static const uint32_t kFourByteMask = 0xFFFFFFFF;
    if (details->characters() == 2 && compiler->one_byte()) {
      if ((mask & kTwoByteMask) == kTwoByteMask) need_mask = false;
    } else if (details->characters() == 1 && !compiler->one_byte()) {
      if ((mask & kTwoByteMask) == kTwoByteMask) need_mask = false;
    } else {
      if (mask == kFourByteMask) need_mask = false;
    }
  }

  if (fall_through_on_failure) {
    if (need_mask) {
      assembler->CheckCharacterAfterAnd(value, mask, on_possible_success);
    } else {
      assembler->CheckCharacter(value, on_possible_success);
    }
  } else {
    if (need_mask) {
      assembler->CheckNotCharacterAfterAnd(value, mask, trace->backtrack());
    } else {
      assembler->CheckNotCharacter(value, trace->backtrack());
    }
  }
  return true;
}

// Here is the meat of GetQuickCheckDetails (see also the comment on the
// super-class in the .h file).
//
// We iterate along the text object, building up for each character a
// mask and value that can be used to test for a quick failure to match.
// The masks and values for the positions will be combined into a single
// machine word for the current character width in order to be used in
// generating a quick check.
void TextNode::GetQuickCheckDetails(QuickCheckDetails* details,
                                    RegExpCompiler* compiler,
                                    int characters_filled_in,
                                    bool not_at_start) {
  // Do not collect any quick check details if the text node reads backward,
  // since it reads in the opposite direction than we use for quick checks.
  if (read_backward()) return;
  Isolate* isolate = compiler->macro_assembler()->isolate();
  DCHECK(characters_filled_in < details->characters());
  int characters = details->characters();
  int char_mask;
  if (compiler->one_byte()) {
    char_mask = String::kMaxOneByteCharCode;
  } else {
    char_mask = String::kMaxUtf16CodeUnit;
  }
  for (int k = 0; k < elements()->length(); k++) {
    TextElement elm = elements()->at(k);
    if (elm.text_type() == TextElement::ATOM) {
      Vector<const uc16> quarks = elm.atom()->data();
      for (int i = 0; i < characters && i < quarks.length(); i++) {
        QuickCheckDetails::Position* pos =
            details->positions(characters_filled_in);
        uc16 c = quarks[i];
        if (elm.atom()->ignore_case()) {
          unibrow::uchar chars[4];
          int length = GetCaseIndependentLetters(
              isolate, c, compiler->one_byte(), chars, 4);
          if (length == 0) {
            // This can happen because all case variants are non-Latin1, but we
            // know the input is Latin1.
            details->set_cannot_match();
            pos->determines_perfectly = false;
            return;
          }
          if (length == 1) {
            // This letter has no case equivalents, so it's nice and simple
            // and the mask-compare will determine definitely whether we have
            // a match at this character position.
            pos->mask = char_mask;
1594
            pos->value = chars[0];
1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778
            pos->determines_perfectly = true;
          } else {
            uint32_t common_bits = char_mask;
            uint32_t bits = chars[0];
            for (int j = 1; j < length; j++) {
              uint32_t differing_bits = ((chars[j] & common_bits) ^ bits);
              common_bits ^= differing_bits;
              bits &= common_bits;
            }
            // If length is 2 and common bits has only one zero in it then
            // our mask and compare instruction will determine definitely
            // whether we have a match at this character position.  Otherwise
            // it can only be an approximate check.
            uint32_t one_zero = (common_bits | ~char_mask);
            if (length == 2 && ((~one_zero) & ((~one_zero) - 1)) == 0) {
              pos->determines_perfectly = true;
            }
            pos->mask = common_bits;
            pos->value = bits;
          }
        } else {
          // Don't ignore case.  Nice simple case where the mask-compare will
          // determine definitely whether we have a match at this character
          // position.
          if (c > char_mask) {
            details->set_cannot_match();
            pos->determines_perfectly = false;
            return;
          }
          pos->mask = char_mask;
          pos->value = c;
          pos->determines_perfectly = true;
        }
        characters_filled_in++;
        DCHECK(characters_filled_in <= details->characters());
        if (characters_filled_in == details->characters()) {
          return;
        }
      }
    } else {
      QuickCheckDetails::Position* pos =
          details->positions(characters_filled_in);
      RegExpCharacterClass* tree = elm.char_class();
      ZoneList<CharacterRange>* ranges = tree->ranges(zone());
      DCHECK(!ranges->is_empty());
      if (tree->is_negated()) {
        // A quick check uses multi-character mask and compare.  There is no
        // useful way to incorporate a negative char class into this scheme
        // so we just conservatively create a mask and value that will always
        // succeed.
        pos->mask = 0;
        pos->value = 0;
      } else {
        int first_range = 0;
        while (ranges->at(first_range).from() > char_mask) {
          first_range++;
          if (first_range == ranges->length()) {
            details->set_cannot_match();
            pos->determines_perfectly = false;
            return;
          }
        }
        CharacterRange range = ranges->at(first_range);
        uc16 from = range.from();
        uc16 to = range.to();
        if (to > char_mask) {
          to = char_mask;
        }
        uint32_t differing_bits = (from ^ to);
        // A mask and compare is only perfect if the differing bits form a
        // number like 00011111 with one single block of trailing 1s.
        if ((differing_bits & (differing_bits + 1)) == 0 &&
            from + differing_bits == to) {
          pos->determines_perfectly = true;
        }
        uint32_t common_bits = ~SmearBitsRight(differing_bits);
        uint32_t bits = (from & common_bits);
        for (int i = first_range + 1; i < ranges->length(); i++) {
          CharacterRange range = ranges->at(i);
          uc16 from = range.from();
          uc16 to = range.to();
          if (from > char_mask) continue;
          if (to > char_mask) to = char_mask;
          // Here we are combining more ranges into the mask and compare
          // value.  With each new range the mask becomes more sparse and
          // so the chances of a false positive rise.  A character class
          // with multiple ranges is assumed never to be equivalent to a
          // mask and compare operation.
          pos->determines_perfectly = false;
          uint32_t new_common_bits = (from ^ to);
          new_common_bits = ~SmearBitsRight(new_common_bits);
          common_bits &= new_common_bits;
          bits &= new_common_bits;
          uint32_t differing_bits = (from & common_bits) ^ bits;
          common_bits ^= differing_bits;
          bits &= common_bits;
        }
        pos->mask = common_bits;
        pos->value = bits;
      }
      characters_filled_in++;
      DCHECK(characters_filled_in <= details->characters());
      if (characters_filled_in == details->characters()) {
        return;
      }
    }
  }
  DCHECK(characters_filled_in != details->characters());
  if (!details->cannot_match()) {
    on_success()->GetQuickCheckDetails(details, compiler, characters_filled_in,
                                       true);
  }
}

void QuickCheckDetails::Clear() {
  for (int i = 0; i < characters_; i++) {
    positions_[i].mask = 0;
    positions_[i].value = 0;
    positions_[i].determines_perfectly = false;
  }
  characters_ = 0;
}

void QuickCheckDetails::Advance(int by, bool one_byte) {
  if (by >= characters_ || by < 0) {
    DCHECK_IMPLIES(by < 0, characters_ == 0);
    Clear();
    return;
  }
  DCHECK_LE(characters_ - by, 4);
  DCHECK_LE(characters_, 4);
  for (int i = 0; i < characters_ - by; i++) {
    positions_[i] = positions_[by + i];
  }
  for (int i = characters_ - by; i < characters_; i++) {
    positions_[i].mask = 0;
    positions_[i].value = 0;
    positions_[i].determines_perfectly = false;
  }
  characters_ -= by;
  // We could change mask_ and value_ here but we would never advance unless
  // they had already been used in a check and they won't be used again because
  // it would gain us nothing.  So there's no point.
}

void QuickCheckDetails::Merge(QuickCheckDetails* other, int from_index) {
  DCHECK(characters_ == other->characters_);
  if (other->cannot_match_) {
    return;
  }
  if (cannot_match_) {
    *this = *other;
    return;
  }
  for (int i = from_index; i < characters_; i++) {
    QuickCheckDetails::Position* pos = positions(i);
    QuickCheckDetails::Position* other_pos = other->positions(i);
    if (pos->mask != other_pos->mask || pos->value != other_pos->value ||
        !other_pos->determines_perfectly) {
      // Our mask-compare operation will be approximate unless we have the
      // exact same operation on both sides of the alternation.
      pos->determines_perfectly = false;
    }
    pos->mask &= other_pos->mask;
    pos->value &= pos->mask;
    other_pos->value &= pos->mask;
    uc16 differing_bits = (pos->value ^ other_pos->value);
    pos->mask &= ~differing_bits;
    pos->value &= pos->mask;
  }
}

class VisitMarker {
 public:
  explicit VisitMarker(NodeInfo* info) : info_(info) {
    DCHECK(!info->visited);
    info->visited = true;
  }
  ~VisitMarker() { info_->visited = false; }

 private:
  NodeInfo* info_;
};

1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809
// Temporarily sets traversed_loop_initialization_node_.
class LoopInitializationMarker {
 public:
  explicit LoopInitializationMarker(LoopChoiceNode* node) : node_(node) {
    DCHECK(!node_->traversed_loop_initialization_node_);
    node_->traversed_loop_initialization_node_ = true;
  }
  ~LoopInitializationMarker() {
    DCHECK(node_->traversed_loop_initialization_node_);
    node_->traversed_loop_initialization_node_ = false;
  }

 private:
  LoopChoiceNode* node_;
  DISALLOW_COPY_AND_ASSIGN(LoopInitializationMarker);
};

// Temporarily decrements min_loop_iterations_.
class IterationDecrementer {
 public:
  explicit IterationDecrementer(LoopChoiceNode* node) : node_(node) {
    DCHECK_GT(node_->min_loop_iterations_, 0);
    --node_->min_loop_iterations_;
  }
  ~IterationDecrementer() { ++node_->min_loop_iterations_; }

 private:
  LoopChoiceNode* node_;
  DISALLOW_COPY_AND_ASSIGN(IterationDecrementer);
};

1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961
RegExpNode* SeqRegExpNode::FilterOneByte(int depth) {
  if (info()->replacement_calculated) return replacement();
  if (depth < 0) return this;
  DCHECK(!info()->visited);
  VisitMarker marker(info());
  return FilterSuccessor(depth - 1);
}

RegExpNode* SeqRegExpNode::FilterSuccessor(int depth) {
  RegExpNode* next = on_success_->FilterOneByte(depth - 1);
  if (next == nullptr) return set_replacement(nullptr);
  on_success_ = next;
  return set_replacement(this);
}

// We need to check for the following characters: 0x39C 0x3BC 0x178.
bool RangeContainsLatin1Equivalents(CharacterRange range) {
  // TODO(dcarney): this could be a lot more efficient.
  return range.Contains(0x039C) || range.Contains(0x03BC) ||
         range.Contains(0x0178);
}

static bool RangesContainLatin1Equivalents(ZoneList<CharacterRange>* ranges) {
  for (int i = 0; i < ranges->length(); i++) {
    // TODO(dcarney): this could be a lot more efficient.
    if (RangeContainsLatin1Equivalents(ranges->at(i))) return true;
  }
  return false;
}

