From 5f8de423f190bbb79a62f804151bc24824fa32d8 Mon Sep 17 00:00:00 2001 From: "Matt A. Tobin" Date: Fri, 2 Feb 2018 04:16:08 -0500 Subject: Add m-esr52 at 52.6.0 --- js/src/irregexp/NativeRegExpMacroAssembler.cpp | 1427 +++++++ js/src/irregexp/NativeRegExpMacroAssembler.h | 226 ++ js/src/irregexp/RegExpAST.cpp | 265 ++ js/src/irregexp/RegExpAST.h | 449 +++ js/src/irregexp/RegExpBytecode.h | 108 + js/src/irregexp/RegExpEngine.cpp | 5135 ++++++++++++++++++++++++ js/src/irregexp/RegExpEngine.h | 1546 +++++++ js/src/irregexp/RegExpInterpreter.cpp | 501 +++ js/src/irregexp/RegExpMacroAssembler.cpp | 573 +++ js/src/irregexp/RegExpMacroAssembler.h | 312 ++ js/src/irregexp/RegExpParser.cpp | 1905 +++++++++ js/src/irregexp/RegExpParser.h | 310 ++ js/src/irregexp/RegExpStack.cpp | 106 + js/src/irregexp/RegExpStack.h | 122 + 14 files changed, 12985 insertions(+) create mode 100644 js/src/irregexp/NativeRegExpMacroAssembler.cpp create mode 100644 js/src/irregexp/NativeRegExpMacroAssembler.h create mode 100644 js/src/irregexp/RegExpAST.cpp create mode 100644 js/src/irregexp/RegExpAST.h create mode 100644 js/src/irregexp/RegExpBytecode.h create mode 100644 js/src/irregexp/RegExpEngine.cpp create mode 100644 js/src/irregexp/RegExpEngine.h create mode 100644 js/src/irregexp/RegExpInterpreter.cpp create mode 100644 js/src/irregexp/RegExpMacroAssembler.cpp create mode 100644 js/src/irregexp/RegExpMacroAssembler.h create mode 100644 js/src/irregexp/RegExpParser.cpp create mode 100644 js/src/irregexp/RegExpParser.h create mode 100644 js/src/irregexp/RegExpStack.cpp create mode 100644 js/src/irregexp/RegExpStack.h (limited to 'js/src/irregexp') diff --git a/js/src/irregexp/NativeRegExpMacroAssembler.cpp b/js/src/irregexp/NativeRegExpMacroAssembler.cpp new file mode 100644 index 000000000..0fb507297 --- /dev/null +++ b/js/src/irregexp/NativeRegExpMacroAssembler.cpp @@ -0,0 +1,1427 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- + * vim: set ts=8 sts=4 et sw=4 tw=99: */ + +// Copyright 2012 the V8 project authors. All rights reserved. +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived +// from this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include "irregexp/NativeRegExpMacroAssembler.h" + +#include "irregexp/RegExpStack.h" +#include "jit/Linker.h" +#ifdef JS_ION_PERF +# include "jit/PerfSpewer.h" +#endif +#include "vm/MatchPairs.h" + +#include "jit/MacroAssembler-inl.h" + +using namespace js; +using namespace js::irregexp; +using namespace js::jit; + +/* + * This assembler uses the following register assignment convention: + * + * - current_character : + * Must be loaded using LoadCurrentCharacter before using any of the + * dispatch methods. Temporarily stores the index of capture start after a + * matching pass for a global regexp. + * - current_position : + * Current position in input, as negative offset from end of string. + * Please notice that this is the byte offset, not the character offset! + * - input_end_pointer : + * Points to byte after last character in the input. + * - backtrack_stack_pointer : + * Points to tip of the heap allocated backtrack stack + * - StackPointer : + * Points to tip of the native stack, used to access arguments, local + * variables and RegExp registers. + * + * The tempN registers are free to use for computations. + */ + +NativeRegExpMacroAssembler::NativeRegExpMacroAssembler(LifoAlloc* alloc, RegExpShared* shared, + JSRuntime* rt, Mode mode, int registers_to_save) + : RegExpMacroAssembler(*alloc, shared, registers_to_save), + runtime(rt), mode_(mode) +{ + // Find physical registers for each compiler register. + AllocatableGeneralRegisterSet regs(GeneralRegisterSet::All()); + + input_end_pointer = regs.takeAny(); + current_character = regs.takeAny(); + current_position = regs.takeAny(); + backtrack_stack_pointer = regs.takeAny(); + temp0 = regs.takeAny(); + temp1 = regs.takeAny(); + temp2 = regs.takeAny(); + + JitSpew(JitSpew_Codegen, + "Starting RegExp (input_end_pointer %s) (current_character %s)" + " (current_position %s) (backtrack_stack_pointer %s) (temp0 %s) temp1 (%s) temp2 (%s)", + input_end_pointer.name(), + current_character.name(), + current_position.name(), + backtrack_stack_pointer.name(), + temp0.name(), + temp1.name(), + temp2.name()); + + savedNonVolatileRegisters = SavedNonVolatileRegisters(regs); + + masm.jump(&entry_label_); + masm.bind(&start_label_); +} + +#define SPEW_PREFIX JitSpew_Codegen, "!!! " + +// The signature of the code which this generates is: +// +// void execute(InputOutputData*); +RegExpCode +NativeRegExpMacroAssembler::GenerateCode(JSContext* cx, bool match_only) +{ + if (!cx->compartment()->ensureJitCompartmentExists(cx)) + return RegExpCode(); + + JitSpew(SPEW_PREFIX "GenerateCode"); + + // We need an even number of registers, for stack alignment. + if (num_registers_ % 2 == 1) + num_registers_++; + + Label return_temp0; + + // Finalize code - write the entry point code now we know how many + // registers we need. + masm.bind(&entry_label_); + +#ifdef JS_CODEGEN_ARM64 + // ARM64 communicates stack address via sp, but uses a pseudo-sp for addressing. + masm.initStackPtr(); +#endif + + // Push non-volatile registers which might be modified by jitcode. + size_t pushedNonVolatileRegisters = 0; + for (GeneralRegisterForwardIterator iter(savedNonVolatileRegisters); iter.more(); ++iter) { + masm.Push(*iter); + pushedNonVolatileRegisters++; + } + +#if defined(XP_IOS) && defined(JS_CODEGEN_ARM) + // The stack is 4-byte aligned on iOS, force 8-byte alignment. + masm.movePtr(StackPointer, temp0); + masm.andPtr(Imm32(~7), StackPointer); + masm.push(temp0); + masm.push(temp0); +#endif + +#ifndef JS_CODEGEN_X86 + // The InputOutputData* is stored as an argument, save it on the stack + // above the frame. + masm.Push(IntArgReg0); +#endif + + size_t frameSize = sizeof(FrameData) + num_registers_ * sizeof(void*); + frameSize = JS_ROUNDUP(frameSize + masm.framePushed(), ABIStackAlignment) - masm.framePushed(); + + // Actually emit code to start a new stack frame. + masm.reserveStack(frameSize); + masm.checkStackAlignment(); + + // Check if we have space on the stack. Use the *NoInterrupt stack limit to + // avoid failing repeatedly when the regex code is called from Ion JIT code, + // see bug 1208819. + Label stack_ok; + void* stack_limit = runtime->addressOfJitStackLimitNoInterrupt(); + masm.branchStackPtrRhs(Assembler::Below, AbsoluteAddress(stack_limit), &stack_ok); + + // Exit with an exception. There is not enough space on the stack + // for our working registers. + masm.movePtr(ImmWord(RegExpRunStatus_Error), temp0); + masm.jump(&return_temp0); + + masm.bind(&stack_ok); + +#ifdef XP_WIN + // Ensure that we write to each stack page, in order. Skipping a page + // on Windows can cause segmentation faults. Assuming page size is 4k. + const int kPageSize = 4096; + for (int i = frameSize - sizeof(void*); i >= 0; i -= kPageSize) + masm.storePtr(temp0, Address(masm.getStackPointer(), i)); +#endif // XP_WIN + +#ifndef JS_CODEGEN_X86 + // The InputOutputData* is stored on the stack immediately above the frame. + Address inputOutputAddress(masm.getStackPointer(), frameSize); +#else + // The InputOutputData* is left in its original on stack location. + Address inputOutputAddress(masm.getStackPointer(), + frameSize + (pushedNonVolatileRegisters + 1) * sizeof(void*)); +#endif + + masm.loadPtr(inputOutputAddress, temp0); + + // Copy output registers to FrameData. + if (!match_only) { + Register matchPairsRegister = input_end_pointer; + masm.loadPtr(Address(temp0, offsetof(InputOutputData, matches)), matchPairsRegister); + masm.loadPtr(Address(matchPairsRegister, MatchPairs::offsetOfPairs()), temp1); + masm.storePtr(temp1, Address(masm.getStackPointer(), offsetof(FrameData, outputRegisters))); + masm.load32(Address(matchPairsRegister, MatchPairs::offsetOfPairCount()), temp1); + masm.lshiftPtr(Imm32(1), temp1); + masm.store32(temp1, Address(masm.getStackPointer(), offsetof(FrameData, numOutputRegisters))); + +#ifdef DEBUG + // Bounds check numOutputRegisters. + Label enoughRegisters; + masm.branchPtr(Assembler::GreaterThanOrEqual, + temp1, ImmWord(num_saved_registers_), &enoughRegisters); + masm.assumeUnreachable("Not enough output registers for RegExp"); + masm.bind(&enoughRegisters); +#endif + } else { + Register endIndexRegister = input_end_pointer; + masm.loadPtr(Address(temp0, offsetof(InputOutputData, endIndex)), endIndexRegister); + masm.storePtr(endIndexRegister, Address(masm.getStackPointer(), offsetof(FrameData, endIndex))); + } + + // Load string end pointer. + masm.loadPtr(Address(temp0, offsetof(InputOutputData, inputEnd)), input_end_pointer); + + // Load input start pointer, and copy to FrameData. + masm.loadPtr(Address(temp0, offsetof(InputOutputData, inputStart)), current_position); + masm.storePtr(current_position, Address(masm.getStackPointer(), offsetof(FrameData, inputStart))); + + // Load start index, and copy to FrameData. + masm.loadPtr(Address(temp0, offsetof(InputOutputData, startIndex)), temp1); + masm.storePtr(temp1, Address(masm.getStackPointer(), offsetof(FrameData, startIndex))); + + // Set up input position to be negative offset from string end. + masm.subPtr(input_end_pointer, current_position); + + // Set temp0 to address of char before start of the string. + // (effectively string position -1). + masm.computeEffectiveAddress(Address(current_position, -char_size()), temp0); + + // Store this value on the frame, for use when clearing + // position registers. + masm.storePtr(temp0, Address(masm.getStackPointer(), offsetof(FrameData, inputStartMinusOne))); + + // Update current position based on start index. + masm.computeEffectiveAddress(BaseIndex(current_position, temp1, factor()), current_position); + + Label load_char_start_regexp, start_regexp; + + // Load newline if index is at start, previous character otherwise. + masm.branchPtr(Assembler::NotEqual, + Address(masm.getStackPointer(), offsetof(FrameData, startIndex)), ImmWord(0), + &load_char_start_regexp); + masm.movePtr(ImmWord('\n'), current_character); + masm.jump(&start_regexp); + + // Global regexp restarts matching here. + masm.bind(&load_char_start_regexp); + + // Load previous char as initial value of current character register. + LoadCurrentCharacterUnchecked(-1, 1); + masm.bind(&start_regexp); + + // Initialize on-stack registers. + MOZ_ASSERT(num_saved_registers_ > 0); + + // Fill saved registers with initial value = start offset - 1 + // Fill in stack push order, to avoid accessing across an unwritten + // page (a problem on Windows). + if (num_saved_registers_ > 8) { + masm.movePtr(ImmWord(register_offset(0)), temp1); + Label init_loop; + masm.bind(&init_loop); + masm.storePtr(temp0, BaseIndex(masm.getStackPointer(), temp1, TimesOne)); + masm.addPtr(ImmWord(sizeof(void*)), temp1); + masm.branchPtr(Assembler::LessThan, temp1, + ImmWord(register_offset(num_saved_registers_)), &init_loop); + } else { + // Unroll the loop. + for (int i = 0; i < num_saved_registers_; i++) + masm.storePtr(temp0, register_location(i)); + } + + // Initialize backtrack stack pointer. + masm.loadPtr(AbsoluteAddress(runtime->regexpStack.addressOfBase()), backtrack_stack_pointer); + masm.storePtr(backtrack_stack_pointer, + Address(masm.getStackPointer(), offsetof(FrameData, backtrackStackBase))); + + masm.jump(&start_label_); + + // Exit code: + if (success_label_.used()) { + MOZ_ASSERT(num_saved_registers_ > 0); + + Address outputRegistersAddress(masm.getStackPointer(), offsetof(FrameData, outputRegisters)); + + // Save captures when successful. + masm.bind(&success_label_); + + if (!match_only) { + Register outputRegisters = temp1; + Register inputByteLength = backtrack_stack_pointer; + + masm.loadPtr(outputRegistersAddress, outputRegisters); + + masm.loadPtr(inputOutputAddress, temp0); + masm.loadPtr(Address(temp0, offsetof(InputOutputData, inputEnd)), inputByteLength); + masm.subPtr(Address(temp0, offsetof(InputOutputData, inputStart)), inputByteLength); + + // Copy captures to output. Note that registers on the C stack are pointer width + // so that they might hold pointers, but output registers are int32_t. + for (int i = 0; i < num_saved_registers_; i++) { + masm.loadPtr(register_location(i), temp0); + if (i == 0 && global_with_zero_length_check()) { + // Keep capture start in current_character for the zero-length check later. + masm.movePtr(temp0, current_character); + } + + // Convert to index from start of string, not end. + masm.addPtr(inputByteLength, temp0); + + // Convert byte index to character index. + if (mode_ == CHAR16) + masm.rshiftPtrArithmetic(Imm32(1), temp0); + + masm.store32(temp0, Address(outputRegisters, i * sizeof(int32_t))); + } + } + + // Restart matching if the regular expression is flagged as global. + if (global()) { + // Increment success counter. + masm.add32(Imm32(1), Address(masm.getStackPointer(), offsetof(FrameData, successfulCaptures))); + + Address numOutputRegistersAddress(masm.getStackPointer(), offsetof(FrameData, numOutputRegisters)); + + // Capture results have been stored, so the number of remaining global + // output registers is reduced by the number of stored captures. + masm.load32(numOutputRegistersAddress, temp0); + + masm.sub32(Imm32(num_saved_registers_), temp0); + + // Check whether we have enough room for another set of capture results. + masm.branch32(Assembler::LessThan, temp0, Imm32(num_saved_registers_), &exit_label_); + + masm.store32(temp0, numOutputRegistersAddress); + + // Advance the location for output. + masm.add32(Imm32(num_saved_registers_ * sizeof(void*)), outputRegistersAddress); + + // Prepare temp0 to initialize registers with its value in the next run. + masm.loadPtr(Address(masm.getStackPointer(), offsetof(FrameData, inputStartMinusOne)), temp0); + + if (global_with_zero_length_check()) { + // Special case for zero-length matches. + + // The capture start index was loaded into current_character above. + masm.branchPtr(Assembler::NotEqual, current_position, current_character, + &load_char_start_regexp); + + // edi (offset from the end) is zero if we already reached the