RegExpNode* TextNode::FilterOneByte(int depth) {
  if (info()->replacement_calculated) return replacement();
  if (depth < 0) return this;
  DCHECK(!info()->visited);
  VisitMarker marker(info());
  int element_count = elements()->length();
  for (int i = 0; i < element_count; i++) {
    TextElement elm = elements()->at(i);
    if (elm.text_type() == TextElement::ATOM) {
      Vector<const uc16> quarks = elm.atom()->data();
      for (int j = 0; j < quarks.length(); j++) {
        uint16_t c = quarks[j];
        if (elm.atom()->ignore_case()) {
          c = unibrow::Latin1::TryConvertToLatin1(c);
        }
        if (c > unibrow::Latin1::kMaxChar) return set_replacement(nullptr);
        // Replace quark in case we converted to Latin-1.
        uint16_t* writable_quarks = const_cast<uint16_t*>(quarks.begin());
        writable_quarks[j] = c;
      }
    } else {
      DCHECK(elm.text_type() == TextElement::CHAR_CLASS);
      RegExpCharacterClass* cc = elm.char_class();
      ZoneList<CharacterRange>* ranges = cc->ranges(zone());
      CharacterRange::Canonicalize(ranges);
      // Now they are in order so we only need to look at the first.
      int range_count = ranges->length();
      if (cc->is_negated()) {
        if (range_count != 0 && ranges->at(0).from() == 0 &&
            ranges->at(0).to() >= String::kMaxOneByteCharCode) {
          // This will be handled in a later filter.
          if (IgnoreCase(cc->flags()) && RangesContainLatin1Equivalents(ranges))
            continue;
          return set_replacement(nullptr);
        }
      } else {
        if (range_count == 0 ||
            ranges->at(0).from() > String::kMaxOneByteCharCode) {
          // This will be handled in a later filter.
          if (IgnoreCase(cc->flags()) && RangesContainLatin1Equivalents(ranges))
            continue;
          return set_replacement(nullptr);
        }
      }
    }
  }
  return FilterSuccessor(depth - 1);
}

RegExpNode* LoopChoiceNode::FilterOneByte(int depth) {
  if (info()->replacement_calculated) return replacement();
  if (depth < 0) return this;
  if (info()->visited) return this;
  {
    VisitMarker marker(info());

    RegExpNode* continue_replacement = continue_node_->FilterOneByte(depth - 1);
    // If we can't continue after the loop then there is no sense in doing the
    // loop.
    if (continue_replacement == nullptr) return set_replacement(nullptr);
  }

  return ChoiceNode::FilterOneByte(depth - 1);
}

RegExpNode* ChoiceNode::FilterOneByte(int depth) {
  if (info()->replacement_calculated) return replacement();
  if (depth < 0) return this;
  if (info()->visited) return this;
  VisitMarker marker(info());
  int choice_count = alternatives_->length();

  for (int i = 0; i < choice_count; i++) {
    GuardedAlternative alternative = alternatives_->at(i);
    if (alternative.guards() != nullptr &&
        alternative.guards()->length() != 0) {
      set_replacement(this);
      return this;
    }
  }

  int surviving = 0;
  RegExpNode* survivor = nullptr;
  for (int i = 0; i < choice_count; i++) {
    GuardedAlternative alternative = alternatives_->at(i);
    RegExpNode* replacement = alternative.node()->FilterOneByte(depth - 1);
    DCHECK(replacement != this);  // No missing EMPTY_MATCH_CHECK.
    if (replacement != nullptr) {
      alternatives_->at(i).set_node(replacement);
      surviving++;
      survivor = replacement;
    }
  }
  if (surviving < 2) return set_replacement(survivor);

  set_replacement(this);
  if (surviving == choice_count) {
    return this;
  }
  // Only some of the nodes survived the filtering.  We need to rebuild the
  // alternatives list.
  ZoneList<GuardedAlternative>* new_alternatives =
      new (zone()) ZoneList<GuardedAlternative>(surviving, zone());
  for (int i = 0; i < choice_count; i++) {
    RegExpNode* replacement =
        alternatives_->at(i).node()->FilterOneByte(depth - 1);
    if (replacement != nullptr) {
      alternatives_->at(i).set_node(replacement);
      new_alternatives->Add(alternatives_->at(i), zone());
    }
  }
  alternatives_ = new_alternatives;
  return this;
}

RegExpNode* NegativeLookaroundChoiceNode::FilterOneByte(int depth) {
  if (info()->replacement_calculated) return replacement();
  if (depth < 0) return this;
  if (info()->visited) return this;
  VisitMarker marker(info());
  // Alternative 0 is the negative lookahead, alternative 1 is what comes
  // afterwards.
1962
  RegExpNode* node = continue_node();
1963 1964
  RegExpNode* replacement = node->FilterOneByte(depth - 1);
  if (replacement == nullptr) return set_replacement(nullptr);
1965
  alternatives_->at(kContinueIndex).set_node(replacement);
1966

1967
  RegExpNode* neg_node = lookaround_node();
1968 1969 1970 1971
  RegExpNode* neg_replacement = neg_node->FilterOneByte(depth - 1);
  // If the negative lookahead is always going to fail then
  // we don't need to check it.
  if (neg_replacement == nullptr) return set_replacement(replacement);
1972
  alternatives_->at(kLookaroundIndex).set_node(neg_replacement);
1973 1974 1975 1976 1977 1978 1979 1980
  return set_replacement(this);
}

void LoopChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details,
                                          RegExpCompiler* compiler,
                                          int characters_filled_in,
                                          bool not_at_start) {
  if (body_can_be_zero_length_ || info()->visited) return;
1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022
  not_at_start = not_at_start || this->not_at_start();
  DCHECK_EQ(alternatives_->length(), 2);  // There's just loop and continue.
  if (traversed_loop_initialization_node_ && min_loop_iterations_ > 0 &&
      loop_node_->EatsAtLeast(not_at_start) >
          continue_node_->EatsAtLeast(true)) {
    // Loop body is guaranteed to execute at least once, and consume characters
    // when it does, meaning the only possible quick checks from this point
    // begin with the loop body. We may recursively visit this LoopChoiceNode,
    // but we temporarily decrease its minimum iteration counter so we know when
    // to check the continue case.
    IterationDecrementer next_iteration(this);
    loop_node_->GetQuickCheckDetails(details, compiler, characters_filled_in,
                                     not_at_start);
  } else {
    // Might not consume anything in the loop body, so treat it like a normal
    // ChoiceNode (and don't recursively visit this node again).
    VisitMarker marker(info());
    ChoiceNode::GetQuickCheckDetails(details, compiler, characters_filled_in,
                                     not_at_start);
  }
}

void LoopChoiceNode::GetQuickCheckDetailsFromLoopEntry(
    QuickCheckDetails* details, RegExpCompiler* compiler,
    int characters_filled_in, bool not_at_start) {
  if (traversed_loop_initialization_node_) {
    // We already entered this loop once, exited via its continuation node, and
    // followed an outer loop's back-edge to before the loop entry point. We
    // could try to reset the minimum iteration count to its starting value at
    // this point, but that seems like more trouble than it's worth. It's safe
    // to keep going with the current (possibly reduced) minimum iteration
    // count.
    GetQuickCheckDetails(details, compiler, characters_filled_in, not_at_start);
  } else {
    // We are entering a loop via its counter initialization action, meaning we
    // are guaranteed to run the loop body at least some minimum number of times
    // before running the continuation node. Set a flag so that this node knows
    // (now and any times we visit it again recursively) that it was entered
    // from the top.
    LoopInitializationMarker marker(this);
    GetQuickCheckDetails(details, compiler, characters_filled_in, not_at_start);
  }
2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054
}

void LoopChoiceNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
                                  BoyerMooreLookahead* bm, bool not_at_start) {
  if (body_can_be_zero_length_ || budget <= 0) {
    bm->SetRest(offset);
    SaveBMInfo(bm, not_at_start, offset);
    return;
  }
  ChoiceNode::FillInBMInfo(isolate, offset, budget - 1, bm, not_at_start);
  SaveBMInfo(bm, not_at_start, offset);
}

void ChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details,
                                      RegExpCompiler* compiler,
                                      int characters_filled_in,
                                      bool not_at_start) {
  not_at_start = (not_at_start || not_at_start_);
  int choice_count = alternatives_->length();
  DCHECK_LT(0, choice_count);
  alternatives_->at(0).node()->GetQuickCheckDetails(
      details, compiler, characters_filled_in, not_at_start);
  for (int i = 1; i < choice_count; i++) {
    QuickCheckDetails new_details(details->characters());
    RegExpNode* node = alternatives_->at(i).node();
    node->GetQuickCheckDetails(&new_details, compiler, characters_filled_in,
                               not_at_start);
    // Here we merge the quick match details of the two branches.
    details->Merge(&new_details, characters_filled_in);
  }
}

2055 2056
namespace {

2057
// Check for [0-9A-Z_a-z].
2058 2059
void EmitWordCheck(RegExpMacroAssembler* assembler, Label* word,
                   Label* non_word, bool fall_through_on_word) {
2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080
  if (assembler->CheckSpecialCharacterClass(
          fall_through_on_word ? 'w' : 'W',
          fall_through_on_word ? non_word : word)) {
    // Optimized implementation available.
    return;
  }
  assembler->CheckCharacterGT('z', non_word);
  assembler->CheckCharacterLT('0', non_word);
  assembler->CheckCharacterGT('a' - 1, word);
  assembler->CheckCharacterLT('9' + 1, word);
  assembler->CheckCharacterLT('A', non_word);
  assembler->CheckCharacterLT('Z' + 1, word);
  if (fall_through_on_word) {
    assembler->CheckNotCharacter('_', non_word);
  } else {
    assembler->CheckCharacter('_', word);
  }
}

// Emit the code to check for a ^ in multiline mode (1-character lookbehind
// that matches newline or the start of input).
2081
void EmitHat(RegExpCompiler* compiler, RegExpNode* on_success, Trace* trace) {
2082
  RegExpMacroAssembler* assembler = compiler->macro_assembler();
2083 2084

  // We will load the previous character into the current character register.
2085 2086 2087
  Trace new_trace(*trace);
  new_trace.InvalidateCurrentCharacter();

2088 2089 2090 2091 2092 2093 2094
  // A positive (> 0) cp_offset means we've already successfully matched a
  // non-empty-width part of the pattern, and thus cannot be at or before the
  // start of the subject string. We can thus skip both at-start and
  // bounds-checks when loading the one-character lookbehind.
  const bool may_be_at_or_before_subject_string_start =
      new_trace.cp_offset() <= 0;

2095
  Label ok;
2096 2097 2098
  if (may_be_at_or_before_subject_string_start) {
    // The start of input counts as a newline in this context, so skip to ok if
    // we are at the start.
2099
    assembler->CheckAtStart(new_trace.cp_offset(), &ok);
2100
  }
2101 2102 2103 2104

  // If we've already checked that we are not at the start of input, it's okay
  // to load the previous character without bounds checks.
  const bool can_skip_bounds_check = !may_be_at_or_before_subject_string_start;
2105
  assembler->LoadCurrentCharacter(new_trace.cp_offset() - 1,
2106
                                  new_trace.backtrack(), can_skip_bounds_check);
2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118
  if (!assembler->CheckSpecialCharacterClass('n', new_trace.backtrack())) {
    // Newline means \n, \r, 0x2028 or 0x2029.
    if (!compiler->one_byte()) {
      assembler->CheckCharacterAfterAnd(0x2028, 0xFFFE, &ok);
    }
    assembler->CheckCharacter('\n', &ok);
    assembler->CheckNotCharacter('\r', new_trace.backtrack());
  }
  assembler->Bind(&ok);
  on_success->Emit(compiler, &new_trace);
}

2119 2120
}  // namespace

2121 2122 2123 2124 2125 2126 2127 2128
// Emit the code to handle \b and \B (word-boundary or non-word-boundary).
void AssertionNode::EmitBoundaryCheck(RegExpCompiler* compiler, Trace* trace) {
  RegExpMacroAssembler* assembler = compiler->macro_assembler();
  Isolate* isolate = assembler->isolate();
  Trace::TriBool next_is_word_character = Trace::UNKNOWN;
  bool not_at_start = (trace->at_start() == Trace::FALSE_VALUE);
  BoyerMooreLookahead* lookahead = bm_info(not_at_start);
  if (lookahead == nullptr) {
2129 2130
    int eats_at_least =
        Min(kMaxLookaheadForBoyerMoore, EatsAtLeast(not_at_start));
2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175
    if (eats_at_least >= 1) {
      BoyerMooreLookahead* bm =
          new (zone()) BoyerMooreLookahead(eats_at_least, compiler, zone());
      FillInBMInfo(isolate, 0, kRecursionBudget, bm, not_at_start);
      if (bm->at(0)->is_non_word()) next_is_word_character = Trace::FALSE_VALUE;
      if (bm->at(0)->is_word()) next_is_word_character = Trace::TRUE_VALUE;
    }
  } else {
    if (lookahead->at(0)->is_non_word())
      next_is_word_character = Trace::FALSE_VALUE;
    if (lookahead->at(0)->is_word()) next_is_word_character = Trace::TRUE_VALUE;
  }
  bool at_boundary = (assertion_type_ == AssertionNode::AT_BOUNDARY);
  if (next_is_word_character == Trace::UNKNOWN) {
    Label before_non_word;
    Label before_word;
    if (trace->characters_preloaded() != 1) {
      assembler->LoadCurrentCharacter(trace->cp_offset(), &before_non_word);
    }
    // Fall through on non-word.
    EmitWordCheck(assembler, &before_word, &before_non_word, false);
    // Next character is not a word character.
    assembler->Bind(&before_non_word);
    Label ok;
    BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord);
    assembler->GoTo(&ok);

    assembler->Bind(&before_word);
    BacktrackIfPrevious(compiler, trace, at_boundary ? kIsWord : kIsNonWord);
    assembler->Bind(&ok);
  } else if (next_is_word_character == Trace::TRUE_VALUE) {
    BacktrackIfPrevious(compiler, trace, at_boundary ? kIsWord : kIsNonWord);
  } else {
    DCHECK(next_is_word_character == Trace::FALSE_VALUE);
    BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord);
  }
}

void AssertionNode::BacktrackIfPrevious(
    RegExpCompiler* compiler, Trace* trace,
    AssertionNode::IfPrevious backtrack_if_previous) {
  RegExpMacroAssembler* assembler = compiler->macro_assembler();
  Trace new_trace(*trace);
  new_trace.InvalidateCurrentCharacter();

2176
  Label fall_through;
2177 2178 2179 2180 2181
  Label* non_word = backtrack_if_previous == kIsNonWord ? new_trace.backtrack()
                                                        : &fall_through;
  Label* word = backtrack_if_previous == kIsNonWord ? &fall_through
                                                    : new_trace.backtrack();

2182 2183 2184 2185 2186 2187 2188 2189
  // A positive (> 0) cp_offset means we've already successfully matched a
  // non-empty-width part of the pattern, and thus cannot be at or before the
  // start of the subject string. We can thus skip both at-start and
  // bounds-checks when loading the one-character lookbehind.
  const bool may_be_at_or_before_subject_string_start =
      new_trace.cp_offset() <= 0;

  if (may_be_at_or_before_subject_string_start) {
2190 2191
    // The start of input counts as a non-word character, so the question is
    // decided if we are at the start.
2192
    assembler->CheckAtStart(new_trace.cp_offset(), non_word);
2193 2194 2195 2196 2197 2198 2199
  }

  // If we've already checked that we are not at the start of input, it's okay
  // to load the previous character without bounds checks.
  const bool can_skip_bounds_check = !may_be_at_or_before_subject_string_start;
  assembler->LoadCurrentCharacter(new_trace.cp_offset() - 1, non_word,
                                  can_skip_bounds_check);
2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649
  EmitWordCheck(assembler, word, non_word, backtrack_if_previous == kIsNonWord);

  assembler->Bind(&fall_through);
  on_success()->Emit(compiler, &new_trace);
}

void AssertionNode::GetQuickCheckDetails(QuickCheckDetails* details,
                                         RegExpCompiler* compiler,
                                         int filled_in, bool not_at_start) {
  if (assertion_type_ == AT_START && not_at_start) {
    details->set_cannot_match();
    return;
  }
  return on_success()->GetQuickCheckDetails(details, compiler, filled_in,
                                            not_at_start);
}

void AssertionNode::Emit(RegExpCompiler* compiler, Trace* trace) {
  RegExpMacroAssembler* assembler = compiler->macro_assembler();
  switch (assertion_type_) {
    case AT_END: {
      Label ok;
      assembler->CheckPosition(trace->cp_offset(), &ok);
      assembler->GoTo(trace->backtrack());
      assembler->Bind(&ok);
      break;
    }
    case AT_START: {
      if (trace->at_start() == Trace::FALSE_VALUE) {
        assembler->GoTo(trace->backtrack());
        return;
      }
      if (trace->at_start() == Trace::UNKNOWN) {
        assembler->CheckNotAtStart(trace->cp_offset(), trace->backtrack());
        Trace at_start_trace = *trace;
        at_start_trace.set_at_start(Trace::TRUE_VALUE);
        on_success()->Emit(compiler, &at_start_trace);
        return;
      }
    } break;
    case AFTER_NEWLINE:
      EmitHat(compiler, on_success(), trace);
      return;
    case AT_BOUNDARY:
    case AT_NON_BOUNDARY: {
      EmitBoundaryCheck(compiler, trace);
      return;
    }
  }
  on_success()->Emit(compiler, trace);
}

static bool DeterminedAlready(QuickCheckDetails* quick_check, int offset) {
  if (quick_check == nullptr) return false;
  if (offset >= quick_check->characters()) return false;
  return quick_check->positions(offset)->determines_perfectly;
}

static void UpdateBoundsCheck(int index, int* checked_up_to) {
  if (index > *checked_up_to) {
    *checked_up_to = index;
  }
}

// We call this repeatedly to generate code for each pass over the text node.
// The passes are in increasing order of difficulty because we hope one
// of the first passes will fail in which case we are saved the work of the
// later passes.  for example for the case independent regexp /%[asdfghjkl]a/
// we will check the '%' in the first pass, the case independent 'a' in the
// second pass and the character class in the last pass.
//
// The passes are done from right to left, so for example to test for /bar/
// we will first test for an 'r' with offset 2, then an 'a' with offset 1
// and then a 'b' with offset 0.  This means we can avoid the end-of-input
// bounds check most of the time.  In the example we only need to check for
// end-of-input when loading the putative 'r'.
//
// A slight complication involves the fact that the first character may already
// be fetched into a register by the previous node.  In this case we want to
// do the test for that character first.  We do this in separate passes.  The
// 'preloaded' argument indicates that we are doing such a 'pass'.  If such a
// pass has been performed then subsequent passes will have true in
// first_element_checked to indicate that that character does not need to be
// checked again.
//
// In addition to all this we are passed a Trace, which can
// contain an AlternativeGeneration object.  In this AlternativeGeneration
// object we can see details of any quick check that was already passed in
// order to get to the code we are now generating.  The quick check can involve
// loading characters, which means we do not need to recheck the bounds
// up to the limit the quick check already checked.  In addition the quick
// check can have involved a mask and compare operation which may simplify
// or obviate the need for further checks at some character positions.
void TextNode::TextEmitPass(RegExpCompiler* compiler, TextEmitPassType pass,
                            bool preloaded, Trace* trace,
                            bool first_element_checked, int* checked_up_to) {
  RegExpMacroAssembler* assembler = compiler->macro_assembler();
  Isolate* isolate = assembler->isolate();
  bool one_byte = compiler->one_byte();
  Label* backtrack = trace->backtrack();
  QuickCheckDetails* quick_check = trace->quick_check_performed();
  int element_count = elements()->length();
  int backward_offset = read_backward() ? -Length() : 0;
  for (int i = preloaded ? 0 : element_count - 1; i >= 0; i--) {
    TextElement elm = elements()->at(i);
    int cp_offset = trace->cp_offset() + elm.cp_offset() + backward_offset;
    if (elm.text_type() == TextElement::ATOM) {
      if (SkipPass(pass, elm.atom()->ignore_case())) continue;
      Vector<const uc16> quarks = elm.atom()->data();
      for (int j = preloaded ? 0 : quarks.length() - 1; j >= 0; j--) {
        if (first_element_checked && i == 0 && j == 0) continue;
        if (DeterminedAlready(quick_check, elm.cp_offset() + j)) continue;
        EmitCharacterFunction* emit_function = nullptr;
        uc16 quark = quarks[j];
        if (elm.atom()->ignore_case()) {
          // Everywhere else we assume that a non-Latin-1 character cannot match
          // a Latin-1 character. Avoid the cases where this is assumption is
          // invalid by using the Latin1 equivalent instead.
          quark = unibrow::Latin1::TryConvertToLatin1(quark);
        }
        switch (pass) {
          case NON_LATIN1_MATCH:
            DCHECK(one_byte);
            if (quark > String::kMaxOneByteCharCode) {
              assembler->GoTo(backtrack);
              return;
            }
            break;
          case NON_LETTER_CHARACTER_MATCH:
            emit_function = &EmitAtomNonLetter;
            break;
          case SIMPLE_CHARACTER_MATCH:
            emit_function = &EmitSimpleCharacter;
            break;
          case CASE_CHARACTER_MATCH:
            emit_function = &EmitAtomLetter;
            break;
          default:
            break;
        }
        if (emit_function != nullptr) {
          bool bounds_check = *checked_up_to < cp_offset + j || read_backward();
          bool bound_checked =
              emit_function(isolate, compiler, quark, backtrack, cp_offset + j,
                            bounds_check, preloaded);
          if (bound_checked) UpdateBoundsCheck(cp_offset + j, checked_up_to);
        }
      }
    } else {
      DCHECK_EQ(TextElement::CHAR_CLASS, elm.text_type());
      if (pass == CHARACTER_CLASS_MATCH) {
        if (first_element_checked && i == 0) continue;
        if (DeterminedAlready(quick_check, elm.cp_offset())) continue;
        RegExpCharacterClass* cc = elm.char_class();
        bool bounds_check = *checked_up_to < cp_offset || read_backward();
        EmitCharClass(assembler, cc, one_byte, backtrack, cp_offset,
                      bounds_check, preloaded, zone());
        UpdateBoundsCheck(cp_offset, checked_up_to);
      }
    }
  }
}

int TextNode::Length() {
  TextElement elm = elements()->last();
  DCHECK_LE(0, elm.cp_offset());
  return elm.cp_offset() + elm.length();
}

bool TextNode::SkipPass(TextEmitPassType pass, bool ignore_case) {
  if (ignore_case) {
    return pass == SIMPLE_CHARACTER_MATCH;
  } else {
    return pass == NON_LETTER_CHARACTER_MATCH || pass == CASE_CHARACTER_MATCH;
  }
}