end. + masm.branchTestPtr(Assembler::Zero, current_position, current_position, + &exit_label_); + + // Advance current position after a zero-length match. + masm.addPtr(Imm32(char_size()), current_position); + } + + masm.jump(&load_char_start_regexp); + } else { + if (match_only) { + // Store endIndex. + + Register endIndexRegister = temp1; + Register inputByteLength = backtrack_stack_pointer; + + masm.loadPtr(Address(masm.getStackPointer(), offsetof(FrameData, endIndex)), endIndexRegister); + + masm.loadPtr(inputOutputAddress, temp0); + masm.loadPtr(Address(temp0, offsetof(InputOutputData, inputEnd)), inputByteLength); + masm.subPtr(Address(temp0, offsetof(InputOutputData, inputStart)), inputByteLength); + + masm.loadPtr(register_location(1), temp0); + + // Convert to index from start of string, not end. + masm.addPtr(inputByteLength, temp0); + + // Convert byte index to character index. + if (mode_ == CHAR16) + masm.rshiftPtrArithmetic(Imm32(1), temp0); + + masm.store32(temp0, Address(endIndexRegister, 0)); + } + + masm.movePtr(ImmWord(RegExpRunStatus_Success), temp0); + } + } + + masm.bind(&exit_label_); + + if (global()) { + // Return the number of successful captures. + masm.load32(Address(masm.getStackPointer(), offsetof(FrameData, successfulCaptures)), temp0); + } + + masm.bind(&return_temp0); + + // Store the result to the input structure. + masm.loadPtr(inputOutputAddress, temp1); + masm.storePtr(temp0, Address(temp1, offsetof(InputOutputData, result))); + +#ifndef JS_CODEGEN_X86 + // Include the InputOutputData* when adjusting the stack size. + masm.freeStack(frameSize + sizeof(void*)); +#else + masm.freeStack(frameSize); +#endif + +#if defined(XP_IOS) && defined(JS_CODEGEN_ARM) + masm.pop(temp0); + masm.movePtr(temp0, StackPointer); +#endif + + // Restore non-volatile registers which were saved on entry. + for (GeneralRegisterBackwardIterator iter(savedNonVolatileRegisters); iter.more(); ++iter) + masm.Pop(*iter); + + masm.abiret(); + + // Backtrack code (branch target for conditional backtracks). + if (backtrack_label_.used()) { + masm.bind(&backtrack_label_); + Backtrack(); + } + + // Backtrack stack overflow code. + if (stack_overflow_label_.used()) { + // Reached if the backtrack-stack limit has been hit. temp2 holds the + // StackPointer to use for accessing FrameData. + masm.bind(&stack_overflow_label_); + + Label grow_failed; + + masm.movePtr(ImmPtr(runtime), temp1); + + // Save registers before calling C function + LiveGeneralRegisterSet volatileRegs(GeneralRegisterSet::Volatile()); +#if defined(JS_CODEGEN_ARM) || defined(JS_CODEGEN_ARM64) + volatileRegs.add(Register::FromCode(Registers::lr)); +#elif defined(JS_CODEGEN_MIPS32) || defined(JS_CODEGEN_MIPS64) + volatileRegs.add(Register::FromCode(Registers::ra)); +#endif + volatileRegs.takeUnchecked(temp0); + volatileRegs.takeUnchecked(temp1); + masm.PushRegsInMask(volatileRegs); + + masm.setupUnalignedABICall(temp0); + masm.passABIArg(temp1); + masm.callWithABI(JS_FUNC_TO_DATA_PTR(void*, GrowBacktrackStack)); + masm.storeCallBoolResult(temp0); + + masm.PopRegsInMask(volatileRegs); + + // If return false, we have failed to grow the stack, and + // must exit with a stack-overflow exception. Do this in the caller + // so that the stack is adjusted by our return instruction. + Label return_from_overflow_handler; + masm.branchTest32(Assembler::Zero, temp0, temp0, &return_from_overflow_handler); + + // Otherwise, store the new backtrack stack base and recompute the new + // top of the stack. + Address backtrackStackBaseAddress(temp2, offsetof(FrameData, backtrackStackBase)); + masm.subPtr(backtrackStackBaseAddress, backtrack_stack_pointer); + + masm.loadPtr(AbsoluteAddress(runtime->regexpStack.addressOfBase()), temp1); + masm.storePtr(temp1, backtrackStackBaseAddress); + masm.addPtr(temp1, backtrack_stack_pointer); + + // Resume execution in calling code. + masm.bind(&return_from_overflow_handler); + masm.abiret(); + } + + if (exit_with_exception_label_.used()) { + // If any of the code above needed to exit with an exception. + masm.bind(&exit_with_exception_label_); + + // Exit with an error result to signal thrown exception. + masm.movePtr(ImmWord(RegExpRunStatus_Error), temp0); + masm.jump(&return_temp0); + } + + Linker linker(masm); + AutoFlushICache afc("RegExp"); + JitCode* code = linker.newCode(cx, REGEXP_CODE); + if (!code) { + ReportOutOfMemory(cx); + return RegExpCode(); + } + +#ifdef JS_ION_PERF + writePerfSpewerJitCodeProfile(code, "RegExp"); +#endif + + for (size_t i = 0; i < labelPatches.length(); i++) { + LabelPatch& v = labelPatches[i]; + MOZ_ASSERT(!v.label); + Assembler::PatchDataWithValueCheck(CodeLocationLabel(code, v.patchOffset), + ImmPtr(code->raw() + v.labelOffset), + ImmPtr(0)); + } + + JitSpew(JitSpew_Codegen, "Created RegExp (raw %p length %d)", + (void*) code->raw(), (int) masm.bytesNeeded()); + + RegExpCode res; + res.jitCode = code; + return res; +} + +int +NativeRegExpMacroAssembler::stack_limit_slack() +{ + return RegExpStack::kStackLimitSlack; +} + +void +NativeRegExpMacroAssembler::AdvanceCurrentPosition(int by) +{ + JitSpew(SPEW_PREFIX "AdvanceCurrentPosition(%d)", by); + + if (by != 0) + masm.addPtr(Imm32(by * char_size()), current_position); +} + +void +NativeRegExpMacroAssembler::AdvanceRegister(int reg, int by) +{ + JitSpew(SPEW_PREFIX "AdvanceRegister(%d, %d)", reg, by); + + MOZ_ASSERT(reg >= 0); + MOZ_ASSERT(reg < num_registers_); + if (by != 0) + masm.addPtr(Imm32(by), register_location(reg)); +} + +void +NativeRegExpMacroAssembler::Backtrack() +{ + JitSpew(SPEW_PREFIX "Backtrack"); + + // Check for an interrupt. + Label noInterrupt; + masm.branch32(Assembler::Equal, + AbsoluteAddress(runtime->addressOfInterruptUint32()), Imm32(0), + &noInterrupt); + masm.movePtr(ImmWord(RegExpRunStatus_Error), temp0); + masm.jump(&exit_label_); + masm.bind(&noInterrupt); + + // Pop code location from backtrack stack and jump to location. + PopBacktrack(temp0); + masm.jump(temp0); +} + +void +NativeRegExpMacroAssembler::Bind(Label* label) +{ + JitSpew(SPEW_PREFIX "Bind"); + + masm.bind(label); +} + +void +NativeRegExpMacroAssembler::CheckAtStart(Label* on_at_start) +{ + JitSpew(SPEW_PREFIX "CheckAtStart"); + + Label not_at_start; + + // Did we start the match at the start of the string at all? + Address startIndex(masm.getStackPointer(), offsetof(FrameData, startIndex)); + masm.branchPtr(Assembler::NotEqual, startIndex, ImmWord(0), ¬_at_start); + + // If we did, are we still at the start of the input? + masm.computeEffectiveAddress(BaseIndex(input_end_pointer, current_position, TimesOne), temp0); + + Address inputStart(masm.getStackPointer(), offsetof(FrameData, inputStart)); + masm.branchPtr(Assembler::Equal, inputStart, temp0, BranchOrBacktrack(on_at_start)); + + masm.bind(¬_at_start); +} + +void +NativeRegExpMacroAssembler::CheckNotAtStart(Label* on_not_at_start) +{ + JitSpew(SPEW_PREFIX "CheckNotAtStart"); + + // Did we start the match at the start of the string at all? + Address startIndex(masm.getStackPointer(), offsetof(FrameData, startIndex)); + masm.branchPtr(Assembler::NotEqual, startIndex, ImmWord(0), BranchOrBacktrack(on_not_at_start)); + + // If we did, are we still at the start of the input? + masm.computeEffectiveAddress(BaseIndex(input_end_pointer, current_position, TimesOne), temp0); + + Address inputStart(masm.getStackPointer(), offsetof(FrameData, inputStart)); + masm.branchPtr(Assembler::NotEqual, inputStart, temp0, BranchOrBacktrack(on_not_at_start)); +} + +void +NativeRegExpMacroAssembler::CheckCharacter(unsigned c, Label* on_equal) +{ + JitSpew(SPEW_PREFIX "CheckCharacter(%d)", (int) c); + masm.branch32(Assembler::Equal, current_character, Imm32(c), BranchOrBacktrack(on_equal)); +} + +void +NativeRegExpMacroAssembler::CheckNotCharacter(unsigned c, Label* on_not_equal) +{ + JitSpew(SPEW_PREFIX "CheckNotCharacter(%d)", (int) c); + masm.branch32(Assembler::NotEqual, current_character, Imm32(c), BranchOrBacktrack(on_not_equal)); +} + +void +NativeRegExpMacroAssembler::CheckCharacterAfterAnd(unsigned c, unsigned and_with, + Label* on_equal) +{ + JitSpew(SPEW_PREFIX "CheckCharacterAfterAnd(%d, %d)", (int) c, (int) and_with); + + if (c == 0) { + masm.branchTest32(Assembler::Zero, current_character, Imm32(and_with), + BranchOrBacktrack(on_equal)); + } else { + masm.move32(Imm32(and_with), temp0); + masm.and32(current_character, temp0); + masm.branch32(Assembler::Equal, temp0, Imm32(c), BranchOrBacktrack(on_equal)); + } +} + +void +NativeRegExpMacroAssembler::CheckNotCharacterAfterAnd(unsigned c, unsigned and_with, + Label* on_not_equal) +{ + JitSpew(SPEW_PREFIX "CheckNotCharacterAfterAnd(%d, %d)", (int) c, (int) and_with); + + if (c == 0) { + masm.branchTest32(Assembler::NonZero, current_character, Imm32(and_with), + BranchOrBacktrack(on_not_equal)); + } else { + masm.move32(Imm32(and_with), temp0); + masm.and32(current_character, temp0); + masm.branch32(Assembler::NotEqual, temp0, Imm32(c), BranchOrBacktrack(on_not_equal)); + } +} + +void +NativeRegExpMacroAssembler::CheckCharacterGT(char16_t c, Label* on_greater) +{ + JitSpew(SPEW_PREFIX "CheckCharacterGT(%d)", (int) c); + masm.branch32(Assembler::GreaterThan, current_character, Imm32(c), + BranchOrBacktrack(on_greater)); +} + +void +NativeRegExpMacroAssembler::CheckCharacterLT(char16_t c, Label* on_less) +{ + JitSpew(SPEW_PREFIX "CheckCharacterLT(%d)", (int) c); + masm.branch32(Assembler::LessThan, current_character, Imm32(c), BranchOrBacktrack(on_less)); +} + +void +NativeRegExpMacroAssembler::CheckGreedyLoop(Label* on_tos_equals_current_position) +{ + JitSpew(SPEW_PREFIX "CheckGreedyLoop"); + + Label fallthrough; + masm.branchPtr(Assembler::NotEqual, + Address(backtrack_stack_pointer, -int(sizeof(void*))), current_position, + &fallthrough); + masm.subPtr(Imm32(sizeof(void*)), backtrack_stack_pointer); // Pop. + JumpOrBacktrack(on_tos_equals_current_position); + masm.bind(&fallthrough); +} + +void +NativeRegExpMacroAssembler::CheckNotBackReference(int start_reg, Label* on_no_match) +{ + JitSpew(SPEW_PREFIX "CheckNotBackReference(%d)", start_reg); + + Label fallthrough; + Label success; + Label fail; + + // Find length of back-referenced capture. + masm.loadPtr(register_location(start_reg), current_character); + masm.loadPtr(register_location(start_reg + 1), temp0); + masm.subPtr(current_character, temp0); // Length to check. + + // Fail on partial or illegal capture (start of capture after end of capture). + masm.branchPtr(Assembler::LessThan, temp0, ImmWord(0), BranchOrBacktrack(on_no_match)); + + // Succeed on empty capture (including no capture). + masm.branchPtr(Assembler::Equal, temp0, ImmWord(0), &fallthrough); + + // Check that there are sufficient characters left in the input. + masm.movePtr(current_position, temp1); + masm.addPtr(temp0, temp1); + masm.branchPtr(Assembler::GreaterThan, temp1, ImmWord(0), BranchOrBacktrack(on_no_match)); + + // Save register to make it available below. + masm.push(backtrack_stack_pointer); + + // Compute pointers to match string and capture string + masm.computeEffectiveAddress(BaseIndex(input_end_pointer, current_position, TimesOne), temp1); // Start of match. + masm.addPtr(input_end_pointer, current_character); // Start of capture. + masm.computeEffectiveAddress(BaseIndex(temp0, temp1, TimesOne), backtrack_stack_pointer); // End of match. + + Label loop; + masm.bind(&loop); + if (mode_ == ASCII) { + masm.load8ZeroExtend(Address(current_character, 0), temp0); + masm.load8ZeroExtend(Address(temp1, 0), temp2); + } else { + MOZ_ASSERT(mode_ == CHAR16); + masm.load16ZeroExtend(Address(current_character, 0), temp0); + masm.load16ZeroExtend(Address(temp1, 0), temp2); + } + masm.branch32(Assembler::NotEqual, temp0, temp2, &fail); + + // Increment pointers into capture and match string. + masm.addPtr(Imm32(char_size()), current_character); + masm.addPtr(Imm32(char_size()), temp1); + + // Check if we have reached end of match area. + masm.branchPtr(Assembler::Below, temp1, backtrack_stack_pointer, &loop); + masm.jump(&success); + + masm.bind(&fail); + + // Restore backtrack stack pointer. + masm.pop(backtrack_stack_pointer); + JumpOrBacktrack(on_no_match); + + masm.bind(&success); + + // Move current character position to position after match. + masm.movePtr(backtrack_stack_pointer, current_position); + masm.subPtr(input_end_pointer, current_position); + + // Restore backtrack stack pointer. + masm.pop(backtrack_stack_pointer); + + masm.bind(&fallthrough); +} + +void +NativeRegExpMacroAssembler::CheckNotBackReferenceIgnoreCase(int start_reg, Label* on_no_match, + bool unicode) +{ + JitSpew(SPEW_PREFIX "CheckNotBackReferenceIgnoreCase(%d, %d)", start_reg, unicode); + + Label fallthrough; + + masm.loadPtr(register_location(start_reg), current_character); // Index of start of capture + masm.loadPtr(register_location(start_reg + 1), temp1); // Index of end of capture + masm.subPtr(current_character, temp1); // Length of capture. + + // The length of a capture should not be negative. This can only happen + // if the end of the capture is unrecorded, or at a point earlier than + // the start of the capture. + masm.branchPtr(Assembler::LessThan, temp1, ImmWord(0), BranchOrBacktrack(on_no_match)); + + // If length is zero, either the capture is empty or it is completely + // uncaptured. In either case succeed immediately. + masm.branchPtr(Assembler::Equal, temp1, ImmWord(0), &fallthrough); + + // Check that there are sufficient characters left in the input. + masm.movePtr(current_position, temp0); + masm.addPtr(temp1, temp0); + masm.branchPtr(Assembler::GreaterThan, temp0, ImmWord(0), BranchOrBacktrack(on_no_match)); + + if (mode_ == ASCII) { + Label success, fail; + + // Save register contents to make the registers available below. After + // this, the temp0, temp2, and current_position registers are available. + masm.push(current_position); + + masm.addPtr(input_end_pointer, current_character); // Start of capture. + masm.addPtr(input_end_pointer, current_position); // Start of text to match against capture. + masm.addPtr(current_position, temp1); // End of text to match against