TextNode* TextNode::CreateForCharacterRanges(Zone* zone,
                                             ZoneList<CharacterRange>* ranges,
                                             bool read_backward,
                                             RegExpNode* on_success,
                                             JSRegExp::Flags flags) {
  DCHECK_NOT_NULL(ranges);
  ZoneList<TextElement>* elms = new (zone) ZoneList<TextElement>(1, zone);
  elms->Add(TextElement::CharClass(
                new (zone) RegExpCharacterClass(zone, ranges, flags)),
            zone);
  return new (zone) TextNode(elms, read_backward, on_success);
}

TextNode* TextNode::CreateForSurrogatePair(Zone* zone, CharacterRange lead,
                                           CharacterRange trail,
                                           bool read_backward,
                                           RegExpNode* on_success,
                                           JSRegExp::Flags flags) {
  ZoneList<CharacterRange>* lead_ranges = CharacterRange::List(zone, lead);
  ZoneList<CharacterRange>* trail_ranges = CharacterRange::List(zone, trail);
  ZoneList<TextElement>* elms = new (zone) ZoneList<TextElement>(2, zone);
  elms->Add(TextElement::CharClass(
                new (zone) RegExpCharacterClass(zone, lead_ranges, flags)),
            zone);
  elms->Add(TextElement::CharClass(
                new (zone) RegExpCharacterClass(zone, trail_ranges, flags)),
            zone);
  return new (zone) TextNode(elms, read_backward, on_success);
}

// This generates the code to match a text node.  A text node can contain
// straight character sequences (possibly to be matched in a case-independent
// way) and character classes.  For efficiency we do not do this in a single
// pass from left to right.  Instead we pass over the text node several times,
// emitting code for some character positions every time.  See the comment on
// TextEmitPass for details.
void TextNode::Emit(RegExpCompiler* compiler, Trace* trace) {
  LimitResult limit_result = LimitVersions(compiler, trace);
  if (limit_result == DONE) return;
  DCHECK(limit_result == CONTINUE);

  if (trace->cp_offset() + Length() > RegExpMacroAssembler::kMaxCPOffset) {
    compiler->SetRegExpTooBig();
    return;
  }

  if (compiler->one_byte()) {
    int dummy = 0;
    TextEmitPass(compiler, NON_LATIN1_MATCH, false, trace, false, &dummy);
  }

  bool first_elt_done = false;
  int bound_checked_to = trace->cp_offset() - 1;
  bound_checked_to += trace->bound_checked_up_to();

  // If a character is preloaded into the current character register then
  // check that now.
  if (trace->characters_preloaded() == 1) {
    for (int pass = kFirstRealPass; pass <= kLastPass; pass++) {
      TextEmitPass(compiler, static_cast<TextEmitPassType>(pass), true, trace,
                   false, &bound_checked_to);
    }
    first_elt_done = true;
  }

  for (int pass = kFirstRealPass; pass <= kLastPass; pass++) {
    TextEmitPass(compiler, static_cast<TextEmitPassType>(pass), false, trace,
                 first_elt_done, &bound_checked_to);
  }

  Trace successor_trace(*trace);
  // If we advance backward, we may end up at the start.
  successor_trace.AdvanceCurrentPositionInTrace(
      read_backward() ? -Length() : Length(), compiler);
  successor_trace.set_at_start(read_backward() ? Trace::UNKNOWN
                                               : Trace::FALSE_VALUE);
  RecursionCheck rc(compiler);
  on_success()->Emit(compiler, &successor_trace);
}

void Trace::InvalidateCurrentCharacter() { characters_preloaded_ = 0; }

void Trace::AdvanceCurrentPositionInTrace(int by, RegExpCompiler* compiler) {
  // We don't have an instruction for shifting the current character register
  // down or for using a shifted value for anything so lets just forget that
  // we preloaded any characters into it.
  characters_preloaded_ = 0;
  // Adjust the offsets of the quick check performed information.  This
  // information is used to find out what we already determined about the
  // characters by means of mask and compare.
  quick_check_performed_.Advance(by, compiler->one_byte());
  cp_offset_ += by;
  if (cp_offset_ > RegExpMacroAssembler::kMaxCPOffset) {
    compiler->SetRegExpTooBig();
    cp_offset_ = 0;
  }
  bound_checked_up_to_ = Max(0, bound_checked_up_to_ - by);
}

void TextNode::MakeCaseIndependent(Isolate* isolate, bool is_one_byte) {
  int element_count = elements()->length();
  for (int i = 0; i < element_count; i++) {
    TextElement elm = elements()->at(i);
    if (elm.text_type() == TextElement::CHAR_CLASS) {
      RegExpCharacterClass* cc = elm.char_class();
#ifdef V8_INTL_SUPPORT
      bool case_equivalents_already_added =
          NeedsUnicodeCaseEquivalents(cc->flags());
#else
      bool case_equivalents_already_added = false;
#endif
      if (IgnoreCase(cc->flags()) && !case_equivalents_already_added) {
        // None of the standard character classes is different in the case
        // independent case and it slows us down if we don't know that.
        if (cc->is_standard(zone())) continue;
        ZoneList<CharacterRange>* ranges = cc->ranges(zone());
        CharacterRange::AddCaseEquivalents(isolate, zone(), ranges,
                                           is_one_byte);
      }
    }
  }
}

int TextNode::GreedyLoopTextLength() { return Length(); }

RegExpNode* TextNode::GetSuccessorOfOmnivorousTextNode(
    RegExpCompiler* compiler) {
  if (read_backward()) return nullptr;
  if (elements()->length() != 1) return nullptr;
  TextElement elm = elements()->at(0);
  if (elm.text_type() != TextElement::CHAR_CLASS) return nullptr;
  RegExpCharacterClass* node = elm.char_class();
  ZoneList<CharacterRange>* ranges = node->ranges(zone());
  CharacterRange::Canonicalize(ranges);
  if (node->is_negated()) {
    return ranges->length() == 0 ? on_success() : nullptr;
  }
  if (ranges->length() != 1) return nullptr;
  uint32_t max_char;
  if (compiler->one_byte()) {
    max_char = String::kMaxOneByteCharCode;
  } else {
    max_char = String::kMaxUtf16CodeUnit;
  }
  return ranges->at(0).IsEverything(max_char) ? on_success() : nullptr;
}

// Finds the fixed match length of a sequence of nodes that goes from
// this alternative and back to this choice node.  If there are variable
// length nodes or other complications in the way then return a sentinel
// value indicating that a greedy loop cannot be constructed.
int ChoiceNode::GreedyLoopTextLengthForAlternative(
    GuardedAlternative* alternative) {
  int length = 0;
  RegExpNode* node = alternative->node();
  // Later we will generate code for all these text nodes using recursion
  // so we have to limit the max number.
  int recursion_depth = 0;
  while (node != this) {
    if (recursion_depth++ > RegExpCompiler::kMaxRecursion) {
      return kNodeIsTooComplexForGreedyLoops;
    }
    int node_length = node->GreedyLoopTextLength();
    if (node_length == kNodeIsTooComplexForGreedyLoops) {
      return kNodeIsTooComplexForGreedyLoops;
    }
    length += node_length;
    SeqRegExpNode* seq_node = static_cast<SeqRegExpNode*>(node);
    node = seq_node->on_success();
  }
  return read_backward() ? -length : length;
}

void LoopChoiceNode::AddLoopAlternative(GuardedAlternative alt) {
  DCHECK_NULL(loop_node_);
  AddAlternative(alt);
  loop_node_ = alt.node();
}

void LoopChoiceNode::AddContinueAlternative(GuardedAlternative alt) {
  DCHECK_NULL(continue_node_);
  AddAlternative(alt);
  continue_node_ = alt.node();
}

void LoopChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
  RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
  if (trace->stop_node() == this) {
    // Back edge of greedy optimized loop node graph.
    int text_length =
        GreedyLoopTextLengthForAlternative(&(alternatives_->at(0)));
    DCHECK_NE(kNodeIsTooComplexForGreedyLoops, text_length);
    // Update the counter-based backtracking info on the stack.  This is an
    // optimization for greedy loops (see below).
    DCHECK(trace->cp_offset() == text_length);
    macro_assembler->AdvanceCurrentPosition(text_length);
    macro_assembler->GoTo(trace->loop_label());
    return;
  }
  DCHECK_NULL(trace->stop_node());
  if (!trace->is_trivial()) {
    trace->Flush(compiler, this);
    return;
  }
  ChoiceNode::Emit(compiler, trace);
}

int ChoiceNode::CalculatePreloadCharacters(RegExpCompiler* compiler,
                                           int eats_at_least) {
  int preload_characters = Min(4, eats_at_least);
  DCHECK_LE(preload_characters, 4);
  if (compiler->macro_assembler()->CanReadUnaligned()) {
    bool one_byte = compiler->one_byte();
    if (one_byte) {
      // We can't preload 3 characters because there is no machine instruction
      // to do that.  We can't just load 4 because we could be reading
      // beyond the end of the string, which could cause a memory fault.
      if (preload_characters == 3) preload_characters = 2;
    } else {
      if (preload_characters > 2) preload_characters = 2;
    }
  } else {
    if (preload_characters > 1) preload_characters = 1;
  }
  return preload_characters;
}

// This class is used when generating the alternatives in a choice node.  It
// records the way the alternative is being code generated.
class AlternativeGeneration : public Malloced {
 public:
  AlternativeGeneration()
      : possible_success(),
        expects_preload(false),
        after(),
        quick_check_details() {}
  Label possible_success;
  bool expects_preload;
  Label after;
  QuickCheckDetails quick_check_details;
};

// Creates a list of AlternativeGenerations.  If the list has a reasonable
// size then it is on the stack, otherwise the excess is on the heap.
class AlternativeGenerationList {
 public:
  AlternativeGenerationList(int count, Zone* zone) : alt_gens_(count, zone) {
    for (int i = 0; i < count && i < kAFew; i++) {
      alt_gens_.Add(a_few_alt_gens_ + i, zone);
    }
    for (int i = kAFew; i < count; i++) {
      alt_gens_.Add(new AlternativeGeneration(), zone);
    }
  }
  ~AlternativeGenerationList() {
    for (int i = kAFew; i < alt_gens_.length(); i++) {
      delete alt_gens_[i];
      alt_gens_[i] = nullptr;
    }
  }

  AlternativeGeneration* at(int i) { return alt_gens_[i]; }

 private:
  static const int kAFew = 10;
  ZoneList<AlternativeGeneration*> alt_gens_;
  AlternativeGeneration a_few_alt_gens_[kAFew];
};

void BoyerMoorePositionInfo::Set(int character) {
  SetInterval(Interval(character, character));
}

2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672
namespace {

ContainedInLattice AddRange(ContainedInLattice containment, const int* ranges,
                            int ranges_length, Interval new_range) {
  DCHECK_EQ(1, ranges_length & 1);
  DCHECK_EQ(String::kMaxCodePoint + 1, ranges[ranges_length - 1]);
  if (containment == kLatticeUnknown) return containment;
  bool inside = false;
  int last = 0;
  for (int i = 0; i < ranges_length; inside = !inside, last = ranges[i], i++) {
    // Consider the range from last to ranges[i].
    // We haven't got to the new range yet.
    if (ranges[i] <= new_range.from()) continue;
    // New range is wholly inside last-ranges[i].  Note that new_range.to() is
    // inclusive, but the values in ranges are not.
    if (last <= new_range.from() && new_range.to() < ranges[i]) {
      return Combine(containment, inside ? kLatticeIn : kLatticeOut);
    }
    return kLatticeUnknown;
  }
  return containment;
}

2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698
int BitsetFirstSetBit(BoyerMoorePositionInfo::Bitset bitset) {
  STATIC_ASSERT(BoyerMoorePositionInfo::kMapSize ==
                2 * kInt64Size * kBitsPerByte);

  // Slight fiddling is needed here, since the bitset is of length 128 while
  // CountTrailingZeros requires an integral type and std::bitset can only
  // convert to unsigned long long. So we handle the most- and least-significant
  // bits separately.

  {
    static constexpr BoyerMoorePositionInfo::Bitset mask(~uint64_t{0});
    BoyerMoorePositionInfo::Bitset masked_bitset = bitset & mask;
    STATIC_ASSERT(kInt64Size >= sizeof(decltype(masked_bitset.to_ullong())));
    uint64_t lsb = masked_bitset.to_ullong();
    if (lsb != 0) return base::bits::CountTrailingZeros(lsb);
  }

  {
    BoyerMoorePositionInfo::Bitset masked_bitset = bitset >> 64;
    uint64_t msb = masked_bitset.to_ullong();
    if (msb != 0) return 64 + base::bits::CountTrailingZeros(msb);
  }

  return -1;
}

2699 2700
}  // namespace

2701 2702
void BoyerMoorePositionInfo::SetInterval(const Interval& interval) {
  w_ = AddRange(w_, kWordRanges, kWordRangeCount, interval);
2703 2704 2705 2706

  if (interval.size() >= kMapSize) {
    map_count_ = kMapSize;
    map_.set();
2707 2708
    return;
  }
2709

2710 2711
  for (int i = interval.from(); i <= interval.to(); i++) {
    int mod_character = (i & kMask);
2712
    if (!map_[mod_character]) {
2713
      map_count_++;
2714
      map_.set(mod_character);
2715 2716 2717 2718 2719 2720
    }
    if (map_count_ == kMapSize) return;
  }
}

void BoyerMoorePositionInfo::SetAll() {
2721
  w_ = kLatticeUnknown;
2722 2723
  if (map_count_ != kMapSize) {
    map_count_ = kMapSize;
2724
    map_.set();
2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737
  }
}

BoyerMooreLookahead::BoyerMooreLookahead(int length, RegExpCompiler* compiler,
                                         Zone* zone)
    : length_(length), compiler_(compiler) {
  if (compiler->one_byte()) {
    max_char_ = String::kMaxOneByteCharCode;
  } else {
    max_char_ = String::kMaxUtf16CodeUnit;
  }
  bitmaps_ = new (zone) ZoneList<BoyerMoorePositionInfo*>(length, zone);
  for (int i = 0; i < length; i++) {
2738
    bitmaps_->Add(new (zone) BoyerMoorePositionInfo(), zone);
2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773
  }
}

// Find the longest range of lookahead that has the fewest number of different
// characters that can occur at a given position.  Since we are optimizing two
// different parameters at once this is a tradeoff.
bool BoyerMooreLookahead::FindWorthwhileInterval(int* from, int* to) {
  int biggest_points = 0;
  // If more than 32 characters out of 128 can occur it is unlikely that we can
  // be lucky enough to step forwards much of the time.
  const int kMaxMax = 32;
  for (int max_number_of_chars = 4; max_number_of_chars < kMaxMax;
       max_number_of_chars *= 2) {
    biggest_points =
        FindBestInterval(max_number_of_chars, biggest_points, from, to);
  }
  if (biggest_points == 0) return false;
  return true;
}

// Find the highest-points range between 0 and length_ where the character
// information is not too vague.  'Too vague' means that there are more than
// max_number_of_chars that can occur at this position.  Calculates the number
// of points as the product of width-of-the-range and
// probability-of-finding-one-of-the-characters, where the probability is
// calculated using the frequency distribution of the sample subject string.
int BoyerMooreLookahead::FindBestInterval(int max_number_of_chars,
                                          int old_biggest_points, int* from,
                                          int* to) {
  int biggest_points = old_biggest_points;
  static const int kSize = RegExpMacroAssembler::kTableSize;
  for (int i = 0; i < length_;) {
    while (i < length_ && Count(i) > max_number_of_chars) i++;
    if (i == length_) break;
    int remembered_from = i;
2774 2775 2776 2777

    BoyerMoorePositionInfo::Bitset union_bitset;
    for (; i < length_ && Count(i) <= max_number_of_chars; i++) {
      union_bitset |= bitmaps_->at(i)->raw_bitset();
2778
    }
2779

2780
    int frequency = 0;
2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792

    // Iterate only over set bits.
    int j;
    while ((j = BitsetFirstSetBit(union_bitset)) != -1) {
      DCHECK(union_bitset[j]);  // Sanity check.
      // Add 1 to the frequency to give a small per-character boost for
      // the cases where our sampling is not good enough and many
      // characters have a frequency of zero.  This means the frequency
      // can theoretically be up to 2*kSize though we treat it mostly as
      // a fraction of kSize.
      frequency += compiler_->frequency_collator()->Frequency(j) + 1;
      union_bitset.reset(j);
2793
    }
2794

2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825
    // We use the probability of skipping times the distance we are skipping to
    // judge the effectiveness of this.  Actually we have a cut-off:  By
    // dividing by 2 we switch off the skipping if the probability of skipping
    // is less than 50%.  This is because the multibyte mask-and-compare
    // skipping in quickcheck is more likely to do well on this case.
    bool in_quickcheck_range =
        ((i - remembered_from < 4) ||
         (compiler_->one_byte() ? remembered_from <= 4 : remembered_from <= 2));
    // Called 'probability' but it is only a rough estimate and can actually
    // be outside the 0-kSize range.
    int probability = (in_quickcheck_range ? kSize / 2 : kSize) - frequency;
    int points = (i - remembered_from) * probability;
    if (points > biggest_points) {
      *from = remembered_from;
      *to = i - 1;
      biggest_points = points;
    }
  }
  return biggest_points;
}

// Take all the characters that will not prevent a successful match if they
// occur in the subject string in the range between min_lookahead and
// max_lookahead (inclusive) measured from the current position.  If the
// character at max_lookahead offset is not one of these characters, then we
// can safely skip forwards by the number of characters in the range.
int BoyerMooreLookahead::GetSkipTable(int min_lookahead, int max_lookahead,
                                      Handle<ByteArray> boolean_skip_table) {
  const int kSkipArrayEntry = 0;
  const int kDontSkipArrayEntry = 1;

2826 2827
  std::memset(boolean_skip_table->GetDataStartAddress(), kSkipArrayEntry,
              boolean_skip_table->length());
2828 2829

  for (int i = max_lookahead; i >= min_lookahead; i--) {
2830 2831 2832 2833 2834 2835 2836 2837
    BoyerMoorePositionInfo::Bitset bitset = bitmaps_->at(i)->raw_bitset();

    // Iterate only over set bits.
    int j;
    while ((j = BitsetFirstSetBit(bitset)) != -1) {
      DCHECK(bitset[j]);  // Sanity check.
      boolean_skip_table->set(j, kDontSkipArrayEntry);
      bitset.reset(j);
2838 2839 2840
    }
  }

2841
  const int skip = max_lookahead + 1 - min_lookahead;
2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853
  return skip;
}

// See comment above on the implementation of GetSkipTable.
void BoyerMooreLookahead::EmitSkipInstructions(RegExpMacroAssembler* masm) {
  const int kSize = RegExpMacroAssembler::kTableSize;

  int min_lookahead = 0;
  int max_lookahead = 0;

  if (!FindWorthwhileInterval(&min_lookahead, &max_lookahead)) return;

2854 2855
  // Check if we only have a single non-empty position info, and that info
  // contains precisely one character.
2856 2857 2858 2859
  bool found_single_character = false;
  int single_character = 0;
  for (int i = max_lookahead; i >= min_lookahead; i--) {
    BoyerMoorePositionInfo* map = bitmaps_->at(i);
2860 2861 2862
    if (map->map_count() == 0) continue;

    if (found_single_character || map->map_count() > 1) {
2863 2864 2865
      found_single_character = false;
      break;
    }
2866 2867 2868 2869 2870 2871 2872 2873

    DCHECK(!found_single_character);
    DCHECK_EQ(map->map_count(), 1);

    found_single_character = true;
    single_character = BitsetFirstSetBit(map->raw_bitset());

    DCHECK_NE(single_character, -1);
2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012
  }

  int lookahead_width = max_lookahead + 1 - min_lookahead;

  if (found_single_character && lookahead_width == 1 && max_lookahead < 3) {
    // The mask-compare can probably handle this better.
    return;
  }

  if (found_single_character) {
    Label cont, again;
    masm->Bind(&again);
    masm->LoadCurrentCharacter(max_lookahead, &cont, true);
    if (max_char_ > kSize) {
      masm->CheckCharacterAfterAnd(single_character,
                                   RegExpMacroAssembler::kTableMask, &cont);
    } else {
      masm->CheckCharacter(single_character, &cont);
    }
    masm->AdvanceCurrentPosition(lookahead_width);
    masm->GoTo(&again);
    masm->Bind(&cont);
    return;
  }

  Factory* factory = masm->isolate()->factory();
  Handle<ByteArray> boolean_skip_table =
      factory->NewByteArray(kSize, AllocationType::kOld);
  int skip_distance =
      GetSkipTable(min_lookahead, max_lookahead, boolean_skip_table);
  DCHECK_NE(0, skip_distance);

  Label cont, again;
  masm->Bind(&again);
  masm->LoadCurrentCharacter(max_lookahead, &cont, true);
  masm->CheckBitInTable(boolean_skip_table, &cont);
  masm->AdvanceCurrentPosition(skip_distance);
  masm->GoTo(&again);
  masm->Bind(&cont);
}

/* Code generation for choice nodes.
 *
 * We generate quick checks that do a mask and compare to eliminate a
 * choice.  If the quick check succeeds then it jumps to the continuation to
 * do slow checks and check subsequent nodes.  If it fails (the common case)
 * it falls through to the next choice.
 *
 * Here is the desired flow graph.  Nodes directly below each other imply
 * fallthrough.  Alternatives 1 and 2 have quick checks.  Alternative
 * 3 doesn't have a quick check so we have to call the slow check.
 * Nodes are marked Qn for quick checks and Sn for slow checks.  The entire
 * regexp continuation is generated directly after the Sn node, up to the
 * next GoTo if we decide to reuse some already generated code.  Some
 * nodes expect preload_characters to be preloaded into the current
 * character register.  R nodes do this preloading.  Vertices are marked
 * F for failures and S for success (possible success in the case of quick
 * nodes).  L, V, < and > are used as arrow heads.
 *
 * ----------> R
 *             |
 *             V
 *            Q1 -----> S1
 *             |   S   /
 *            F|      /
 *             |    F/
 *             |    /
 *             |   R
 *             |  /
 *             V L
 *            Q2 -----> S2
 *             |   S   /
 *            F|      /
 *             |    F/
 *             |    /
 *             |   R
 *             |  /
 *             V L
 *            S3
 *             |
 *            F|
 *             |
 *             R
 *             |
 * backtrack   V
 * <----------Q4
 *   \    F    |
 *    \        |S
 *     \   F   V
 *      \-----S4
 *
 * For greedy loops we push the current position, then generate the code that
 * eats the input specially in EmitGreedyLoop.  The other choice (the
 * continuation) is generated by the normal code in EmitChoices, and steps back
 * in the input to the starting position when it fails to match.  The loop code
 * looks like this (U is the unwind code that steps back in the greedy loop).
 *
 *              _____
 *             /     \
 *             V     |
 * ----------> S1    |
 *            /|     |
 *           / |S    |
 *         F/  \_____/
 *         /
 *        |<-----
 *        |      \
 *        V       |S
 *        Q2 ---> U----->backtrack
 *        |  F   /
 *       S|     /
 *        V  F /
 *        S2--/
 */