capture. + + Label loop, loop_increment; + masm.bind(&loop); + masm.load8ZeroExtend(Address(current_position, 0), temp0); + masm.load8ZeroExtend(Address(current_character, 0), temp2); + masm.branch32(Assembler::Equal, temp0, temp2, &loop_increment); + + // Mismatch, try case-insensitive match (converting letters to lower-case). + masm.or32(Imm32(0x20), temp0); // Convert match character to lower-case. + + // Is temp0 a lowercase letter? + Label convert_capture; + masm.computeEffectiveAddress(Address(temp0, -'a'), temp2); + masm.branch32(Assembler::BelowOrEqual, temp2, Imm32(static_cast('z' - 'a')), + &convert_capture); + + // Latin-1: Check for values in range [224,254] but not 247. + masm.sub32(Imm32(224 - 'a'), temp2); + masm.branch32(Assembler::Above, temp2, Imm32(254 - 224), &fail); + + // Check for 247. + masm.branch32(Assembler::Equal, temp2, Imm32(247 - 224), &fail); + + masm.bind(&convert_capture); + + // Also convert capture character. + masm.load8ZeroExtend(Address(current_character, 0), temp2); + masm.or32(Imm32(0x20), temp2); + + masm.branch32(Assembler::NotEqual, temp0, temp2, &fail); + + masm.bind(&loop_increment); + + // Increment pointers into match and capture strings. + masm.addPtr(Imm32(1), current_character); + masm.addPtr(Imm32(1), current_position); + + // Compare to end of match, and loop if not done. + masm.branchPtr(Assembler::Below, current_position, temp1, &loop); + masm.jump(&success); + + masm.bind(&fail); + + // Restore original values before failing. + masm.pop(current_position); + JumpOrBacktrack(on_no_match); + + masm.bind(&success); + + // Drop original character position value. + masm.addToStackPtr(Imm32(sizeof(uintptr_t))); + + // Compute new value of character position after the matched part. + masm.subPtr(input_end_pointer, current_position); + } else { + MOZ_ASSERT(mode_ == CHAR16); + + // Note: temp1 needs to be saved/restored if it is volatile, as it is used after the call. + LiveGeneralRegisterSet volatileRegs(GeneralRegisterSet::Volatile()); + volatileRegs.takeUnchecked(temp0); + volatileRegs.takeUnchecked(temp2); + masm.PushRegsInMask(volatileRegs); + + // Set byte_offset1. + // Start of capture, where current_character already holds string-end negative offset. + masm.addPtr(input_end_pointer, current_character); + + // Set byte_offset2. + // Found by adding negative string-end offset of current position + // to end of string. + masm.addPtr(input_end_pointer, current_position); + + // Parameters are + // Address byte_offset1 - Address captured substring's start. + // Address byte_offset2 - Address of current character position. + // size_t byte_length - length of capture in bytes(!) + masm.setupUnalignedABICall(temp0); + masm.passABIArg(current_character); + masm.passABIArg(current_position); + masm.passABIArg(temp1); + if (!unicode) { + int (*fun)(const char16_t*, const char16_t*, size_t) = CaseInsensitiveCompareStrings; + masm.callWithABI(JS_FUNC_TO_DATA_PTR(void*, fun)); + } else { + int (*fun)(const char16_t*, const char16_t*, size_t) = CaseInsensitiveCompareUCStrings; + masm.callWithABI(JS_FUNC_TO_DATA_PTR(void*, fun)); + } + masm.storeCallInt32Result(temp0); + + masm.PopRegsInMask(volatileRegs); + + // Check if function returned non-zero for success or zero for failure. + masm.branchTest32(Assembler::Zero, temp0, temp0, BranchOrBacktrack(on_no_match)); + + // On success, increment position by length of capture. + masm.addPtr(temp1, current_position); + } + + masm.bind(&fallthrough); +} + +void +NativeRegExpMacroAssembler::CheckNotCharacterAfterMinusAnd(char16_t c, char16_t minus, char16_t and_with, + Label* on_not_equal) +{ + JitSpew(SPEW_PREFIX "CheckNotCharacterAfterMinusAnd(%d, %d, %d)", (int) c, + (int) minus, (int) and_with); + + masm.computeEffectiveAddress(Address(current_character, -minus), temp0); + if (c == 0) { + masm.branchTest32(Assembler::NonZero, temp0, Imm32(and_with), + BranchOrBacktrack(on_not_equal)); + } else { + masm.and32(Imm32(and_with), temp0); + masm.branch32(Assembler::NotEqual, temp0, Imm32(c), BranchOrBacktrack(on_not_equal)); + } +} + +void +NativeRegExpMacroAssembler::CheckCharacterInRange(char16_t from, char16_t to, + Label* on_in_range) +{ + JitSpew(SPEW_PREFIX "CheckCharacterInRange(%d, %d)", (int) from, (int) to); + + masm.computeEffectiveAddress(Address(current_character, -from), temp0); + masm.branch32(Assembler::BelowOrEqual, temp0, Imm32(to - from), BranchOrBacktrack(on_in_range)); +} + +void +NativeRegExpMacroAssembler::CheckCharacterNotInRange(char16_t from, char16_t to, + Label* on_not_in_range) +{ + JitSpew(SPEW_PREFIX "CheckCharacterNotInRange(%d, %d)", (int) from, (int) to); + + masm.computeEffectiveAddress(Address(current_character, -from), temp0); + masm.branch32(Assembler::Above, temp0, Imm32(to - from), BranchOrBacktrack(on_not_in_range)); +} + +void +NativeRegExpMacroAssembler::CheckBitInTable(uint8_t* table, Label* on_bit_set) +{ + JitSpew(SPEW_PREFIX "CheckBitInTable"); + + masm.movePtr(ImmPtr(table), temp0); + + // kTableMask is currently 127, so we need to mask even if the input is + // Latin1. V8 has the same issue. + static_assert(JSString::MAX_LATIN1_CHAR > kTableMask, + "No need to mask if MAX_LATIN1_CHAR <= kTableMask"); + masm.move32(Imm32(kTableSize - 1), temp1); + masm.and32(current_character, temp1); + + masm.load8ZeroExtend(BaseIndex(temp0, temp1, TimesOne), temp0); + masm.branchTest32(Assembler::NonZero, temp0, temp0, BranchOrBacktrack(on_bit_set)); +} + +void +NativeRegExpMacroAssembler::Fail() +{ + JitSpew(SPEW_PREFIX "Fail"); + + if (!global()) + masm.movePtr(ImmWord(RegExpRunStatus_Success_NotFound), temp0); + masm.jump(&exit_label_); +} + +void +NativeRegExpMacroAssembler::IfRegisterGE(int reg, int comparand, Label* if_ge) +{ + JitSpew(SPEW_PREFIX "IfRegisterGE(%d, %d)", reg, comparand); + masm.branchPtr(Assembler::GreaterThanOrEqual, register_location(reg), ImmWord(comparand), + BranchOrBacktrack(if_ge)); +} + +void +NativeRegExpMacroAssembler::IfRegisterLT(int reg, int comparand, Label* if_lt) +{ + JitSpew(SPEW_PREFIX "IfRegisterLT(%d, %d)", reg, comparand); + masm.branchPtr(Assembler::LessThan, register_location(reg), ImmWord(comparand), + BranchOrBacktrack(if_lt)); +} + +void +NativeRegExpMacroAssembler::IfRegisterEqPos(int reg, Label* if_eq) +{ + JitSpew(SPEW_PREFIX "IfRegisterEqPos(%d)", reg); + masm.branchPtr(Assembler::Equal, register_location(reg), current_position, + BranchOrBacktrack(if_eq)); +} + +void +NativeRegExpMacroAssembler::LoadCurrentCharacter(int cp_offset, Label* on_end_of_input, + bool check_bounds, int characters) +{ + JitSpew(SPEW_PREFIX "LoadCurrentCharacter(%d, %d)", cp_offset, characters); + + MOZ_ASSERT(cp_offset >= -1); // ^ and \b can look behind one character. + MOZ_ASSERT(cp_offset < (1<<30)); // Be sane! (And ensure negation works) + if (check_bounds) + CheckPosition(cp_offset + characters - 1, on_end_of_input); + LoadCurrentCharacterUnchecked(cp_offset, characters); +} + +void +NativeRegExpMacroAssembler::LoadCurrentCharacterUnchecked(int cp_offset, int characters) +{ + JitSpew(SPEW_PREFIX "LoadCurrentCharacterUnchecked(%d, %d)", cp_offset, characters); + + if (mode_ == ASCII) { + BaseIndex address(input_end_pointer, current_position, TimesOne, cp_offset); + if (characters == 4) { + masm.load32(address, current_character); + } else if (characters == 2) { + masm.load16ZeroExtend(address, current_character); + } else { + MOZ_ASSERT(characters == 1); + masm.load8ZeroExtend(address, current_character); + } + } else { + MOZ_ASSERT(mode_ == CHAR16); + MOZ_ASSERT(characters <= 2); + BaseIndex address(input_end_pointer, current_position, TimesOne, cp_offset * sizeof(char16_t)); + if (characters == 2) + masm.load32(address, current_character); + else + masm.load16ZeroExtend(address, current_character); + } +} + +void +NativeRegExpMacroAssembler::PopCurrentPosition() +{ + JitSpew(SPEW_PREFIX "PopCurrentPosition"); + + PopBacktrack(current_position); +} + +void +NativeRegExpMacroAssembler::PopRegister(int register_index) +{ + JitSpew(SPEW_PREFIX "PopRegister(%d)", register_index); + + PopBacktrack(temp0); + masm.storePtr(temp0, register_location(register_index)); +} + +void +NativeRegExpMacroAssembler::PushBacktrack(Label* label) +{ + JitSpew(SPEW_PREFIX "PushBacktrack"); + + CodeOffset patchOffset = masm.movWithPatch(ImmPtr(nullptr), temp0); + + MOZ_ASSERT(!label->bound()); + + { + AutoEnterOOMUnsafeRegion oomUnsafe; + if (!labelPatches.append(LabelPatch(label, patchOffset))) + oomUnsafe.crash("NativeRegExpMacroAssembler::PushBacktrack"); + } + + PushBacktrack(temp0); + CheckBacktrackStackLimit(); +} + +void +NativeRegExpMacroAssembler::BindBacktrack(Label* label) +{ + JitSpew(SPEW_PREFIX "BindBacktrack"); + + Bind(label); + + for (size_t i = 0; i < labelPatches.length(); i++) { + LabelPatch& v = labelPatches[i]; + if (v.label == label) { + v.labelOffset = label->offset(); + v.label = nullptr; + break; + } + } +} + +void +NativeRegExpMacroAssembler::PushBacktrack(Register source) +{ + JitSpew(SPEW_PREFIX "PushBacktrack"); + + MOZ_ASSERT(source != backtrack_stack_pointer); + + // Notice: This updates flags, unlike normal Push. + masm.storePtr(source, Address(backtrack_stack_pointer, 0)); + masm.addPtr(Imm32(sizeof(void*)), backtrack_stack_pointer); +} + +void +NativeRegExpMacroAssembler::PushBacktrack(int32_t value) +{ + JitSpew(SPEW_PREFIX "PushBacktrack(%d)", (int) value); + + // Notice: This updates flags, unlike normal Push. + masm.storePtr(ImmWord(value), Address(backtrack_stack_pointer, 0)); + masm.addPtr(Imm32(sizeof(void*)), backtrack_stack_pointer); +} + +void +NativeRegExpMacroAssembler::PopBacktrack(Register target) +{ + JitSpew(SPEW_PREFIX "PopBacktrack"); + + MOZ_ASSERT(target != backtrack_stack_pointer); + + // Notice: This updates flags, unlike normal Pop. + masm.subPtr(Imm32(sizeof(void*)), backtrack_stack_pointer); + masm.loadPtr(Address(backtrack_stack_pointer, 0), target); +} + +void +NativeRegExpMacroAssembler::CheckBacktrackStackLimit() +{ + JitSpew(SPEW_PREFIX "CheckBacktrackStackLimit"); + + const void* limitAddr = runtime->regexpStack.addressOfLimit(); + + Label no_stack_overflow; + masm.branchPtr(Assembler::AboveOrEqual, AbsoluteAddress(limitAddr), + backtrack_stack_pointer, &no_stack_overflow); + + // Copy the stack pointer before the call() instruction modifies it. + masm.moveStackPtrTo(temp2); + + masm.call(&stack_overflow_label_); + masm.bind(&no_stack_overflow); + + // Exit with an exception if the call failed. + masm.branchTest32(Assembler::Zero, temp0, temp0, &exit_with_exception_label_); +} + +void +NativeRegExpMacroAssembler::PushCurrentPosition() +{ + JitSpew(SPEW_PREFIX "PushCurrentPosition"); + + PushBacktrack(current_position); +} + +void +NativeRegExpMacroAssembler::PushRegister(int register_index, StackCheckFlag check_stack_limit) +{ + JitSpew(SPEW_PREFIX "PushRegister(%d)", register_index); + + masm.loadPtr(register_location(register_index), temp0); + PushBacktrack(temp0); + if (check_stack_limit) + CheckBacktrackStackLimit(); +} + +void +NativeRegExpMacroAssembler::ReadCurrentPositionFromRegister(int reg) +{ + JitSpew(SPEW_PREFIX "ReadCurrentPositionFromRegister(%d)", reg); + + masm.loadPtr(register_location(reg), current_position); +} + +void +NativeRegExpMacroAssembler::WriteCurrentPositionToRegister(int reg, int cp_offset) +{ + JitSpew(SPEW_PREFIX "WriteCurrentPositionToRegister(%d, %d)", reg, cp_offset); + + if (cp_offset == 0) { + masm.storePtr(current_position, register_location(reg)); + } else { + masm.computeEffectiveAddress(Address(current_position, cp_offset * char_size()), temp0); + masm.storePtr(temp0, register_location(reg)); + } +} + +void +NativeRegExpMacroAssembler::ReadBacktrackStackPointerFromRegister(int reg) +{ + JitSpew(SPEW_PREFIX "ReadBacktrackStackPointerFromRegister(%d)", reg); + + masm.loadPtr(register_location(reg), backtrack_stack_pointer); + masm.addPtr(Address(masm.getStackPointer(), + offsetof(FrameData, backtrackStackBase)), backtrack_stack_pointer); +} + +void +NativeRegExpMacroAssembler::WriteBacktrackStackPointerToRegister(int reg) +{ + JitSpew(SPEW_PREFIX "WriteBacktrackStackPointerToRegister(%d)", reg); + + masm.movePtr(backtrack_stack_pointer, temp0); + masm.subPtr(Address(masm.getStackPointer(), + offsetof(FrameData, backtrackStackBase)), temp0); + masm.storePtr(temp0, register_location(reg)); +} + +void +NativeRegExpMacroAssembler::SetCurrentPositionFromEnd(int by) +{ + JitSpew(SPEW_PREFIX "SetCurrentPositionFromEnd(%d)", by); + + Label after_position; + masm.branchPtr(Assembler::GreaterThanOrEqual, current_position, + ImmWord(-by * char_size()), &after_position); + masm.movePtr(ImmWord(-by * char_size()), current_position); + + // On RegExp code entry (where this operation is used), the character before + // the current position is expected to be already loaded. + // We have advanced the position, so it's safe to read backwards. + LoadCurrentCharacterUnchecked(-1, 1); + masm.bind(&after_position); +} + +void +NativeRegExpMacroAssembler::SetRegister(int register_index, int to) +{ + JitSpew(SPEW_PREFIX "SetRegister(%d, %d)", register_index, to); + + MOZ_ASSERT(register_index >= num_saved_registers_); // Reserved for positions! + masm.storePtr(ImmWord(to), register_location(register_index)); +} + +bool +NativeRegExpMacroAssembler::Succeed() +{ + JitSpew(SPEW_PREFIX "Succeed"); + + masm.jump(&success_label_); + return global(); +} + +void +NativeRegExpMacroAssembler::ClearRegisters(int reg_from, int reg_to) +{ + JitSpew(SPEW_PREFIX "ClearRegisters(%d, %d)", reg_from, reg_to); + + MOZ_ASSERT(reg_from <= reg_to); + masm.loadPtr(Address(masm.getStackPointer(), offsetof(FrameData, inputStartMinusOne)), temp0); + for (int reg = reg_from; reg <= reg_to; reg++) + masm.storePtr(temp0, register_location(reg)); +} + +void +NativeRegExpMacroAssembler::CheckPosition(int cp_offset, Label* on_outside_input) +{ + JitSpew(SPEW_PREFIX "CheckPosition(%d)", cp_offset); + masm.branchPtr(Assembler::GreaterThanOrEqual, current_position, + ImmWord(-cp_offset * char_size()), BranchOrBacktrack(on_outside_input)); +} + +Label* +NativeRegExpMacroAssembler::BranchOrBacktrack(Label* branch) +{ + if (branch) + return branch; + return &backtrack_label_; +} + +void +NativeRegExpMacroAssembler::JumpOrBacktrack(Label* to) +{ + JitSpew(SPEW_PREFIX "JumpOrBacktrack"); + + if (to) + masm.jump(to); + else + Backtrack(); +} + +bool +NativeRegExpMacroAssembler::CheckSpecialCharacterClass(char16_t type, Label* on_no_match) +{ + JitSpew(SPEW_PREFIX "CheckSpecialCharacterClass(%d)", (int) type); + + Label* branch = BranchOrBacktrack(on_no_match); + + // Range checks (c in min..max) are generally implemented by an unsigned + // (c - min) <= (max - min) check + switch (type) { + case 's': + // Match space-characters. + if (mode_ == ASCII) { + // One byte space characters are '\t'..'