GreedyLoopState::GreedyLoopState(bool not_at_start) {
  counter_backtrack_trace_.set_backtrack(&label_);
  if (not_at_start) counter_backtrack_trace_.set_at_start(Trace::FALSE_VALUE);
}

void ChoiceNode::AssertGuardsMentionRegisters(Trace* trace) {
#ifdef DEBUG
  int choice_count = alternatives_->length();
  for (int i = 0; i < choice_count - 1; i++) {
    GuardedAlternative alternative = alternatives_->at(i);
    ZoneList<Guard*>* guards = alternative.guards();
    int guard_count = (guards == nullptr) ? 0 : guards->length();
    for (int j = 0; j < guard_count; j++) {
      DCHECK(!trace->mentions_reg(guards->at(j)->reg()));
    }
  }
#endif
}

void ChoiceNode::SetUpPreLoad(RegExpCompiler* compiler, Trace* current_trace,
                              PreloadState* state) {
  if (state->eats_at_least_ == PreloadState::kEatsAtLeastNotYetInitialized) {
    // Save some time by looking at most one machine word ahead.
    state->eats_at_least_ =
3013
        EatsAtLeast(current_trace->at_start() == Trace::FALSE_VALUE);
3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162
  }
  state->preload_characters_ =
      CalculatePreloadCharacters(compiler, state->eats_at_least_);

  state->preload_is_current_ =
      (current_trace->characters_preloaded() == state->preload_characters_);
  state->preload_has_checked_bounds_ = state->preload_is_current_;
}

void ChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
  int choice_count = alternatives_->length();

  if (choice_count == 1 && alternatives_->at(0).guards() == nullptr) {
    alternatives_->at(0).node()->Emit(compiler, trace);
    return;
  }

  AssertGuardsMentionRegisters(trace);

  LimitResult limit_result = LimitVersions(compiler, trace);
  if (limit_result == DONE) return;
  DCHECK(limit_result == CONTINUE);

  // For loop nodes we already flushed (see LoopChoiceNode::Emit), but for
  // other choice nodes we only flush if we are out of code size budget.
  if (trace->flush_budget() == 0 && trace->actions() != nullptr) {
    trace->Flush(compiler, this);
    return;
  }

  RecursionCheck rc(compiler);

  PreloadState preload;
  preload.init();
  GreedyLoopState greedy_loop_state(not_at_start());

  int text_length = GreedyLoopTextLengthForAlternative(&alternatives_->at(0));
  AlternativeGenerationList alt_gens(choice_count, zone());

  if (choice_count > 1 && text_length != kNodeIsTooComplexForGreedyLoops) {
    trace = EmitGreedyLoop(compiler, trace, &alt_gens, &preload,
                           &greedy_loop_state, text_length);
  } else {
    // TODO(erikcorry): Delete this.  We don't need this label, but it makes us
    // match the traces produced pre-cleanup.
    Label second_choice;
    compiler->macro_assembler()->Bind(&second_choice);

    preload.eats_at_least_ = EmitOptimizedUnanchoredSearch(compiler, trace);

    EmitChoices(compiler, &alt_gens, 0, trace, &preload);
  }

  // At this point we need to generate slow checks for the alternatives where
  // the quick check was inlined.  We can recognize these because the associated
  // label was bound.
  int new_flush_budget = trace->flush_budget() / choice_count;
  for (int i = 0; i < choice_count; i++) {
    AlternativeGeneration* alt_gen = alt_gens.at(i);
    Trace new_trace(*trace);
    // If there are actions to be flushed we have to limit how many times
    // they are flushed.  Take the budget of the parent trace and distribute
    // it fairly amongst the children.
    if (new_trace.actions() != nullptr) {
      new_trace.set_flush_budget(new_flush_budget);
    }
    bool next_expects_preload =
        i == choice_count - 1 ? false : alt_gens.at(i + 1)->expects_preload;
    EmitOutOfLineContinuation(compiler, &new_trace, alternatives_->at(i),
                              alt_gen, preload.preload_characters_,
                              next_expects_preload);
  }
}

Trace* ChoiceNode::EmitGreedyLoop(RegExpCompiler* compiler, Trace* trace,
                                  AlternativeGenerationList* alt_gens,
                                  PreloadState* preload,
                                  GreedyLoopState* greedy_loop_state,
                                  int text_length) {
  RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
  // Here we have special handling for greedy loops containing only text nodes
  // and other simple nodes.  These are handled by pushing the current
  // position on the stack and then incrementing the current position each
  // time around the switch.  On backtrack we decrement the current position
  // and check it against the pushed value.  This avoids pushing backtrack
  // information for each iteration of the loop, which could take up a lot of
  // space.
  DCHECK(trace->stop_node() == nullptr);
  macro_assembler->PushCurrentPosition();
  Label greedy_match_failed;
  Trace greedy_match_trace;
  if (not_at_start()) greedy_match_trace.set_at_start(Trace::FALSE_VALUE);
  greedy_match_trace.set_backtrack(&greedy_match_failed);
  Label loop_label;
  macro_assembler->Bind(&loop_label);
  greedy_match_trace.set_stop_node(this);
  greedy_match_trace.set_loop_label(&loop_label);
  alternatives_->at(0).node()->Emit(compiler, &greedy_match_trace);
  macro_assembler->Bind(&greedy_match_failed);

  Label second_choice;  // For use in greedy matches.
  macro_assembler->Bind(&second_choice);

  Trace* new_trace = greedy_loop_state->counter_backtrack_trace();

  EmitChoices(compiler, alt_gens, 1, new_trace, preload);

  macro_assembler->Bind(greedy_loop_state->label());
  // If we have unwound to the bottom then backtrack.
  macro_assembler->CheckGreedyLoop(trace->backtrack());
  // Otherwise try the second priority at an earlier position.
  macro_assembler->AdvanceCurrentPosition(-text_length);
  macro_assembler->GoTo(&second_choice);
  return new_trace;
}

int ChoiceNode::EmitOptimizedUnanchoredSearch(RegExpCompiler* compiler,
                                              Trace* trace) {
  int eats_at_least = PreloadState::kEatsAtLeastNotYetInitialized;
  if (alternatives_->length() != 2) return eats_at_least;

  GuardedAlternative alt1 = alternatives_->at(1);
  if (alt1.guards() != nullptr && alt1.guards()->length() != 0) {
    return eats_at_least;
  }
  RegExpNode* eats_anything_node = alt1.node();
  if (eats_anything_node->GetSuccessorOfOmnivorousTextNode(compiler) != this) {
    return eats_at_least;
  }

  // Really we should be creating a new trace when we execute this function,
  // but there is no need, because the code it generates cannot backtrack, and
  // we always arrive here with a trivial trace (since it's the entry to a
  // loop.  That also implies that there are no preloaded characters, which is
  // good, because it means we won't be violating any assumptions by
  // overwriting those characters with new load instructions.
  DCHECK(trace->is_trivial());

  RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
  Isolate* isolate = macro_assembler->isolate();
  // At this point we know that we are at a non-greedy loop that will eat
  // any character one at a time.  Any non-anchored regexp has such a
  // loop prepended to it in order to find where it starts.  We look for
  // a pattern of the form ...abc... where we can look 6 characters ahead
  // and step forwards 3 if the character is not one of abc.  Abc need
  // not be atoms, they can be any reasonably limited character class or
  // small alternation.
  BoyerMooreLookahead* bm = bm_info(false);
  if (bm == nullptr) {
3163
    eats_at_least = Min(kMaxLookaheadForBoyerMoore, EatsAtLeast(false));
3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214
    if (eats_at_least >= 1) {
      bm = new (zone()) BoyerMooreLookahead(eats_at_least, compiler, zone());
      GuardedAlternative alt0 = alternatives_->at(0);
      alt0.node()->FillInBMInfo(isolate, 0, kRecursionBudget, bm, false);
    }
  }
  if (bm != nullptr) {
    bm->EmitSkipInstructions(macro_assembler);
  }
  return eats_at_least;
}

void ChoiceNode::EmitChoices(RegExpCompiler* compiler,
                             AlternativeGenerationList* alt_gens,
                             int first_choice, Trace* trace,
                             PreloadState* preload) {
  RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
  SetUpPreLoad(compiler, trace, preload);

  // For now we just call all choices one after the other.  The idea ultimately
  // is to use the Dispatch table to try only the relevant ones.
  int choice_count = alternatives_->length();

  int new_flush_budget = trace->flush_budget() / choice_count;

  for (int i = first_choice; i < choice_count; i++) {
    bool is_last = i == choice_count - 1;
    bool fall_through_on_failure = !is_last;
    GuardedAlternative alternative = alternatives_->at(i);
    AlternativeGeneration* alt_gen = alt_gens->at(i);
    alt_gen->quick_check_details.set_characters(preload->preload_characters_);
    ZoneList<Guard*>* guards = alternative.guards();
    int guard_count = (guards == nullptr) ? 0 : guards->length();
    Trace new_trace(*trace);
    new_trace.set_characters_preloaded(
        preload->preload_is_current_ ? preload->preload_characters_ : 0);
    if (preload->preload_has_checked_bounds_) {
      new_trace.set_bound_checked_up_to(preload->preload_characters_);
    }
    new_trace.quick_check_performed()->Clear();
    if (not_at_start_) new_trace.set_at_start(Trace::FALSE_VALUE);
    if (!is_last) {
      new_trace.set_backtrack(&alt_gen->after);
    }
    alt_gen->expects_preload = preload->preload_is_current_;
    bool generate_full_check_inline = false;
    if (compiler->optimize() &&
        try_to_emit_quick_check_for_alternative(i == 0) &&
        alternative.node()->EmitQuickCheck(
            compiler, trace, &new_trace, preload->preload_has_checked_bounds_,
            &alt_gen->possible_success, &alt_gen->quick_check_details,
3215
            fall_through_on_failure, this)) {
3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323
      // Quick check was generated for this choice.
      preload->preload_is_current_ = true;
      preload->preload_has_checked_bounds_ = true;
      // If we generated the quick check to fall through on possible success,
      // we now need to generate the full check inline.
      if (!fall_through_on_failure) {
        macro_assembler->Bind(&alt_gen->possible_success);
        new_trace.set_quick_check_performed(&alt_gen->quick_check_details);
        new_trace.set_characters_preloaded(preload->preload_characters_);
        new_trace.set_bound_checked_up_to(preload->preload_characters_);
        generate_full_check_inline = true;
      }
    } else if (alt_gen->quick_check_details.cannot_match()) {
      if (!fall_through_on_failure) {
        macro_assembler->GoTo(trace->backtrack());
      }
      continue;
    } else {
      // No quick check was generated.  Put the full code here.
      // If this is not the first choice then there could be slow checks from
      // previous cases that go here when they fail.  There's no reason to
      // insist that they preload characters since the slow check we are about
      // to generate probably can't use it.
      if (i != first_choice) {
        alt_gen->expects_preload = false;
        new_trace.InvalidateCurrentCharacter();
      }
      generate_full_check_inline = true;
    }
    if (generate_full_check_inline) {
      if (new_trace.actions() != nullptr) {
        new_trace.set_flush_budget(new_flush_budget);
      }
      for (int j = 0; j < guard_count; j++) {
        GenerateGuard(macro_assembler, guards->at(j), &new_trace);
      }
      alternative.node()->Emit(compiler, &new_trace);
      preload->preload_is_current_ = false;
    }
    macro_assembler->Bind(&alt_gen->after);
  }
}

void ChoiceNode::EmitOutOfLineContinuation(RegExpCompiler* compiler,
                                           Trace* trace,
                                           GuardedAlternative alternative,
                                           AlternativeGeneration* alt_gen,
                                           int preload_characters,
                                           bool next_expects_preload) {
  if (!alt_gen->possible_success.is_linked()) return;

  RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
  macro_assembler->Bind(&alt_gen->possible_success);
  Trace out_of_line_trace(*trace);
  out_of_line_trace.set_characters_preloaded(preload_characters);
  out_of_line_trace.set_quick_check_performed(&alt_gen->quick_check_details);
  if (not_at_start_) out_of_line_trace.set_at_start(Trace::FALSE_VALUE);
  ZoneList<Guard*>* guards = alternative.guards();
  int guard_count = (guards == nullptr) ? 0 : guards->length();
  if (next_expects_preload) {
    Label reload_current_char;
    out_of_line_trace.set_backtrack(&reload_current_char);
    for (int j = 0; j < guard_count; j++) {
      GenerateGuard(macro_assembler, guards->at(j), &out_of_line_trace);
    }
    alternative.node()->Emit(compiler, &out_of_line_trace);
    macro_assembler->Bind(&reload_current_char);
    // Reload the current character, since the next quick check expects that.
    // We don't need to check bounds here because we only get into this
    // code through a quick check which already did the checked load.
    macro_assembler->LoadCurrentCharacter(trace->cp_offset(), nullptr, false,
                                          preload_characters);
    macro_assembler->GoTo(&(alt_gen->after));
  } else {
    out_of_line_trace.set_backtrack(&(alt_gen->after));
    for (int j = 0; j < guard_count; j++) {
      GenerateGuard(macro_assembler, guards->at(j), &out_of_line_trace);
    }
    alternative.node()->Emit(compiler, &out_of_line_trace);
  }
}

void ActionNode::Emit(RegExpCompiler* compiler, Trace* trace) {
  RegExpMacroAssembler* assembler = compiler->macro_assembler();
  LimitResult limit_result = LimitVersions(compiler, trace);
  if (limit_result == DONE) return;
  DCHECK(limit_result == CONTINUE);

  RecursionCheck rc(compiler);

  switch (action_type_) {
    case STORE_POSITION: {
      Trace::DeferredCapture new_capture(data_.u_position_register.reg,
                                         data_.u_position_register.is_capture,
                                         trace);
      Trace new_trace = *trace;
      new_trace.add_action(&new_capture);
      on_success()->Emit(compiler, &new_trace);
      break;
    }
    case INCREMENT_REGISTER: {
      Trace::DeferredIncrementRegister new_increment(
          data_.u_increment_register.reg);
      Trace new_trace = *trace;
      new_trace.add_action(&new_increment);
      on_success()->Emit(compiler, &new_trace);
      break;
    }
3324 3325 3326
    case SET_REGISTER_FOR_LOOP: {
      Trace::DeferredSetRegisterForLoop new_set(data_.u_store_register.reg,
                                                data_.u_store_register.value);
3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459
      Trace new_trace = *trace;
      new_trace.add_action(&new_set);
      on_success()->Emit(compiler, &new_trace);
      break;
    }
    case CLEAR_CAPTURES: {
      Trace::DeferredClearCaptures new_capture(Interval(
          data_.u_clear_captures.range_from, data_.u_clear_captures.range_to));
      Trace new_trace = *trace;
      new_trace.add_action(&new_capture);
      on_success()->Emit(compiler, &new_trace);
      break;
    }
    case BEGIN_SUBMATCH:
      if (!trace->is_trivial()) {
        trace->Flush(compiler, this);
      } else {
        assembler->WriteCurrentPositionToRegister(
            data_.u_submatch.current_position_register, 0);
        assembler->WriteStackPointerToRegister(
            data_.u_submatch.stack_pointer_register);
        on_success()->Emit(compiler, trace);
      }
      break;
    case EMPTY_MATCH_CHECK: {
      int start_pos_reg = data_.u_empty_match_check.start_register;
      int stored_pos = 0;
      int rep_reg = data_.u_empty_match_check.repetition_register;
      bool has_minimum = (rep_reg != RegExpCompiler::kNoRegister);
      bool know_dist = trace->GetStoredPosition(start_pos_reg, &stored_pos);
      if (know_dist && !has_minimum && stored_pos == trace->cp_offset()) {
        // If we know we haven't advanced and there is no minimum we
        // can just backtrack immediately.
        assembler->GoTo(trace->backtrack());
      } else if (know_dist && stored_pos < trace->cp_offset()) {
        // If we know we've advanced we can generate the continuation
        // immediately.
        on_success()->Emit(compiler, trace);
      } else if (!trace->is_trivial()) {
        trace->Flush(compiler, this);
      } else {
        Label skip_empty_check;
        // If we have a minimum number of repetitions we check the current
        // number first and skip the empty check if it's not enough.
        if (has_minimum) {
          int limit = data_.u_empty_match_check.repetition_limit;
          assembler->IfRegisterLT(rep_reg, limit, &skip_empty_check);
        }
        // If the match is empty we bail out, otherwise we fall through
        // to the on-success continuation.
        assembler->IfRegisterEqPos(data_.u_empty_match_check.start_register,
                                   trace->backtrack());
        assembler->Bind(&skip_empty_check);
        on_success()->Emit(compiler, trace);
      }
      break;
    }
    case POSITIVE_SUBMATCH_SUCCESS: {
      if (!trace->is_trivial()) {
        trace->Flush(compiler, this);
        return;
      }
      assembler->ReadCurrentPositionFromRegister(
          data_.u_submatch.current_position_register);
      assembler->ReadStackPointerFromRegister(
          data_.u_submatch.stack_pointer_register);
      int clear_register_count = data_.u_submatch.clear_register_count;
      if (clear_register_count == 0) {
        on_success()->Emit(compiler, trace);
        return;
      }
      int clear_registers_from = data_.u_submatch.clear_register_from;
      Label clear_registers_backtrack;
      Trace new_trace = *trace;
      new_trace.set_backtrack(&clear_registers_backtrack);
      on_success()->Emit(compiler, &new_trace);

      assembler->Bind(&clear_registers_backtrack);
      int clear_registers_to = clear_registers_from + clear_register_count - 1;
      assembler->ClearRegisters(clear_registers_from, clear_registers_to);

      DCHECK(trace->backtrack() == nullptr);
      assembler->Backtrack();
      return;
    }
    default:
      UNREACHABLE();
  }
}

void BackReferenceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
  RegExpMacroAssembler* assembler = compiler->macro_assembler();
  if (!trace->is_trivial()) {
    trace->Flush(compiler, this);
    return;
  }

  LimitResult limit_result = LimitVersions(compiler, trace);
  if (limit_result == DONE) return;
  DCHECK(limit_result == CONTINUE);

  RecursionCheck rc(compiler);

  DCHECK_EQ(start_reg_ + 1, end_reg_);
  if (IgnoreCase(flags_)) {
    assembler->CheckNotBackReferenceIgnoreCase(
        start_reg_, read_backward(), IsUnicode(flags_), trace->backtrack());
  } else {
    assembler->CheckNotBackReference(start_reg_, read_backward(),
                                     trace->backtrack());
  }
  // We are going to advance backward, so we may end up at the start.
  if (read_backward()) trace->set_at_start(Trace::UNKNOWN);

  // Check that the back reference does not end inside a surrogate pair.
  if (IsUnicode(flags_) && !compiler->one_byte()) {
    assembler->CheckNotInSurrogatePair(trace->cp_offset(), trace->backtrack());
  }
  on_success()->Emit(compiler, trace);
}

void TextNode::CalculateOffsets() {
  int element_count = elements()->length();
  // Set up the offsets of the elements relative to the start.  This is a fixed
  // quantity since a TextNode can only contain fixed-width things.
  int cp_offset = 0;
  for (int i = 0; i < element_count; i++) {
    TextElement& elm = elements()->at(i);
    elm.set_cp_offset(cp_offset);
    cp_offset += elm.length();
  }
}

3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476
namespace {

// Assertion propagation moves information about assertions such as
// \b to the affected nodes.  For instance, in /.\b./ information must
// be propagated to the first '.' that whatever follows needs to know
// if it matched a word or a non-word, and to the second '.' that it
// has to check if it succeeds a word or non-word.  In this case the
// result will be something like:
//
//   +-------+        +------------+
//   |   .   |        |      .     |
//   +-------+  --->  +------------+
//   | word? |        | check word |
//   +-------+        +------------+
class AssertionPropagator : public AllStatic {
 public:
  static void VisitText(TextNode* that) {}
3477

3478
  static void VisitAction(ActionNode* that) {
3479 3480
    // If the next node is interested in what it follows then this node
    // has to be interested too so it can pass the information on.
3481
    that->info()->AddFromFollowing(that->on_success()->info());
3482 3483
  }

3484
  static void VisitChoice(ChoiceNode* that, int i) {
3485 3486
    // Anything the following nodes need to know has to be known by
    // this node also, so it can pass it on.
3487
    that->info()->AddFromFollowing(that->alternatives()->at(i).node()->info());
3488 3489
  }

3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532
  static void VisitLoopChoiceContinueNode(LoopChoiceNode* that) {
    that->info()->AddFromFollowing(that->continue_node()->info());
  }

  static void VisitLoopChoiceLoopNode(LoopChoiceNode* that) {
    that->info()->AddFromFollowing(that->loop_node()->info());
  }

  static void VisitNegativeLookaroundChoiceLookaroundNode(
      NegativeLookaroundChoiceNode* that) {
    VisitChoice(that, NegativeLookaroundChoiceNode::kLookaroundIndex);
  }

  static void VisitNegativeLookaroundChoiceContinueNode(
      NegativeLookaroundChoiceNode* that) {
    VisitChoice(that, NegativeLookaroundChoiceNode::kContinueIndex);
  }

  static void VisitBackReference(BackReferenceNode* that) {}

  static void VisitAssertion(AssertionNode* that) {}
};