\r', ' ' and \u00a0. + Label success; + masm.branch32(Assembler::Equal, current_character, Imm32(' '), &success); + + // Check range 0x09..0x0d. + masm.computeEffectiveAddress(Address(current_character, -'\t'), temp0); + masm.branch32(Assembler::BelowOrEqual, temp0, Imm32('\r' - '\t'), &success); + + // \u00a0 (NBSP). + masm.branch32(Assembler::NotEqual, temp0, Imm32(0x00a0 - '\t'), branch); + + masm.bind(&success); + return true; + } + return false; + case 'S': + // The emitted code for generic character classes is good enough. + return false; + case 'd': + // Match ASCII digits ('0'..'9') + masm.computeEffectiveAddress(Address(current_character, -'0'), temp0); + masm.branch32(Assembler::Above, temp0, Imm32('9' - '0'), branch); + return true; + case 'D': + // Match non ASCII-digits + masm.computeEffectiveAddress(Address(current_character, -'0'), temp0); + masm.branch32(Assembler::BelowOrEqual, temp0, Imm32('9' - '0'), branch); + return true; + case '.': { + // Match non-newlines (not 0x0a('\n'), 0x0d('\r'), 0x2028 and 0x2029) + masm.move32(current_character, temp0); + masm.xor32(Imm32(0x01), temp0); + + // See if current character is '\n'^1 or '\r'^1, i.e., 0x0b or 0x0c + masm.sub32(Imm32(0x0b), temp0); + masm.branch32(Assembler::BelowOrEqual, temp0, Imm32(0x0c - 0x0b), branch); + if (mode_ == CHAR16) { + // Compare original value to 0x2028 and 0x2029, using the already + // computed (current_char ^ 0x01 - 0x0b). I.e., check for + // 0x201d (0x2028 - 0x0b) or 0x201e. + masm.sub32(Imm32(0x2028 - 0x0b), temp0); + masm.branch32(Assembler::BelowOrEqual, temp0, Imm32(0x2029 - 0x2028), branch); + } + return true; + } + case 'w': { + if (mode_ != ASCII) { + // Table is 128 entries, so all ASCII characters can be tested. + masm.branch32(Assembler::Above, current_character, Imm32('z'), branch); + } + MOZ_ASSERT(0 == word_character_map[0]); // Character '\0' is not a word char. + masm.movePtr(ImmPtr(word_character_map), temp0); + masm.load8ZeroExtend(BaseIndex(temp0, current_character, TimesOne), temp0); + masm.branchTest32(Assembler::Zero, temp0, temp0, branch); + return true; + } + case 'W': { + Label done; + if (mode_ != ASCII) { + // Table is 128 entries, so all ASCII characters can be tested. + masm.branch32(Assembler::Above, current_character, Imm32('z'), &done); + } + MOZ_ASSERT(0 == word_character_map[0]); // Character '\0' is not a word char. + masm.movePtr(ImmPtr(word_character_map), temp0); + masm.load8ZeroExtend(BaseIndex(temp0, current_character, TimesOne), temp0); + masm.branchTest32(Assembler::NonZero, temp0, temp0, branch); + if (mode_ != ASCII) + masm.bind(&done); + return true; + } + // Non-standard classes (with no syntactic shorthand) used internally. + case '*': + // Match any character. + return true; + case 'n': { + // Match newlines (0x0a('\n'), 0x0d('\r'), 0x2028 or 0x2029). + // The opposite of '.'. + masm.move32(current_character, temp0); + masm.xor32(Imm32(0x01), temp0); + + // See if current character is '\n'^1 or '\r'^1, i.e., 0x0b or 0x0c + masm.sub32(Imm32(0x0b), temp0); + + if (mode_ == ASCII) { + masm.branch32(Assembler::Above, temp0, Imm32(0x0c - 0x0b), branch); + } else { + Label done; + masm.branch32(Assembler::BelowOrEqual, temp0, Imm32(0x0c - 0x0b), &done); + MOZ_ASSERT(CHAR16 == mode_); + + // Compare original value to 0x2028 and 0x2029, using the already + // computed (current_char ^ 0x01 - 0x0b). I.e., check for + // 0x201d (0x2028 - 0x0b) or 0x201e. + masm.sub32(Imm32(0x2028 - 0x0b), temp0); + masm.branch32(Assembler::Above, temp0, Imm32(1), branch); + + masm.bind(&done); + } + return true; + } + // No custom implementation (yet): + default: + return false; + } +} + +bool +NativeRegExpMacroAssembler::CanReadUnaligned() +{ +#if defined(JS_CODEGEN_ARM) + return !jit::HasAlignmentFault(); +#elif defined(JS_CODEGEN_MIPS32) || defined(JS_CODEGEN_MIPS64) + return false; +#else + return true; +#endif +} + +const uint8_t +NativeRegExpMacroAssembler::word_character_map[] = +{ + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, // '0' - '7' + 0xffu, 0xffu, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, // '8' - '9' + + 0x00u, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, // 'A' - 'G' + 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, // 'H' - 'O' + 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, // 'P' - 'W' + 0xffu, 0xffu, 0xffu, 0x00u, 0x00u, 0x00u, 0x00u, 0xffu, // 'X' - 'Z', '_' + + 0x00u, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, // 'a' - 'g' + 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, // 'h' - 'o' + 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, // 'p' - 'w' + 0xffu, 0xffu, 0xffu, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, // 'x' - 'z' + + // Latin-1 range + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, + 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, +}; diff --git a/js/src/irregexp/NativeRegExpMacroAssembler.h b/js/src/irregexp/NativeRegExpMacroAssembler.h new file mode 100644 index 000000000..7a72e252f --- /dev/null +++ b/js/src/irregexp/NativeRegExpMacroAssembler.h @@ -0,0 +1,226 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- + * vim: set ts=8 sts=4 et sw=4 tw=99: */ + +// Copyright 2012 the V8 project authors. All rights reserved. +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived +// from this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#ifndef V8_NATIVE_REGEXP_MACRO_ASSEMBLER_H_ +#define V8_NATIVE_REGEXP_MACRO_ASSEMBLER_H_ + +#include "irregexp/RegExpMacroAssembler.h" + +namespace js { +namespace irregexp { + +struct InputOutputData +{ + const void* inputStart; + const void* inputEnd; + + // Index into inputStart (in chars) at which to begin matching. + size_t startIndex; + size_t* endIndex; + + MatchPairs* matches; + + // RegExpMacroAssembler::Result for non-global regexps, number of captures + // for global regexps. + int32_t result; + + template + InputOutputData(const CharT* inputStart, const CharT* inputEnd, + size_t startIndex, MatchPairs* matches, size_t* endIndex) + : inputStart(inputStart), + inputEnd(inputEnd), + startIndex(startIndex), + endIndex(endIndex), + matches(matches), + result(0) + {} +}; + +struct FrameData +{ + // Copy of the input/output data's data. + char16_t* inputStart; + size_t startIndex; + size_t* endIndex; + + // Pointer to the character before the input start. + char16_t* inputStartMinusOne; + + // Copy of the input MatchPairs registers, may be modified by JIT code. + int32_t* outputRegisters; + int32_t numOutputRegisters; + + int32_t successfulCaptures; + + void* backtrackStackBase; +}; + +class MOZ_STACK_CLASS NativeRegExpMacroAssembler final : public RegExpMacroAssembler +{ + public: + // Type of input string to generate code for. + enum Mode { ASCII = 1, CHAR16 = 2 }; + + NativeRegExpMacroAssembler(LifoAlloc* alloc, RegExpShared* shared, + JSRuntime* rt, Mode mode, int registers_to_save); + + // Inherited virtual methods. + RegExpCode GenerateCode(JSContext* cx, bool match_only); + int stack_limit_slack(); + bool CanReadUnaligned(); + void AdvanceCurrentPosition(int by); + void AdvanceRegister(int reg, int by); + void Backtrack(); + void Bind(jit::Label* label); + void CheckAtStart(jit::Label* on_at_start); + void CheckCharacter(unsigned c, jit::Label* on_equal); + void CheckCharacterAfterAnd(unsigned c, unsigned and_with, jit::Label* on_equal); + void CheckCharacterGT(char16_t limit, jit::Label* on_greater); + void CheckCharacterLT(char16_t limit, jit::Label* on_less); + void CheckGreedyLoop(jit::Label* on_tos_equals_current_position); + void CheckNotAtStart(jit::Label* on_not_at_start); + void CheckNotBackReference(int start_reg, jit::Label* on_no_match); + void CheckNotBackReferenceIgnoreCase(int start_reg, jit::Label* on_no_match, bool unicode); + void CheckNotCharacter(unsigned c, jit::Label* on_not_equal); + void CheckNotCharacterAfterAnd(unsigned c, unsigned and_with, jit::Label* on_not_equal); + void CheckNotCharacterAfterMinusAnd(char16_t c, char16_t minus, char16_t and_with, + jit::Label* on_not_equal); + void CheckCharacterInRange(char16_t from, char16_t to, + jit::Label* on_in_range); + void CheckCharacterNotInRange(char16_t from, char16_t to, + jit::Label* on_not_in_range); + void CheckBitInTable(uint8_t* table, jit::Label* on_bit_set); + void CheckPosition(int cp_offset, jit::Label* on_outside_input); + void JumpOrBacktrack(jit::Label* to); + bool CheckSpecialCharacterClass(char16_t type, jit::Label* on_no_match); + void Fail(); + void IfRegisterGE(int reg, int comparand, jit::Label* if_ge); + void IfRegisterLT(int reg, int comparand, jit::Label* if_lt); + void IfRegisterEqPos(int reg, jit::Label* if_eq); + void LoadCurrentCharacter(int cp_offset, jit::Label* on_end_of_input, + bool check_bounds = true, int characters = 1); + void PopCurrentPosition(); + void PopRegister(int register_index); + void PushCurrentPosition(); + void PushRegister(int register_index, StackCheckFlag check_stack_limit); + void ReadCurrentPositionFromRegister(int reg); + void ReadBacktrackStackPointerFromRegister(int reg); + void SetCurrentPositionFromEnd(int by); + void SetRegister(int register_index, int to); + bool Succeed(); + void WriteCurrentPositionToRegister(int reg, int cp_offset); + void ClearRegisters(int reg_from, int reg_to); + void WriteBacktrackStackPointerToRegister(int reg); + void PushBacktrack(jit::Label* label); + void BindBacktrack(jit::Label* label); + + // Compares two-byte strings case insensitively. + // Called from generated RegExp code. + static int CaseInsensitiveCompareUC16(jit::Address byte_offset1, + jit::Address byte_offset2, + size_t byte_length); + + // Byte map of one byte characters with a 0xff if the character is a word + // character (digit, letter or underscore) and 0x00 otherwise. + // Used by generated RegExp code. + static const uint8_t word_character_map[256]; + + // Byte size of chars in the string to match (decided by the Mode argument) + inline int char_size() { return static_cast(mode_); } + inline jit::Scale factor() { return mode_ == CHAR16 ? jit::TimesTwo : jit::TimesOne; } + + jit::Label* BranchOrBacktrack(jit::Label* branch); + + // Pushes a register or constant on the backtrack stack. Decrements the + // stack pointer by a word size and stores the register's value there. + void PushBacktrack(jit::Register value); + void PushBacktrack(int32_t value); + + // Pop a value from the backtrack stack. + void PopBacktrack(jit::Register target); + + // Check whether we are exceeding the stack limit on the backtrack stack. + void CheckBacktrackStackLimit(); + + void LoadCurrentCharacterUnchecked(int cp_offset, int characters); + + private: + jit::MacroAssembler masm; + + JSRuntime* runtime; + Mode mode_; + jit::Label entry_label_; + jit::Label start_label_; + jit::Label backtrack_label_; + jit::Label success_label_; + jit::Label exit_label_; + jit::Label stack_overflow_label_; + jit::Label exit_with_exception_label_; + + // Set of registers which are used by the code generator, and as such which + // are saved. + jit::LiveGeneralRegisterSet savedNonVolatileRegisters; + + struct LabelPatch { + // Once it is bound via BindBacktrack, |label| becomes null and + // |labelOffset| is set. + jit::Label* label; + size_t labelOffset; + + jit::CodeOffset patchOffset; + + LabelPatch(jit::Label* label, jit::CodeOffset patchOffset) + : label(label), labelOffset(0), patchOffset(patchOffset) + {} + }; + + Vector labelPatches; + + // See RegExpMacroAssembler.cpp for the meaning of these registers. + jit::Register input_end_pointer; + jit::Register current_character; + jit::Register current_position; + jit::Register backtrack_stack_pointer; + jit::Register temp0, temp1, temp2; + + // The frame_pointer-relative location of a regexp register. + jit::Address register_location(int register_index) { + checkRegister(register_index); + return jit::Address(masm.getStackPointer(), register_offset(register_index)); + } + + int32_t register_offset(int register_index) { + return sizeof(FrameData) + register_index * sizeof(void*); + } +}; + +} } // namespace js::irregexp + +#endif // V8_NATIVE_REGEXP_MACRO_ASSEMBLER_H_ diff --git a/js/src/irregexp/RegExpAST.cpp b/js/src/irregexp/RegExpAST.cpp new file mode 100644 index 000000000..8dfd99057 --- /dev/null +++ b/js/src/irregexp/RegExpAST.cpp @@ -0,0 +1,265 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- + * vim: set ts=8 sts=4 et sw=4 tw=99: */ + +// Copyright 2012 the V8 project authors. All rights reserved. +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived +// from this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include "irregexp/RegExpAST.h" + +using namespace js; +using namespace js::irregexp; + +#define MAKE_ACCEPT(Name) \ + void* RegExp##Name::Accept(RegExpVisitor* visitor, void* data) { \ + return visitor->Visit##Name(this, data); \ + } +FOR_EACH_REG_EXP_TREE_TYPE(MAKE_ACCEPT) +#undef MAKE_ACCEPT + +#define MAKE_TYPE_CASE(Name) \ + RegExp##Name* RegExpTree::As##Name() { \ + return nullptr; \ + } \ + bool RegExpTree::Is##Name() { return false; } +FOR_EACH_REG_EXP_TREE_TYPE(MAKE_TYPE_CASE) +#undef MAKE_TYPE_CASE + +#define MAKE_TYPE_CASE(Name) \ + RegExp##Name* RegExp##Name::As##Name() { \ + return this; \ + } \ + bool RegExp##Name::Is##Name() { return true; } +FOR_EACH_REG_EXP_TREE_TYPE(MAKE_TYPE_CASE) +#undef MAKE_TYPE_CASE + +static Interval +ListCaptureRegisters(const RegExpTreeVector& children) +{ + Interval result = Interval::Empty(); + for (size_t i = 0; i < children.length(); i++) + result = result.Union(children[i]->CaptureRegisters()); + return result; +} + +// ---------------------------------------------------------------------------- +// RegExpDisjunction + +RegExpDisjunction::RegExpDisjunction(RegExpTreeVector* alternatives) + : alternatives_(alternatives) +{ + MOZ_ASSERT(alternatives->length() > 1); + RegExpTree* first_alternative = (*alternatives)[0]; + min_match_ = first_alternative->min_match(); + max_match_ = first_alternative->max_match(); + for (size_t i = 1; i < alternatives->length(); i++) { + RegExpTree* alternative = (*alternatives)[i]; + min_match_ = Min(min_match_, alternative->min_match()); + max_match_ = Max(max_match_, alternative->max_match()); + } +} + +Interval +RegExpDisjunction::CaptureRegisters() +{ + return ListCaptureRegisters(alternatives()); +} + +bool +RegExpDisjunction::IsAnchoredAtStart() +{ + const RegExpTreeVector& alternatives = this->alternatives(); + for (size_t i = 0; i < alternatives.length(); i++) { + if (!alternatives[i]->IsAnchoredAtStart()) + return false; + } + return true; +} + +bool +RegExpDisjunction::IsAnchoredAtEnd() +{ + const RegExpTreeVector& alternatives = this->alternatives(); + for (size_t i = 0; i < alternatives.length(); i++) { + if (!alternatives[i]->IsAnchoredAtEnd()) + return false; + } + return true; +} + +// ---------------------------------------------------------------------------- +// RegExpAlternative + +static int IncreaseBy(int previous, int increase) +{ + if (RegExpTree::kInfinity - previous < increase) + return RegExpTree::kInfinity; + return previous + increase; +} + +RegExpAlternative::RegExpAlternative(RegExpTreeVector* nodes) + : nodes_(nodes), + min_match_(0), + max_match_(0) +{ + MOZ_ASSERT(nodes->length() > 1); + for (size_t i = 0; i < nodes->length(); i++) { + RegExpTree* node = (*nodes)[i]; + int node_min_match = node->min_match(); + min_match_ = IncreaseBy(min_match_, node_min_match); + int node_max_match = node->max_match(); + max_match_ = IncreaseBy(max_match_, node_max_match); + } +} + +Interval +RegExpAlternative::CaptureRegisters() +{ + return ListCaptureRegisters(nodes()); +} + +bool +RegExpAlternative::IsAnchoredAtStart() +{ + const RegExpTreeVector& nodes = this->nodes(); + for (size_t i = 0; i < nodes.length(); i++) { + RegExpTree* node = nodes[i]; + if (node->IsAnchoredAtStart()) { return true; } + if (node->max_match() > 0) { return false; } + } + return false; +} + +bool +RegExpAlternative::IsAnchoredAtEnd() +{ + const RegExpTreeVector& nodes = this->nodes(); + for (int i = nodes.length() - 1; i >= 0; i--) { + RegExpTree* node = nodes[i]; + if (node->IsAnchoredAtEnd()) { return true; } + if (node->max_match() > 0) { return false; } + } + return false; +} + +// ---------------------------------------------------------------------------- +// RegExpAssertion + +bool +RegExpAssertion::IsAnchoredAtStart() +{ + return assertion_type() == RegExpAssertion::START_OF_INPUT; +} + +bool +RegExpAssertion::IsAnchoredAtEnd() +{ + return assertion_type() == RegExpAssertion::END_OF_INPUT; +} + +// ---------------------------------------------------------------------------- +// RegExpCharacterClass + +void +RegExpCharacterClass::AppendToText(RegExpText* text) +{ + text->AddElement(TextElement::CharClass(this)); +} + +CharacterRangeVector& +CharacterSet::ranges(LifoAlloc* alloc) +{ + if (ranges_ == nullptr) { + ranges_ = alloc->newInfallible(*alloc); + CharacterRange::AddClassEscape(alloc, standard_set_type_, ranges_); + } + return *ranges_; +} + +// ---------------------------------------------------------------------------- +// RegExpAtom + +void +RegExpAtom::AppendToText(RegExpText* text) +{ + text->AddElement(TextElement::Atom(this)); +} + +// ---------------------------------------------------------------------------- +// RegExpText + +void +RegExpText::AppendToText(RegExpText* text) +{ + for (size_t i = 0; i < elements().length(); i++) + text->AddElement(elements()[i]); +} + +// ---------------------------------------------------------------------------- +// RegExpQuantifier + +Interval +RegExpQuantifier::CaptureRegisters() +{ + return body()->CaptureRegisters(); +} + +// ---------------------------------------------------------------------------- +// RegExpCapture + +bool +RegExpCapture::IsAnchoredAtStart() +{ + return body()->IsAnchoredAtStart(); +} + +bool +RegExpCapture::IsAnchoredAtEnd() +{ + return body()->IsAnchoredAtEnd(); +} + +Interval +RegExpCapture::CaptureRegisters() +{ + Interval self(StartRegister(index()), EndRegister(index())); + return self.Union(body()->CaptureRegisters()); +} + +// ---------------------------------------------------------------------------- +// RegExpLookahead + +Interval +RegExpLookahead::CaptureRegisters() +{ + return body()->CaptureRegisters(); +} + +bool +RegExpLookahead::IsAnchoredAtStart() +{ + return is_positive() && body()->IsAnchoredAtStart(); +} diff --git a/js/src/irregexp/RegExpAST.h b/js/src/irregexp/RegExpAST.h new file mode 100644 index 000000000..7bda6fc7e --- /dev/null +++ b/js/src/irregexp/RegExpAST.h @@ -0,0 +1,449 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- + * vim: set ts=8 sts=4 et sw=4 tw=99: */ + +// Copyright 2012 the V8 project authors. All rights reserved. +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived +// from this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#ifndef V8_REGEXP_AST_H_ +#define V8_REGEXP_AST_H_ + +// Prevent msvc build failures as indicated in bug 1205328 +#ifdef min +# undef min +#endif +#ifdef max +# undef max +#endif + +#include "irregexp/RegExpEngine.h" + +namespace js { +namespace irregexp { + +class RegExpCompiler; +class RegExpNode; + +class RegExpVisitor +{ + public: + virtual ~RegExpVisitor() { } +#define MAKE_CASE(Name) \ + virtual void* Visit##Name(RegExp##Name*, void* data) = 0; + FOR_EACH_REG_EXP_TREE_TYPE(MAKE_CASE) +#undef MAKE_CASE +}; + +class RegExpTree +{ + public: + static const int kInfinity = INT32_MAX; + virtual ~RegExpTree() {} + virtual RegExpNode* ToNode(RegExpCompiler* compiler, + RegExpNode* on_success) = 0; + virtual bool IsTextElement() { return false; } + virtual bool IsAnchoredAtStart() { return false; } + virtual bool IsAnchoredAtEnd() { return false; } + virtual int min_match() = 0; + virtual int max_match() = 0; + // Returns the interval of registers used for captures within this + // expression. + virtual Interval CaptureRegisters() { return Interval::Empty(); } + virtual void AppendToText(RegExpText* text) { + MOZ_CRASH("Bad call"); + } +#define MAKE_ASTYPE(Name) \ + virtual RegExp##Name* As##Name(); \ + virtual bool Is##Name(); + FOR_EACH_REG_EXP_TREE_TYPE(MAKE_ASTYPE) +#undef MAKE_ASTYPE +}; + +typedef InfallibleVector RegExpTreeVector; + +class RegExpDisjunction : public RegExpTree +{ + public: + explicit RegExpDisjunction(RegExpTreeVector* alternatives); + virtual void* Accept(RegExpVisitor* visitor, void* data); + virtual RegExpNode* ToNode(RegExpCompiler* compiler, + RegExpNode* on_success); + virtual RegExpDisjunction* AsDisjunction(); + virtual Interval CaptureRegisters(); + virtual bool IsDisjunction(); + virtual bool IsAnchoredAtStart(); + virtual bool IsAnchoredAtEnd(); + virtual int min_match() { return min_match_; } + virtual int max_match() { return max_match_; } + + const RegExpTreeVector& alternatives() { return *alternatives_; } + + private: + RegExpTreeVector* alternatives_; + int min_match_; + int max_match_; +}; + +class RegExpAlternative : public RegExpTree +{ + public: + explicit RegExpAlternative(RegExpTreeVector* nodes); + virtual void* Accept(RegExpVisitor* visitor, void* data); + virtual RegExpNode* ToNode(RegExpCompiler* compiler, + RegExpNode* on_success); + virtual RegExpAlternative* AsAlternative(); + virtual Interval CaptureRegisters(); + virtual bool IsAlternative(); + virtual bool IsAnchoredAtStart(); + virtual bool IsAnchoredAtEnd(); + virtual int min_match() { return min_match_; } + virtual int max_match() { return max_match_; } + + const RegExpTreeVector& nodes() { return *nodes_; } + + private: + RegExpTreeVector* nodes_; + int min_match_; + int max_match_; +}; + +class RegExpAssertion : public RegExpTree { + public: + enum AssertionType { + START_OF_LINE, + START_OF_INPUT, + END_OF_LINE, + END_OF_INPUT, + BOUNDARY, + NON_BOUNDARY, + NOT_AFTER_LEAD_SURROGATE, + NOT_IN_SURROGATE_PAIR + }; + explicit RegExpAssertion(AssertionType type) : assertion_type_(type) { } + virtual void* Accept(RegExpVisitor* visitor, void* data); + virtual RegExpNode* ToNode(RegExpCompiler* compiler, + RegExpNode* on_success); + virtual RegExpAssertion* AsAssertion(); + virtual bool IsAssertion(); + virtual bool IsAnchoredAtStart(); + virtual bool IsAnchoredAtEnd(); + virtual int min_match() { return 0; } + virtual int max_match() { return 0; } + AssertionType assertion_type() { return assertion_type_; } + private: + AssertionType assertion_type_; +}; + +class CharacterSet +{ + public: + explicit CharacterSet(char16_t standard_set_type) + : ranges_(nullptr), + standard_set_type_(standard_set_type) + {} + explicit CharacterSet(CharacterRangeVector* ranges) + : ranges_(ranges), + standard_set_type_(0) + {} + + CharacterRangeVector& ranges(LifoAlloc* alloc); + char16_t standard_set_type() { return standard_set_type_; } + void set_standard_set_type(char16_t special_set_type) { + standard_set_type_ = special_set_type; + } + bool is_standard() { return standard_set_type_ != 0; } + void Canonicalize(); + + private: + CharacterRangeVector* ranges_; + + // If non-zero, the value represents a standard set (e.g., all whitespace + // characters) without having to expand the ranges. + char16_t standard_set_type_; +}; + +class RegExpCharacterClass : public RegExpTree +{ + public: + RegExpCharacterClass(CharacterRangeVector* ranges, bool is_negated) + : set_(ranges), + is_negated_(is_negated) + {} + + explicit RegExpCharacterClass(char16_t type) + : set_(type), + is_negated_(false) + {} + + virtual void* Accept(RegExpVisitor* visitor, void* data); + virtual RegExpNode* ToNode(RegExpCompiler* compiler, + RegExpNode* on_success); + virtual RegExpCharacterClass* AsCharacterClass(); + virtual bool IsCharacterClass(); + virtual bool IsTextElement() { return true; } + virtual int min_match() { return 1; } + virtual int max_match() { return 1; } + virtual void AppendToText(RegExpText* text); + + CharacterSet character_set() { return set_; } + + // TODO(lrn): Remove need for complex version if is_standard that + // recognizes a mangled standard set and just do { return set_.is_special(); } + bool is_standard(LifoAlloc* alloc); + + // Returns a value representing the standard character set if is_standard() + // returns true. + // Currently used values are: + // s : unicode whitespace + // S : unicode non-whitespace + // w : ASCII word character (digit, letter, underscore) + // W : non-ASCII word character + // d : ASCII digit + // D : non-ASCII digit + // . : non-unicode non-newline + // * : All characters + char16_t standard_type() { return set_.standard_set_type(); } + + CharacterRangeVector& ranges(LifoAlloc* alloc) { return set_.ranges(alloc); } + bool is_negated() { return is_negated_; } + + private: + CharacterSet set_; + bool is_negated_; +}; + +typedef InfallibleVector CharacterVector; + +class RegExpAtom : public RegExpTree +{ + public: + explicit RegExpAtom(CharacterVector* data) + : data_(data) + {} + + virtual void* Accept(RegExpVisitor* visitor, void* data); + virtual RegExpNode* ToNode(RegExpCompiler* compiler, + RegExpNode* on_success); + virtual RegExpAtom* AsAtom(); + virtual bool IsAtom(); + virtual bool IsTextElement() { return true; } + virtual int min_match() { return data_->length(); } + virtual int max_match() { return data_->length(); } + virtual void AppendToText(RegExpText* text); + + const CharacterVector& data() { return *data_; } + int length() { return data_->length(); } + + private: + CharacterVector* data_; +}; + +class RegExpText : public RegExpTree +{ + public: + explicit RegExpText(LifoAlloc* alloc) + : elements_(*alloc), length_(0) + {} + + virtual void* Accept(RegExpVisitor* visitor, void* data); + virtual RegExpNode* ToNode(RegExpCompiler* compiler, + RegExpNode* on_success); + virtual RegExpText* AsText(); + virtual bool IsText(); + virtual bool IsTextElement() { return true; } + virtual int min_match() { return length_; } + virtual int max_match() { return length_; } + virtual void AppendToText(RegExpText* text); + + void AddElement(TextElement elm) { + elements_.append(elm); + length_ += elm.length(); + } + const TextElementVector& elements() { return elements_; } + + private: + TextElementVector elements_; + int length_; +}; + +class RegExpQuantifier : public RegExpTree +{ + public: + enum QuantifierType { GREEDY, NON_GREEDY, POSSESSIVE }; + RegExpQuantifier(int min, int max, QuantifierType type, RegExpTree* body) + : body_(body), + min_(min), + max_(max), + min_match_(min * body->min_match()), + quantifier_type_(type) + { + if (max > 0 && body->max_match() > kInfinity / max) { + max_match_ = kInfinity; + } else { + max_match_ = max * body->max_match(); + } + } + + virtual void* Accept(RegExpVisitor* visitor, void* data); + virtual RegExpNode* ToNode(RegExpCompiler* compiler, + RegExpNode* on_success); + static RegExpNode* ToNode(int min, + int max, + bool is_greedy, + RegExpTree* body, + RegExpCompiler* compiler, + RegExpNode* on_success, + bool not_at_start = false); + virtual RegExpQuantifier* AsQuantifier(); + virtual Interval CaptureRegisters(); + virtual bool IsQuantifier(); + virtual int min_match() { return min_match_; } + virtual int max_match() { return max_match_; } + int min() { return min_; } + int max() { return max_; } + bool is_possessive() { return quantifier_type_ == POSSESSIVE; } + bool is_non_greedy() { return quantifier_type_ == NON_GREEDY; } + bool is_greedy() { return quantifier_type_ == GREEDY; } + RegExpTree* body() { return body_; } + + private: + RegExpTree* body_; + int min_; + int max_; + int min_match_; + int max_match_; + QuantifierType quantifier_type_; +}; + +class RegExpCapture : public RegExpTree +{ + public: + explicit RegExpCapture(RegExpTree* body, int index) + : body_(body), index_(index) + {} + + virtual void* Accept(RegExpVisitor* visitor, void* data); + virtual RegExpNode* ToNode(RegExpCompiler* compiler, + RegExpNode* on_success); + static RegExpNode* ToNode(RegExpTree* body, + int index, + RegExpCompiler* compiler, + RegExpNode* on_success); + virtual RegExpCapture* AsCapture(); + virtual bool IsAnchoredAtStart(); + virtual bool IsAnchoredAtEnd(); + virtual Interval CaptureRegisters(); + virtual bool IsCapture(); + virtual int min_match() { return body_->min_match(); } + virtual int max_match() { return body_->max_match(); } + RegExpTree* body() { return body_; } + int index() { return index_; } + static int StartRegister(int index) { return index * 2; } + static int EndRegister(int index) { return index * 2 + 1; } + + private: + RegExpTree* body_; + int index_; +}; + +class RegExpLookahead : public RegExpTree +{ + public: + RegExpLookahead(RegExpTree* body, + bool is_positive, + int capture_count, + int capture_from) + : body_(body), + is_positive_(is_positive), + capture_count_(capture_count), + capture_from_(capture_from) + {} + + virtual void* Accept(RegExpVisitor* visitor, void* data); + virtual RegExpNode* ToNode(RegExpCompiler* compiler, + RegExpNode* on_success); + virtual RegExpLookahead* AsLookahead(); + virtual Interval CaptureRegisters(); + virtual bool IsLookahead(); + virtual bool IsAnchoredAtStart(); + virtual int min_match() { return 0; } + virtual int max_match() { return 0; } + RegExpTree* body() { return body_; } + bool is_positive() { return is_positive_; } + int capture_count() { return capture_count_; } + int capture_from() { return capture_from_; } + + private: + RegExpTree* body_; + bool is_positive_; + int capture_count_; + int capture_from_; +}; + +typedef InfallibleVector RegExpCaptureVector; + +class RegExpBackReference : public RegExpTree +{ + public: + explicit RegExpBackReference(RegExpCapture* capture) + : capture_(capture) + {} + + virtual void* Accept(RegExpVisitor* visitor, void* data); + virtual RegExpNode* ToNode(RegExpCompiler* compiler, + RegExpNode* on_success); + virtual RegExpBackReference* AsBackReference(); + virtual bool IsBackReference(); + virtual int min_match() { return 0; } + virtual int max_match() { return capture_->max_match(); } + int index() { return capture_->index(); } + RegExpCapture* capture() { return capture_; } + private: + RegExpCapture* capture_; +}; + +class RegExpEmpty : public RegExpTree +{ + public: + RegExpEmpty() + {} + + virtual void* Accept(RegExpVisitor* visitor, void* data); + virtual RegExpNode* ToNode(RegExpCompiler* compiler, + RegExpNode* on_success); + virtual RegExpEmpty* AsEmpty(); + virtual bool IsEmpty(); + virtual int min_match() { return 0; } + virtual int max_match() { return 0; } + static RegExpEmpty* GetInstance() { + static RegExpEmpty instance; + return &instance; + } +}; + +} } // namespace js::irregexp + +#endif // V8_REGEXP_AST_H_ diff --git a/js/src/irregexp/RegExpBytecode.h b/js/src/irregexp/RegExpBytecode.h new file mode 100644 index 000000000..f31b78c59 --- /dev/null +++ b/js/src/irregexp/RegExpBytecode.h @@ -0,0 +1,108 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- + * vim: set ts=8 sts=4 et sw=4 tw=99: */ + +// Copyright 2011 the V8 project authors. All rights reserved. +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived +// from this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#ifndef V8_BYTECODES_IRREGEXP_H_ +#define V8_BYTECODES_IRREGEXP_H_ + +namespace js { +namespace irregexp { + +const int BYTECODE_MASK = 0xff; + +// The first argument is packed in with the byte code in one word, so it +// has 24 bits, but it can be positive and negative so only use 23 bits for +// positive values. +const unsigned int MAX_FIRST_ARG = 0x7fffffu; +const int BYTECODE_SHIFT = 8; + +#define BYTECODE_ITERATOR(V) \ +V(BREAK, 0, 4) /* bc8 */ \ +V(PUSH_CP, 1, 4) /* bc8 pad24 */ \ +V(PUSH_BT, 2, 8) /* bc8 pad24 offset32 */ \ +V(PUSH_REGISTER, 3, 4) /* bc8 reg_idx24 */ \ +V(SET_REGISTER_TO_CP, 4, 8) /* bc8 reg_idx24 offset32 */ \ +V(SET_CP_TO_REGISTER, 5, 4) /* bc8 reg_idx24 */ \ +V(SET_REGISTER_TO_SP, 6, 4) /* bc8 reg_idx24 */ \ +V(SET_SP_TO_REGISTER, 7, 4) /* bc8 reg_idx24 */ \ +V(SET_REGISTER, 8, 8) /* bc8 reg_idx24 value32 */ \ +V(ADVANCE_REGISTER, 9, 8) /* bc8 reg_idx24 value32 */ \ +V(POP_CP, 10, 4) /* bc8 pad24 */ \ +V(POP_BT, 11, 4) /* bc8 pad24 */ \ +V(POP_REGISTER, 12, 4) /* bc8 reg_idx24 */ \ +V(FAIL, 13, 4) /* bc8 pad24 */ \ +V(SUCCEED, 14, 4) /* bc8 pad24 */ \ +V(ADVANCE_CP, 15, 4) /* bc8 offset24 */ \ +V(GOTO, 16, 8) /* bc8 pad24 addr32 */ \ +V(LOAD_CURRENT_CHAR, 17, 8) /* bc8 offset24 addr32 */ \ +V(LOAD_CURRENT_CHAR_UNCHECKED, 18, 4) /* bc8 offset24 */ \ +V(LOAD_2_CURRENT_CHARS, 19, 8) /* bc8 offset24 addr32 */ \ +V(LOAD_2_CURRENT_CHARS_UNCHECKED, 20, 4) /* bc8 offset24 */ \ +V(LOAD_4_CURRENT_CHARS, 21, 8) /* bc8 offset24 addr32 */ \ +V(LOAD_4_CURRENT_CHARS_UNCHECKED, 22, 4) /* bc8 offset24 */ \ +V(CHECK_4_CHARS, 23, 12) /* bc8 pad24 uint32 addr32 */ \ +V(CHECK_CHAR, 24, 8) /* bc8 pad8 uint16 addr32 */ \ +V(CHECK_NOT_4_CHARS, 25, 12) /* bc8 pad24 uint32 addr32 */ \ +V(CHECK_NOT_CHAR, 26, 8) /* bc8 pad8 uint16 addr32 */ \ +V(AND_CHECK_4_CHARS, 27, 16) /* bc8 pad24 uint32 uint32 addr32 */ \ +V(AND_CHECK_CHAR, 28, 12) /* bc8 pad8 uint16 uint32 addr32 */ \ +V(AND_CHECK_NOT_4_CHARS, 29, 16) /* bc8 pad24 uint32 uint32 addr32 */ \ +V(AND_CHECK_NOT_CHAR, 30, 12) /* bc8 pad8 uint16 uint32 addr32 */ \ +V(MINUS_AND_CHECK_NOT_CHAR, 31, 12) /* bc8 pad8 uc16 uc16 uc16 addr32 */ \ +V(CHECK_CHAR_IN_RANGE, 32, 12) /* bc8 pad24 uc16 uc16 addr32 */ \ +V(CHECK_CHAR_NOT_IN_RANGE, 33, 12) /* bc8 pad24 uc16 uc16 addr32 */ \ +V(CHECK_BIT_IN_TABLE, 34, 24) /* bc8 pad24 addr32 bits128 */ \ +V(CHECK_LT, 35, 8) /* bc8 pad8 uc16 addr32 */ \ +V(CHECK_GT, 36, 8) /* bc8 pad8 uc16 addr32 */ \ +V(CHECK_NOT_BACK_REF, 37, 8) /* bc8 reg_idx24 addr32 */ \ +V(CHECK_NOT_BACK_REF_NO_CASE, 38, 8) /* bc8 reg_idx24 addr32 */ \ +V(CHECK_NOT_REGS_EQUAL, 39, 12) /* bc8 regidx24 reg_idx32 addr32 */ \ +V(CHECK_REGISTER_LT, 40, 12) /* bc8 reg_idx24 value32 addr32 */ \ +V(CHECK_REGISTER_GE, 41, 12) /* bc8 reg_idx24 value32 addr32 */ \ +V(CHECK_REGISTER_EQ_POS, 42, 8) /* bc8 reg_idx24 addr32 */ \ +V(CHECK_AT_START, 43, 8) /* bc8 pad24 addr32 */ \ +V(CHECK_NOT_AT_START, 44, 8) /* bc8 pad24 addr32 */ \ +V(CHECK_GREEDY, 45, 8) /* bc8 pad24 addr32 */ \ +V(ADVANCE_CP_AND_GOTO, 46, 8) /* bc8 offset24 addr32 */ \ +V(SET_CURRENT_POSITION_FROM_END, 47, 4) /* bc8 idx24 */ \ +V(CHECK_NOT_BACK_REF_NO_CASE_UNICODE, 48, 8) /* bc8 reg_idx24 addr32 */ + +#define DECLARE_BYTECODES(name, code, length) \ + static const int BC_##name = code; +BYTECODE_ITERATOR(DECLARE_BYTECODES) +#undef DECLARE_BYTECODES + +#define DECLARE_BYTECODE_LENGTH(name, code, length) \ + static const int BC_##name##_LENGTH = length; +BYTECODE_ITERATOR(DECLARE_BYTECODE_LENGTH) +#undef DECLARE_BYTECODE_LENGTH + +} } // namespace js::irregexp + +#endif // V8_BYTECODES_IRREGEXP_H_ diff --git a/js/src/irregexp/RegExpEngine.cpp b/js/src/irregexp/RegExpEngine.cpp new file mode 100644 index 000000000..2e19065fd --- /dev/null +++ b/js/src/irregexp/RegExpEngine.cpp @@ -0,0 +1,5135 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- + * vim: set ts=8 sts=4 et sw=4 tw=99: */ + +// Copyright 2012 the V8 project authors. All rights reserved. +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived +// from this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include "irregexp/RegExpEngine.h" + +#include "irregexp/NativeRegExpMacroAssembler.h" +#include "irregexp/RegExpMacroAssembler.h" +#include "jit/ExecutableAllocator.h" +#include "jit/JitCommon.h" + +using namespace js; +using namespace js::irregexp; + +using mozilla::ArrayLength; +using mozilla::DebugOnly; +using mozilla::Maybe; + +#define DEFINE_ACCEPT(Type) \ + void Type##Node::Accept(NodeVisitor* visitor) { \ + visitor->Visit##Type(this); \ + } +FOR_EACH_NODE_TYPE(DEFINE_ACCEPT) +#undef DEFINE_ACCEPT + +void LoopChoiceNode::Accept(NodeVisitor* visitor) { + visitor->VisitLoopChoice(this); +} + +static const int kMaxLookaheadForBoyerMoore = 8; + +RegExpNode::RegExpNode(LifoAlloc* alloc) + : replacement_(nullptr), trace_count_(0), alloc_(alloc) +{ + bm_info_[0] = bm_info_[1] = nullptr; +} + +// ------------------------------------------------------------------- +// CharacterRange + +// The '2' variant has inclusive from and exclusive to. +// This covers \s as defined in ECMA-262 5.1, 15.10.2.12, +// which include WhiteSpace (7.2) or LineTerminator (7.3) values. +static const int kSpaceRanges[] = { '\t', '\r' + 1, ' ', ' ' + 1, + 0x00A0, 0x00A1, 0x1680, 0x1681, 0x180E, 0x180F, 0x2000, 0x200B, + 0x2028, 0x202A, 0x202F, 0x2030, 0x205F, 0x2060, 0x3000, 0x3001, + 0xFEFF, 0xFF00, 0x10000 }; +static const int kSpaceRangeCount = ArrayLength(kSpaceRanges); + +static const int kSpaceAndSurrogateRanges[] = { '\t', '\r' + 1, ' ', ' ' + 1, + 0x00A0, 0x00A1, 0x1680, 0x1681, 0x180E, 0x180F, 0x2000, 0x200B, + 0x2028, 0x202A, 0x202F, 0x2030, 0x205F, 0x2060, 0x3000, 0x3001, + unicode::LeadSurrogateMin, unicode::TrailSurrogateMax + 1, + 0xFEFF, 0xFF00, 0x10000 }; +static const int kSpaceAndSurrogateRangeCount = ArrayLength(kSpaceAndSurrogateRanges); +static const int kWordRanges[] = { + '0', '9' + 1, 'A', 'Z' + 1, '_', '_' + 1, 'a', 'z' + 1, 0x10000 }; +static const int kWordRangeCount = ArrayLength(kWordRanges); +static const int kIgnoreCaseWordRanges[] = { + '0', '9' + 1, 'A', 'Z' + 1, '_', '_' + 1, 'a', 'z' + 1, + 0x017F, 0x017F + 1, 0x212A, 0x212A + 1, + 0x10000 }; +static const int kIgnoreCaseWordCount = ArrayLength(kIgnoreCaseWordRanges); +static const int kWordAndSurrogateRanges[] = { + '0', '9' + 1, 'A', 'Z' + 1, '_', '_' + 1, 'a', 'z' + 1, + unicode::LeadSurrogateMin, unicode::TrailSurrogateMax + 1, + 0x10000 }; +static const int kWordAndSurrogateRangeCount = ArrayLength(kWordAndSurrogateRanges); +static const int kNegatedIgnoreCaseWordAndSurrogateRanges[] = { + 0, '0', '9' + 1, 'A', + 'Z' + 1, '_', '_' + 1, 'a', + 'z' + 1, 0x017F, + 0x017F + 1, 0x212A, + 0x212A + 1, unicode::LeadSurrogateMin, + unicode::TrailSurrogateMax + 1, 0x10000, + 0x10000 }; +static const int kNegatedIgnoreCaseWordAndSurrogateRangeCount = + ArrayLength(kNegatedIgnoreCaseWordAndSurrogateRanges); +static const int kDigitRanges[] = { '0', '9' + 1, 0x10000 }; +static const int kDigitRangeCount = ArrayLength(kDigitRanges); +static const int kDigitAndSurrogateRanges[] = { + '0', '9' + 1, + unicode::LeadSurrogateMin, unicode::TrailSurrogateMax + 1, + 0x10000 }; +static const int kDigitAndSurrogateRangeCount = ArrayLength(kDigitAndSurrogateRanges); +static const int kSurrogateRanges[] = { + unicode::LeadSurrogateMin, unicode::TrailSurrogateMax + 1, + 0x10000 }; +static const int kSurrogateRangeCount = ArrayLength(kSurrogateRanges); +static const int kLineTerminatorRanges[] = { 0x000A, 0x000B, 0x000D, 0x000E, + 0x2028, 0x202A, 0x10000 }; +static const int kLineTerminatorRangeCount = ArrayLength(kLineTerminatorRanges); +static const int kMaxOneByteCharCode = 0xff; +static const int kMaxUtf16CodeUnit = 0xffff; + +static char16_t +MaximumCharacter(bool ascii) +{ + return ascii ? kMaxOneByteCharCode : kMaxUtf16CodeUnit; +} + +static void +AddClass(const int* elmv, int elmc, + CharacterRangeVector* ranges) +{ + elmc--; + MOZ_ASSERT(elmv[elmc] == 0x10000); + for (int i = 0; i < elmc; i += 2) { + MOZ_ASSERT(elmv[i] < elmv[i + 1]); + ranges->append(CharacterRange(elmv[i], elmv[i + 1] - 1)); + } +} + +static void +AddClassNegated(const int* elmv, + int elmc, + CharacterRangeVector* ranges) +{ + elmc--; + MOZ_ASSERT(elmv[elmc] == 0x10000); + MOZ_ASSERT(elmv[0] != 0x0000); + MOZ_ASSERT(elmv[elmc-1] != kMaxUtf16CodeUnit); + char16_t last = 0x0000; + for (int i = 0; i < elmc; i += 2) { + MOZ_ASSERT(last <= elmv[i] - 1); + MOZ_ASSERT(elmv[i] < elmv[i + 1]); + ranges->append(CharacterRange(last, elmv[i] - 1)); + last = elmv[i + 1]; + } + ranges->append(CharacterRange(last, kMaxUtf16CodeUnit)); +} + +void +CharacterRange::AddClassEscape(LifoAlloc* alloc, char16_t type, + CharacterRangeVector* ranges) +{ + switch (type) { + case 's': + AddClass(kSpaceRanges, kSpaceRangeCount, ranges); + break; + case 'S': + AddClassNegated(kSpaceRanges, kSpaceRangeCount, ranges); + break; + case 'w': + AddClass(kWordRanges, kWordRangeCount, ranges); + break; + case 'W': + AddClassNegated(kWordRanges, kWordRangeCount, ranges); + break; + case 'd': + AddClass(kDigitRanges, kDigitRangeCount, ranges); + break; + case 'D': + AddClassNegated(kDigitRanges, kDigitRangeCount, ranges); + break; + case '.': + AddClassNegated(kLineTerminatorRanges, kLineTerminatorRangeCount, ranges); + break; + // This is not a character range as defined by the spec but a + // convenient shorthand for a character class that matches any + // character. + case '*': + ranges->append(CharacterRange::Everything()); + break; + // This is the set of characters matched by the $ and ^ symbols + // in multiline mode. + case 'n': + AddClass(kLineTerminatorRanges, kLineTerminatorRangeCount, ranges); + break; + default: + MOZ_CRASH("Bad character class escape"); + } +} + +// Add class escape, excluding surrogate pair range. +void +CharacterRange::AddClassEscapeUnicode(LifoAlloc* alloc, char16_t type, + CharacterRangeVector* ranges, bool ignore_case) +{ + switch (type) { + case 's': + case 'd': + return AddClassEscape(alloc, type, ranges); + break; + case 'S': + AddClassNegated(kSpaceAndSurrogateRanges, kSpaceAndSurrogateRangeCount, ranges); + break; + case 'w': + if (ignore_case) + AddClass(kIgnoreCaseWordRanges, kIgnoreCaseWordCount, ranges); + else + AddClassEscape(alloc, type, ranges); + break; + case 'W': + if (ignore_case) { + AddClass(kNegatedIgnoreCaseWordAndSurrogateRanges, + kNegatedIgnoreCaseWordAndSurrogateRangeCount, ranges); + } else { + AddClassNegated(kWordAndSurrogateRanges, kWordAndSurrogateRangeCount, ranges); + } + break; + case 'D': + AddClassNegated(kDigitAndSurrogateRanges, kDigitAndSurrogateRangeCount, ranges); + break; + default: + MOZ_CRASH("Bad