// Propagates information about the minimum size of successful matches from
// successor nodes to their predecessors. Note that all eats_at_least values
// are initialized to zero before analysis.
class EatsAtLeastPropagator : public AllStatic {
 public:
  static void VisitText(TextNode* that) {
    // The eats_at_least value is not used if reading backward.
    if (!that->read_backward()) {
      // We are not at the start after this node, and thus we can use the
      // successor's eats_at_least_from_not_start value.
      uint8_t eats_at_least = base::saturated_cast<uint8_t>(
          that->Length() + that->on_success()
                               ->eats_at_least_info()
                               ->eats_at_least_from_not_start);
      that->set_eats_at_least_info(EatsAtLeastInfo(eats_at_least));
    }
  }

  static void VisitAction(ActionNode* that) {
    // POSITIVE_SUBMATCH_SUCCESS rewinds input, so we must not consider
3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546
    // successor nodes for eats_at_least. SET_REGISTER_FOR_LOOP indicates a loop
    // entry point, which means the loop body will run at least the minimum
    // number of times before the continuation case can run. Otherwise the
    // current node eats at least as much as its successor.
    switch (that->action_type()) {
      case ActionNode::POSITIVE_SUBMATCH_SUCCESS:
        break;  // Was already initialized to zero.
      case ActionNode::SET_REGISTER_FOR_LOOP:
        that->set_eats_at_least_info(
            that->on_success()->EatsAtLeastFromLoopEntry());
        break;
      default:
        that->set_eats_at_least_info(*that->on_success()->eats_at_least_info());
        break;
3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657
    }
  }

  static void VisitChoice(ChoiceNode* that, int i) {
    // The minimum possible match from a choice node is the minimum of its
    // successors.
    EatsAtLeastInfo eats_at_least =
        i == 0 ? EatsAtLeastInfo(UINT8_MAX) : *that->eats_at_least_info();
    eats_at_least.SetMin(
        *that->alternatives()->at(i).node()->eats_at_least_info());
    that->set_eats_at_least_info(eats_at_least);
  }

  static void VisitLoopChoiceContinueNode(LoopChoiceNode* that) {
    that->set_eats_at_least_info(*that->continue_node()->eats_at_least_info());
  }

  static void VisitLoopChoiceLoopNode(LoopChoiceNode* that) {}

  static void VisitNegativeLookaroundChoiceLookaroundNode(
      NegativeLookaroundChoiceNode* that) {}

  static void VisitNegativeLookaroundChoiceContinueNode(
      NegativeLookaroundChoiceNode* that) {
    that->set_eats_at_least_info(*that->continue_node()->eats_at_least_info());
  }

  static void VisitBackReference(BackReferenceNode* that) {
    if (!that->read_backward()) {
      that->set_eats_at_least_info(*that->on_success()->eats_at_least_info());
    }
  }

  static void VisitAssertion(AssertionNode* that) {
    EatsAtLeastInfo eats_at_least = *that->on_success()->eats_at_least_info();
    if (that->assertion_type() == AssertionNode::AT_START) {
      // If we know we are not at the start and we are asked "how many
      // characters will you match if you succeed?" then we can answer anything
      // since false implies false.  So let's just set the max answer
      // (UINT8_MAX) since that won't prevent us from preloading a lot of
      // characters for the other branches in the node graph.
      eats_at_least.eats_at_least_from_not_start = UINT8_MAX;
    }
    that->set_eats_at_least_info(eats_at_least);
  }
};

}  // namespace

// -------------------------------------------------------------------
// Analysis

// Iterates the node graph and provides the opportunity for propagators to set
// values that depend on successor nodes.
template <typename... Propagators>
class Analysis : public NodeVisitor {
 public:
  Analysis(Isolate* isolate, bool is_one_byte)
      : isolate_(isolate), is_one_byte_(is_one_byte), error_message_(nullptr) {}

  void EnsureAnalyzed(RegExpNode* that) {
    StackLimitCheck check(isolate());
    if (check.HasOverflowed()) {
      fail("Stack overflow");
      return;
    }
    if (that->info()->been_analyzed || that->info()->being_analyzed) return;
    that->info()->being_analyzed = true;
    that->Accept(this);
    that->info()->being_analyzed = false;
    that->info()->been_analyzed = true;
  }

  bool has_failed() { return error_message_ != nullptr; }
  const char* error_message() {
    DCHECK(error_message_ != nullptr);
    return error_message_;
  }
  void fail(const char* error_message) { error_message_ = error_message; }

  Isolate* isolate() const { return isolate_; }

  void VisitEnd(EndNode* that) override {
    // nothing to do
  }

// Used to call the given static function on each propagator / variadic template
// argument.
#define STATIC_FOR_EACH(expr)       \
  do {                              \
    int dummy[] = {((expr), 0)...}; \
    USE(dummy);                     \
  } while (false)

  void VisitText(TextNode* that) override {
    that->MakeCaseIndependent(isolate(), is_one_byte_);
    EnsureAnalyzed(that->on_success());
    if (has_failed()) return;
    that->CalculateOffsets();
    STATIC_FOR_EACH(Propagators::VisitText(that));
  }

  void VisitAction(ActionNode* that) override {
    EnsureAnalyzed(that->on_success());
    if (has_failed()) return;
    STATIC_FOR_EACH(Propagators::VisitAction(that));
  }

  void VisitChoice(ChoiceNode* that) override {
    for (int i = 0; i < that->alternatives()->length(); i++) {
      EnsureAnalyzed(that->alternatives()->at(i).node());
3658
      if (has_failed()) return;
3659
      STATIC_FOR_EACH(Propagators::VisitChoice(that, i));
3660 3661
    }
  }
3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675

  void VisitLoopChoice(LoopChoiceNode* that) override {
    DCHECK_EQ(that->alternatives()->length(), 2);  // Just loop and continue.

    // First propagate all information from the continuation node.
    EnsureAnalyzed(that->continue_node());
    if (has_failed()) return;
    STATIC_FOR_EACH(Propagators::VisitLoopChoiceContinueNode(that));

    // Check the loop last since it may need the value of this node
    // to get a correct result.
    EnsureAnalyzed(that->loop_node());
    if (has_failed()) return;
    STATIC_FOR_EACH(Propagators::VisitLoopChoiceLoopNode(that));
3676 3677
  }

3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713
  void VisitNegativeLookaroundChoice(
      NegativeLookaroundChoiceNode* that) override {
    DCHECK_EQ(that->alternatives()->length(), 2);  // Lookaround and continue.

    EnsureAnalyzed(that->lookaround_node());
    if (has_failed()) return;
    STATIC_FOR_EACH(
        Propagators::VisitNegativeLookaroundChoiceLookaroundNode(that));

    EnsureAnalyzed(that->continue_node());
    if (has_failed()) return;
    STATIC_FOR_EACH(
        Propagators::VisitNegativeLookaroundChoiceContinueNode(that));
  }

  void VisitBackReference(BackReferenceNode* that) override {
    EnsureAnalyzed(that->on_success());
    if (has_failed()) return;
    STATIC_FOR_EACH(Propagators::VisitBackReference(that));
  }

  void VisitAssertion(AssertionNode* that) override {
    EnsureAnalyzed(that->on_success());
    if (has_failed()) return;
    STATIC_FOR_EACH(Propagators::VisitAssertion(that));
  }

#undef STATIC_FOR_EACH

 private:
  Isolate* isolate_;
  bool is_one_byte_;
  const char* error_message_;

  DISALLOW_IMPLICIT_CONSTRUCTORS(Analysis);
};
3714

3715 3716 3717 3718 3719 3720 3721 3722
const char* AnalyzeRegExp(Isolate* isolate, bool is_one_byte,
                          RegExpNode* node) {
  Analysis<AssertionPropagator, EatsAtLeastPropagator> analysis(isolate,
                                                                is_one_byte);
  DCHECK_EQ(node->info()->been_analyzed, false);
  analysis.EnsureAnalyzed(node);
  DCHECK_IMPLIES(analysis.has_failed(), analysis.error_message() != nullptr);
  return analysis.has_failed() ? analysis.error_message() : nullptr;
3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839
}

void BackReferenceNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
                                     BoyerMooreLookahead* bm,
                                     bool not_at_start) {
  // Working out the set of characters that a backreference can match is too
  // hard, so we just say that any character can match.
  bm->SetRest(offset);
  SaveBMInfo(bm, not_at_start, offset);
}

STATIC_ASSERT(BoyerMoorePositionInfo::kMapSize ==
              RegExpMacroAssembler::kTableSize);

void ChoiceNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
                              BoyerMooreLookahead* bm, bool not_at_start) {
  ZoneList<GuardedAlternative>* alts = alternatives();
  budget = (budget - 1) / alts->length();
  for (int i = 0; i < alts->length(); i++) {
    GuardedAlternative& alt = alts->at(i);
    if (alt.guards() != nullptr && alt.guards()->length() != 0) {
      bm->SetRest(offset);  // Give up trying to fill in info.
      SaveBMInfo(bm, not_at_start, offset);
      return;
    }
    alt.node()->FillInBMInfo(isolate, offset, budget, bm, not_at_start);
  }
  SaveBMInfo(bm, not_at_start, offset);
}

void TextNode::FillInBMInfo(Isolate* isolate, int initial_offset, int budget,
                            BoyerMooreLookahead* bm, bool not_at_start) {
  if (initial_offset >= bm->length()) return;
  int offset = initial_offset;
  int max_char = bm->max_char();
  for (int i = 0; i < elements()->length(); i++) {
    if (offset >= bm->length()) {
      if (initial_offset == 0) set_bm_info(not_at_start, bm);
      return;
    }
    TextElement text = elements()->at(i);
    if (text.text_type() == TextElement::ATOM) {
      RegExpAtom* atom = text.atom();
      for (int j = 0; j < atom->length(); j++, offset++) {
        if (offset >= bm->length()) {
          if (initial_offset == 0) set_bm_info(not_at_start, bm);
          return;
        }
        uc16 character = atom->data()[j];
        if (IgnoreCase(atom->flags())) {
          unibrow::uchar chars[4];
          int length = GetCaseIndependentLetters(
              isolate, character, bm->max_char() == String::kMaxOneByteCharCode,
              chars, 4);
          for (int j = 0; j < length; j++) {
            bm->Set(offset, chars[j]);
          }
        } else {
          if (character <= max_char) bm->Set(offset, character);
        }
      }
    } else {
      DCHECK_EQ(TextElement::CHAR_CLASS, text.text_type());
      RegExpCharacterClass* char_class = text.char_class();
      ZoneList<CharacterRange>* ranges = char_class->ranges(zone());
      if (char_class->is_negated()) {
        bm->SetAll(offset);
      } else {
        for (int k = 0; k < ranges->length(); k++) {
          CharacterRange& range = ranges->at(k);
          if (range.from() > max_char) continue;
          int to = Min(max_char, static_cast<int>(range.to()));
          bm->SetInterval(offset, Interval(range.from(), to));
        }
      }
      offset++;
    }
  }
  if (offset >= bm->length()) {
    if (initial_offset == 0) set_bm_info(not_at_start, bm);
    return;
  }
  on_success()->FillInBMInfo(isolate, offset, budget - 1, bm,
                             true);  // Not at start after a text node.
  if (initial_offset == 0) set_bm_info(not_at_start, bm);
}

// static
RegExpNode* RegExpCompiler::OptionallyStepBackToLeadSurrogate(
    RegExpCompiler* compiler, RegExpNode* on_success, JSRegExp::Flags flags) {
  DCHECK(!compiler->read_backward());
  Zone* zone = compiler->zone();
  ZoneList<CharacterRange>* lead_surrogates = CharacterRange::List(
      zone, CharacterRange::Range(kLeadSurrogateStart, kLeadSurrogateEnd));
  ZoneList<CharacterRange>* trail_surrogates = CharacterRange::List(
      zone, CharacterRange::Range(kTrailSurrogateStart, kTrailSurrogateEnd));

  ChoiceNode* optional_step_back = new (zone) ChoiceNode(2, zone);

  int stack_register = compiler->UnicodeLookaroundStackRegister();
  int position_register = compiler->UnicodeLookaroundPositionRegister();
  RegExpNode* step_back = TextNode::CreateForCharacterRanges(
      zone, lead_surrogates, true, on_success, flags);
  RegExpLookaround::Builder builder(true, step_back, stack_register,
                                    position_register);
  RegExpNode* match_trail = TextNode::CreateForCharacterRanges(
      zone, trail_surrogates, false, builder.on_match_success(), flags);

  optional_step_back->AddAlternative(
      GuardedAlternative(builder.ForMatch(match_trail)));
  optional_step_back->AddAlternative(GuardedAlternative(on_success));

  return optional_step_back;
}

}  // namespace internal
}  // namespace v8