type!"); + } +} + +#define FOR_EACH_NON_ASCII_TO_ASCII_FOLDING(macro) \ + /* LATIN CAPITAL LETTER Y WITH DIAERESIS */ \ + macro(0x0178, 0x00FF) \ + /* LATIN SMALL LETTER LONG S */ \ + macro(0x017F, 0x0073) \ + /* LATIN CAPITAL LETTER SHARP S */ \ + macro(0x1E9E, 0x00DF) \ + /* KELVIN SIGN */ \ + macro(0x212A, 0x006B) \ + /* ANGSTROM SIGN */ \ + macro(0x212B, 0x00E5) + +// We need to check for the following characters: 0x39c 0x3bc 0x178. +static inline bool +RangeContainsLatin1Equivalents(CharacterRange range, bool unicode) +{ + /* TODO(dcarney): this could be a lot more efficient. */ + if (unicode) { +#define CHECK_RANGE(C, F) \ + if (range.Contains(C)) return true; +FOR_EACH_NON_ASCII_TO_ASCII_FOLDING(CHECK_RANGE) +#undef CHECK_RANGE + } + + return range.Contains(0x39c) || range.Contains(0x3bc) || range.Contains(0x178); +} + +static bool +RangesContainLatin1Equivalents(const CharacterRangeVector& ranges, bool unicode) +{ + for (size_t i = 0; i < ranges.length(); i++) { + // TODO(dcarney): this could be a lot more efficient. + if (RangeContainsLatin1Equivalents(ranges[i], unicode)) + return true; + } + return false; +} + +static const size_t kEcma262UnCanonicalizeMaxWidth = 4; + +// Returns the number of characters in the equivalence class, omitting those +// that cannot occur in the source string if it is a one byte string. +static int +GetCaseIndependentLetters(char16_t character, + bool ascii_subject, + bool unicode, + const char16_t* choices, + size_t choices_length, + char16_t* letters) +{ + size_t count = 0; + for (size_t i = 0; i < choices_length; i++) { + char16_t c = choices[i]; + + // Skip characters that can't appear in one byte strings. + if (!unicode && ascii_subject && c > kMaxOneByteCharCode) + continue; + + // Watch for duplicates. + bool found = false; + for (size_t j = 0; j < count; j++) { + if (letters[j] == c) { + found = true; + break; + } + } + if (found) + continue; + + letters[count++] = c; + } + + return count; +} + +static int +GetCaseIndependentLetters(char16_t character, + bool ascii_subject, + bool unicode, + char16_t* letters) +{ + if (unicode) { + const char16_t choices[] = { + character, + unicode::FoldCase(character), + unicode::ReverseFoldCase1(character), + unicode::ReverseFoldCase2(character), + unicode::ReverseFoldCase3(character), + }; + return GetCaseIndependentLetters(character, ascii_subject, unicode, + choices, ArrayLength(choices), letters); + } + + char16_t upper = unicode::ToUpperCase(character); + unicode::CodepointsWithSameUpperCase others(character); + char16_t other1 = others.other1(); + char16_t other2 = others.other2(); + char16_t other3 = others.other3(); + + // ES 2017 draft 996af87b7072b3c3dd2b1def856c66f456102215 21.2.4.2 + // step 3.g. + // The standard requires that non-ASCII characters cannot have ASCII + // character codes in their equivalence class, even though this + // situation occurs multiple times in the unicode tables. + static const unsigned kMaxAsciiCharCode = 127; + if (upper <= kMaxAsciiCharCode) { + if (character > kMaxAsciiCharCode) { + // If Canonicalize(character) == character, all other characters + // should be ignored. + return GetCaseIndependentLetters(character, ascii_subject, unicode, + &character, 1, letters); + } + + if (other1 > kMaxAsciiCharCode) + other1 = character; + if (other2 > kMaxAsciiCharCode) + other2 = character; + if (other3 > kMaxAsciiCharCode) + other3 = character; + } + + const char16_t choices[] = { + character, + upper, + other1, + other2, + other3 + }; + return GetCaseIndependentLetters(character, ascii_subject, unicode, + choices, ArrayLength(choices), letters); +} + +static char16_t +ConvertNonLatin1ToLatin1(char16_t c, bool unicode) +{ + MOZ_ASSERT(c > kMaxOneByteCharCode); + if (unicode) { + switch (c) { +#define CONVERT(C, F) case C: return F; +FOR_EACH_NON_ASCII_TO_ASCII_FOLDING(CONVERT) +#undef CONVERT + } + } + + switch (c) { + // This are equivalent characters in unicode. + case 0x39c: + case 0x3bc: + return 0xb5; + // This is an uppercase of a Latin-1 character + // outside of Latin-1. + case 0x178: + return 0xff; + } + return 0; +} + +void +CharacterRange::AddCaseEquivalents(bool is_ascii, bool unicode, CharacterRangeVector* ranges) +{ + char16_t bottom = from(); + char16_t top = to(); + + if (is_ascii && !RangeContainsLatin1Equivalents(*this, unicode)) { + if (bottom > kMaxOneByteCharCode) + return; + if (top > kMaxOneByteCharCode) + top = kMaxOneByteCharCode; + } + + for (char16_t c = bottom;; c++) { + char16_t chars[kEcma262UnCanonicalizeMaxWidth]; + size_t length = GetCaseIndependentLetters(c, is_ascii, unicode, chars); + + for (size_t i = 0; i < length; i++) { + char16_t other = chars[i]; + if (other == c) + continue; + + // Try to combine with an existing range. + bool found = false; + for (size_t i = 0; i < ranges->length(); i++) { + CharacterRange& range = (*ranges)[i]; + if (range.Contains(other)) { + found = true; + break; + } else if (other == range.from() - 1) { + range.set_from(other); + found = true; + break; + } else if (other == range.to() + 1) { + range.set_to(other); + found = true; + break; + } + } + + if (!found) + ranges->append(CharacterRange::Singleton(other)); + } + + if (c == top) + break; + } +} + +static bool +CompareInverseRanges(const CharacterRangeVector& ranges, const int* special_class, size_t length) +{ + length--; // Remove final 0x10000. + MOZ_ASSERT(special_class[length] == 0x10000); + MOZ_ASSERT(ranges.length() != 0); + MOZ_ASSERT(length != 0); + MOZ_ASSERT(special_class[0] != 0); + if (ranges.length() != (length >> 1) + 1) + return false; + CharacterRange range = ranges[0]; + if (range.from() != 0) + return false; + for (size_t i = 0; i < length; i += 2) { + if (special_class[i] != (range.to() + 1)) + return false; + range = ranges[(i >> 1) + 1]; + if (special_class[i+1] != range.from()) + return false; + } + if (range.to() != 0xffff) + return false; + return true; +} + +static bool +CompareRanges(const CharacterRangeVector& ranges, const int* special_class, size_t length) +{ + length--; // Remove final 0x10000. + MOZ_ASSERT(special_class[length] == 0x10000); + if (ranges.length() * 2 != length) + return false; + for (size_t i = 0; i < length; i += 2) { + CharacterRange range = ranges[i >> 1]; + if (range.from() != special_class[i] || range.to() != special_class[i + 1] - 1) + return false; + } + return true; +} + +bool +RegExpCharacterClass::is_standard(LifoAlloc* alloc) +{ + // TODO(lrn): Remove need for this function, by not throwing away information + // along the way. + if (is_negated_) + return false; + if (set_.is_standard()) + return true; + if (CompareRanges(set_.ranges(alloc), kSpaceRanges, kSpaceRangeCount)) { + set_.set_standard_set_type('s'); + return true; + } + if (CompareInverseRanges(set_.ranges(alloc), kSpaceRanges, kSpaceRangeCount)) { + set_.set_standard_set_type('S'); + return true; + } + if (CompareInverseRanges(set_.ranges(alloc), + kLineTerminatorRanges, + kLineTerminatorRangeCount)) { + set_.set_standard_set_type('.'); + return true; + } + if (CompareRanges(set_.ranges(alloc), + kLineTerminatorRanges, + kLineTerminatorRangeCount)) { + set_.set_standard_set_type('n'); + return true; + } + if (CompareRanges(set_.ranges(alloc), kWordRanges, kWordRangeCount)) { + set_.set_standard_set_type('w'); + return true; + } + if (CompareInverseRanges(set_.ranges(alloc), kWordRanges, kWordRangeCount)) { + set_.set_standard_set_type('W'); + return true; + } + return false; +} + +bool +CharacterRange::IsCanonical(const CharacterRangeVector& ranges) +{ + int n = ranges.length(); + if (n <= 1) + return true; + + int max = ranges[0].to(); + for (int i = 1; i < n; i++) { + CharacterRange next_range = ranges[i]; + if (next_range.from() <= max + 1) + return false; + max = next_range.to(); + } + return true; +} + +// Move a number of elements in a zonelist to another position +// in the same list. Handles overlapping source and target areas. +static +void MoveRanges(CharacterRangeVector& list, int from, int to, int count) +{ + // Ranges are potentially overlapping. + if (from < to) { + for (int i = count - 1; i >= 0; i--) + list[to + i] = list[from + i]; + } else { + for (int i = 0; i < count; i++) + list[to + i] = list[from + i]; + } +} + +static int +InsertRangeInCanonicalList(CharacterRangeVector& list, + int count, + CharacterRange insert) +{ + // Inserts a range into list[0..count[, which must be sorted + // by from value and non-overlapping and non-adjacent, using at most + // list[0..count] for the result. Returns the number of resulting + // canonicalized ranges. Inserting a range may collapse existing ranges into + // fewer ranges, so the return value can be anything in the range 1..count+1. + char16_t from = insert.from(); + char16_t to = insert.to(); + int start_pos = 0; + int end_pos = count; + for (int i = count - 1; i >= 0; i--) { + CharacterRange current = list[i]; + if (current.from() > to + 1) { + end_pos = i; + } else if (current.to() + 1 < from) { + start_pos = i + 1; + break; + } + } + + // Inserted range overlaps, or is adjacent to, ranges at positions + // [start_pos..end_pos[. Ranges before start_pos or at or after end_pos are + // not affected by the insertion. + // If start_pos == end_pos, the range must be inserted before start_pos. + // if start_pos < end_pos, the entire range from start_pos to end_pos + // must be merged with the insert range. + + if (start_pos == end_pos) { + // Insert between existing ranges at position start_pos. + if (start_pos < count) { + MoveRanges(list, start_pos, start_pos + 1, count - start_pos); + } + list[start_pos] = insert; + return count + 1; + } + if (start_pos + 1 == end_pos) { + // Replace single existing range at position start_pos. + CharacterRange to_replace = list[start_pos]; + int new_from = Min(to_replace.from(), from); + int new_to = Max(to_replace.to(), to); + list[start_pos] = CharacterRange(new_from, new_to); + return count; + } + // Replace a number of existing ranges from start_pos to end_pos - 1. + // Move the remaining ranges down. + + int new_from = Min(list[start_pos].from(), from); + int new_to = Max(list[end_pos - 1].to(), to); + if (end_pos < count) { + MoveRanges(list, end_pos, start_pos + 1, count - end_pos); + } + list[start_pos] = CharacterRange(new_from, new_to); + return count - (end_pos - start_pos) + 1; +} + +void +CharacterRange::Canonicalize(CharacterRangeVector& character_ranges) +{ + if (character_ranges.length() <= 1) return; + // Check whether ranges are already canonical (increasing, non-overlapping, + // non-adjacent). + int n = character_ranges.length(); + int max = character_ranges[0].to(); + int i = 1; + while (i < n) { + CharacterRange current = character_ranges[i]; + if (current.from() <= max + 1) { + break; + } + max = current.to(); + i++; + } + // Canonical until the i'th range. If that's all of them, we are done. + if (i == n) return; + + // The ranges at index i and forward are not canonicalized. Make them so by + // doing the equivalent of insertion sort (inserting each into the previous + // list, in order). + // Notice that inserting a range can reduce the number of ranges in the + // result due to combining of adjacent and overlapping ranges. + int read = i; // Range to insert. + size_t num_canonical = i; // Length of canonicalized part of list. + do { + num_canonical = InsertRangeInCanonicalList(character_ranges, + num_canonical, + character_ranges[read]); + read++; + } while (read < n); + + while (character_ranges.length() > num_canonical) + character_ranges.popBack(); + + MOZ_ASSERT(CharacterRange::IsCanonical(character_ranges)); +} + +// ------------------------------------------------------------------- +// SeqRegExpNode + +class VisitMarker +{ + public: + explicit VisitMarker(NodeInfo* info) + : info_(info) + { + MOZ_ASSERT(!info->visited); + info->visited = true; + } + ~VisitMarker() { + info_->visited = false; + } + private: + NodeInfo* info_; +}; + +bool +SeqRegExpNode::FillInBMInfo(int offset, + int budget, + BoyerMooreLookahead* bm, + bool not_at_start) +{ + if (!bm->CheckOverRecursed()) + return false; + if (!on_success_->FillInBMInfo(offset, budget - 1, bm, not_at_start)) + return false; + if (offset == 0) + set_bm_info(not_at_start, bm); + return true; +} + +RegExpNode* +SeqRegExpNode::FilterASCII(int depth, bool ignore_case, bool unicode) +{ + if (info()->replacement_calculated) + return replacement(); + + if (depth < 0) + return this; + + MOZ_ASSERT(!info()->visited); + VisitMarker marker(info()); + return FilterSuccessor(depth - 1, ignore_case, unicode); +} + +RegExpNode* +SeqRegExpNode::FilterSuccessor(int depth, bool ignore_case, bool unicode) +{ + RegExpNode* next = on_success_->FilterASCII(depth - 1, ignore_case, unicode); + if (next == nullptr) + return set_replacement(nullptr); + + on_success_ = next; + return set_replacement(this); +} + +// ------------------------------------------------------------------- +// ActionNode + +int +ActionNode::EatsAtLeast(int still_to_find, int budget, bool not_at_start) +{ + if (budget <= 0) + return 0; + if (action_type_ == POSITIVE_SUBMATCH_SUCCESS) + return 0; // Rewinds input! + return on_success()->EatsAtLeast(still_to_find, + budget - 1, + not_at_start); +} + +bool +ActionNode::FillInBMInfo(int offset, + int budget, + BoyerMooreLookahead* bm, + bool not_at_start) +{ + if (!bm->CheckOverRecursed()) + return false; + + if (action_type_ == BEGIN_SUBMATCH) { + bm->SetRest(offset); + } else if (action_type_ != POSITIVE_SUBMATCH_SUCCESS) { + if (!on_success()->FillInBMInfo(offset, budget - 1, bm, not_at_start)) + return false; + } + SaveBMInfo(bm, not_at_start, offset); + + return true; +} + +/* static */ ActionNode* +ActionNode::SetRegister(int reg, + int val, + RegExpNode* on_success) +{ + ActionNode* result = on_success->alloc()->newInfallible(SET_REGISTER, on_success); + result->data_.u_store_register.reg = reg; + result->data_.u_store_register.value = val; + return result; +} + +/* static */ ActionNode* +ActionNode::IncrementRegister(int reg, RegExpNode* on_success) +{ + ActionNode* result = on_success->alloc()->newInfallible(INCREMENT_REGISTER, on_success); + result->data_.u_increment_register.reg = reg; + return result; +} + +/* static */ ActionNode* +ActionNode::StorePosition(int reg, bool is_capture, RegExpNode* on_success) +{ + ActionNode* result = on_success->alloc()->newInfallible(STORE_POSITION, on_success); + result->data_.u_position_register.reg = reg; + result->data_.u_position_register.is_capture = is_capture; + return result; +} + +/* static */ ActionNode* +ActionNode::ClearCaptures(Interval range, RegExpNode* on_success) +{ + ActionNode* result = on_success->alloc()->newInfallible(CLEAR_CAPTURES, on_success); + result->data_.u_clear_captures.range_from = range.from(); + result->data_.u_clear_captures.range_to = range.to(); + return result; +} + +/* static */ ActionNode* +ActionNode::BeginSubmatch(int stack_pointer_reg, int position_reg, RegExpNode* on_success) +{ + ActionNode* result = on_success->alloc()->newInfallible(BEGIN_SUBMATCH, on_success); + result->data_.u_submatch.stack_pointer_register = stack_pointer_reg; + result->data_.u_submatch.current_position_register = position_reg; + return result; +} + +/* static */ ActionNode* +ActionNode::PositiveSubmatchSuccess(int stack_pointer_reg, + int restore_reg, + int clear_capture_count, + int clear_capture_from, + RegExpNode* on_success) +{ + ActionNode* result = on_success->alloc()->newInfallible(POSITIVE_SUBMATCH_SUCCESS, on_success); + result->data_.u_submatch.stack_pointer_register = stack_pointer_reg; + result->data_.u_submatch.current_position_register = restore_reg; + result->data_.u_submatch.clear_register_count = clear_capture_count; + result->data_.u_submatch.clear_register_from = clear_capture_from; + return result; +} + +/* static */ ActionNode* +ActionNode::EmptyMatchCheck(int start_register, + int repetition_register, + int repetition_limit, + RegExpNode* on_success) +{ + ActionNode* result = on_success->alloc()->newInfallible(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; +} + +// ------------------------------------------------------------------- +// TextNode + +int +TextNode::EatsAtLeast(int still_to_find, int budget, bool not_at_start) +{ + int answer = Length(); + if (answer >= still_to_find) + return answer; + if (budget <= 0) + return answer; + + // We are not at start after this node so we set the last argument to 'true'. + return answer + on_success()->EatsAtLeast(still_to_find - answer, + budget - 1, + true); +} + +int +TextNode::GreedyLoopTextLength() +{ + TextElement elm = elements()[elements().length() - 1]; + return elm.cp_offset() + elm.length(); +} + +RegExpNode* +TextNode::FilterASCII(int depth, bool ignore_case, bool unicode) +{ + if (info()->replacement_calculated) + return replacement(); + + if (depth < 0) + return this; + + MOZ_ASSERT(!info()->visited); + VisitMarker marker(info()); + int element_count = elements().length(); + for (int i = 0; i < element_count; i++) { + TextElement elm = elements()[i]; + if (elm.text_type() == TextElement::ATOM) { + CharacterVector& quarks = const_cast(elm.atom()->data()); + for (size_t j = 0; j < quarks.length(); j++) { + uint16_t c = quarks[j]; + if (c <= kMaxOneByteCharCode) + continue; + if (!ignore_case) + return set_replacement(nullptr); + + // Here, we need to check for characters whose upper and lower cases + // are outside the Latin-1 range. + char16_t converted = ConvertNonLatin1ToLatin1(c, unicode); + if (converted == 0) { + // Character is outside Latin-1 completely + return set_replacement(nullptr); + } + + // Convert quark to Latin-1 in place. + quarks[j] = converted; + } + } else { + MOZ_ASSERT(elm.text_type() == TextElement::CHAR_CLASS); + RegExpCharacterClass* cc = elm.char_class(); + + CharacterRangeVector& ranges = cc->ranges(alloc()); + if (!CharacterRange::IsCanonical(ranges)) + 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[0].from() == 0 && + ranges[0].to() >= kMaxOneByteCharCode) + { + // This will be handled in a later filter. + if (ignore_case && RangesContainLatin1Equivalents(ranges, unicode)) + continue; + return set_replacement(nullptr); + } + } else { + if (range_count == 0 || + ranges[0].from() > kMaxOneByteCharCode) + { + // This will be handled in a later filter. + if (ignore_case && RangesContainLatin1Equivalents(ranges, unicode)) + continue; + return set_replacement(nullptr); + } + } + } + } + return FilterSuccessor(depth - 1, ignore_case, unicode); +} + +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()[i]; + elm.set_cp_offset(cp_offset); + cp_offset += elm.length(); + } +} + +void TextNode::MakeCaseIndependent(bool is_ascii, bool unicode) +{ + int element_count = elements().length(); + for (int i = 0; i < element_count; i++) { + TextElement elm = elements()[i]; + if (elm.text_type() == TextElement::CHAR_CLASS) { + RegExpCharacterClass* cc = elm.char_class(); + + // 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(alloc())) + continue; + + CharacterRangeVector& ranges = cc->ranges(alloc()); + int range_count = ranges.length(); + for (int j = 0; j < range_count; j++) + ranges[j].AddCaseEquivalents(is_ascii, unicode, &ranges); + } + } +} + +// ------------------------------------------------------------------- +// AssertionNode + +int +AssertionNode::EatsAtLeast(int still_to_find, int budget, bool not_at_start) +{ + if (budget <= 0) + return 0; + + // 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 lets just return the max answer (still_to_find) since + // that won't prevent us from preloading a lot of characters for the other + // branches in the node graph. + if (assertion_type() == AT_START && not_at_start) + return still_to_find; + + return on_success()->EatsAtLeast(still_to_find, budget - 1, not_at_start); +} + +bool +AssertionNode::FillInBMInfo(int offset, int budget, BoyerMooreLookahead* bm, bool not_at_start) +{ + if (!bm->CheckOverRecursed()) + return false; + + // Match the behaviour of EatsAtLeast on this node. + if (assertion_type() == AT_START && not_at_start) + return true; + + if (!on_success()->FillInBMInfo(offset, budget - 1, bm, not_at_start)) + return false; + SaveBMInfo(bm, not_at_start, offset); + return true; +} + +// ------------------------------------------------------------------- +// BackReferenceNode + +int +BackReferenceNode::EatsAtLeast(int still_to_find, int budget, bool not_at_start) +{ + if (budget <= 0) + return 0; + return on_success()->EatsAtLeast(still_to_find, budget - 1, not_at_start); +} + +bool +BackReferenceNode::FillInBMInfo(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); + return true; +} + +// ------------------------------------------------------------------- +// ChoiceNode + +int +ChoiceNode::EatsAtLeastHelper(int still_to_find, + int budget, + RegExpNode* ignore_this_node, + bool not_at_start) +{ + if (budget <= 0) + return 0; + + int min = 100; + size_t choice_count = alternatives().length(); + budget = (budget - 1) / choice_count; + for (size_t i = 0; i < choice_count; i++) { + RegExpNode* node = alternatives()[i].node(); + if (node == ignore_this_node) continue; + int node_eats_at_least = + node->EatsAtLeast(still_to_find, budget, not_at_start); + if (node_eats_at_least < min) + min = node_eats_at_least; + if (min == 0) + return 0; + } + return min; +} + +int +ChoiceNode::EatsAtLeast(int still_to_find, int budget, bool not_at_start) +{ + return EatsAtLeastHelper(still_to_find, + budget, + nullptr, + not_at_start); +} + +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(); + MOZ_ASSERT(choice_count > 0); + alternatives()[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()[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); + } +} + +bool +ChoiceNode::FillInBMInfo(int offset, + int budget, + BoyerMooreLookahead* bm, + bool not_at_start) +{ + if (!bm->CheckOverRecursed()) + return false; + + const GuardedAlternativeVector& alts = alternatives(); + budget = (budget - 1) / alts.length(); + for (size_t i = 0; i < alts.length(); i++) { + const GuardedAlternative& alt = alts[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 true; + } + if (!alt.node()->FillInBMInfo(offset, budget, bm, not_at_start)) + return false; + } + SaveBMInfo(bm, not_at_start, offset); + return true; +} + +RegExpNode* +ChoiceNode::FilterASCII(int depth, bool ignore_case, bool unicode) +{ + 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++) { + const GuardedAlternative alternative = alternatives()[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()[i]; + RegExpNode* replacement = + alternative.node()->FilterASCII(depth - 1, ignore_case, unicode); + MOZ_ASSERT(replacement != this); // No missing EMPTY_MATCH_CHECK. + if (replacement != nullptr) { + alternatives()[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. + GuardedAlternativeVector new_alternatives(*alloc()); + new_alternatives.reserve(surviving); + for (int i = 0; i < choice_count; i++) { + RegExpNode* replacement = + alternatives()[i].node()->FilterASCII(depth - 1, ignore_case, unicode); + if (replacement != nullptr) { + alternatives()[i].set_node(replacement); + new_alternatives.append(alternatives()[i]); + } + } + + alternatives_ = Move(new_alternatives); + return this; +} + +// ------------------------------------------------------------------- +// NegativeLookaheadChoiceNode + +bool +NegativeLookaheadChoiceNode::FillInBMInfo(int offset, + int budget, + BoyerMooreLookahead* bm, + bool not_at_start) +{ + if (!bm->CheckOverRecursed()) + return false; + + if (!alternatives()[1].node()->FillInBMInfo(offset, budget - 1, bm, not_at_start)) + return false; + if (offset == 0) + set_bm_info(not_at_start, bm); + return true; +} + +int +NegativeLookaheadChoiceNode::EatsAtLeast(int still_to_find, int budget, bool not_at_start) +{ + if (budget <= 0) + return 0; + + // Alternative 0 is the negative lookahead, alternative 1 is what comes + // afterwards. + RegExpNode* node = alternatives()[1].node(); + return node->EatsAtLeast(still_to_find, budget - 1, not_at_start); +} + +void +NegativeLookaheadChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details, + RegExpCompiler* compiler, + int filled_in, + bool not_at_start) +{ + // Alternative 0 is the negative lookahead, alternative 1 is what comes + // afterwards. + RegExpNode* node = alternatives()[1].node(); + return node->GetQuickCheckDetails(details, compiler, filled_in, not_at_start); +} + +RegExpNode* +NegativeLookaheadChoiceNode::FilterASCII(int depth, bool ignore_case, bool unicode) +{ + 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. + RegExpNode* node = alternatives()[1].node(); + RegExpNode* replacement = node->FilterASCII(depth - 1, ignore_case, unicode); + + if (replacement == nullptr) + return set_replacement(nullptr); + alternatives()[1].set_node(replacement); + + RegExpNode* neg_node = alternatives()[0].node(); + RegExpNode* neg_replacement = neg_node->FilterASCII(depth - 1, ignore_case, unicode); + + // 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); + + alternatives()[0].set_node(neg_replacement); + return set_replacement(this); +} + +// ------------------------------------------------------------------- +// LoopChoiceNode + +void +GuardedAlternative::AddGuard(LifoAlloc* alloc, Guard* guard) +{ + if (guards_ == nullptr) + guards_ = alloc->newInfallible(*alloc); + guards_->append(guard); +} + +void +LoopChoiceNode::AddLoopAlternative(GuardedAlternative alt) +{ + MOZ_ASSERT(loop_node_ == nullptr); + AddAlternative(alt); + loop_node_ = alt.node(); +} + + +void +LoopChoiceNode::AddContinueAlternative(GuardedAlternative alt) +{ + MOZ_ASSERT(continue_node_ == nullptr); + AddAlternative(alt); + continue_node_ = alt.node(); +} + +int +LoopChoiceNode::EatsAtLeast(int still_to_find, int budget, bool not_at_start) +{ + return EatsAtLeastHelper(still_to_find, + budget - 1, + loop_node_, + not_at_start); +} + +void +LoopChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details, + RegExpCompiler* compiler, + int characters_filled_in, + bool not_at_start) +{ + if (body_can_be_zero_length_ || info()->visited) + return; + VisitMarker marker(info()); + return ChoiceNode::GetQuickCheckDetails(details, + compiler, + characters_filled_in, + not_at_start); +} + +bool +LoopChoiceNode::FillInBMInfo(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 true; + } + if (!ChoiceNode::FillInBMInfo(offset, budget - 1, bm, not_at_start)) + return false; + SaveBMInfo(bm, not_at_start, offset); + return true; +} + +RegExpNode* +LoopChoiceNode::FilterASCII(int depth, bool ignore_case, bool unicode) +{ + if (info()->replacement_calculated) + return replacement(); + if (depth < 0) + return this; + if (info()->visited) + return this; + + { + VisitMarker marker(info()); + + RegExpNode* continue_replacement = + continue_node_->FilterASCII(depth - 1, ignore_case, unicode); + + // 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::FilterASCII(depth - 1, ignore_case, unicode); +} + +// ------------------------------------------------------------------- +// Analysis + +void +Analysis::EnsureAnalyzed(RegExpNode* that) +{ + JS_CHECK_RECURSION(cx, failASCII("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; +} + +void +Analysis::VisitEnd(EndNode* that) +{ + // nothing to do +} + +void +Analysis::VisitText(TextNode* that) +{ + if (ignore_case_) + that->MakeCaseIndependent(is_ascii_, unicode_); + EnsureAnalyzed(that->on_success()); + if (!has_failed()) { + that->CalculateOffsets(); + } +} + +void +Analysis::VisitAction(ActionNode* that) +{ + RegExpNode* target = that->on_success(); + EnsureAnalyzed(target); + + if (!has_failed()) { + // 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. + that->info()->AddFromFollowing(target->info()); + } +} + +void +Analysis::VisitChoice(ChoiceNode* that) +{ + NodeInfo* info = that->info(); + + for (size_t i = 0; i < that->alternatives().length(); i++) { + RegExpNode* node = that->alternatives()[i].node(); + EnsureAnalyzed(node); + if (has_failed()) return; + + // Anything the following nodes need to know has to be known by + // this node also, so it can pass it on. + info->AddFromFollowing(node->info()); + } +} + +void +Analysis::VisitLoopChoice(LoopChoiceNode* that) +{ + NodeInfo* info = that->info(); + for (size_t i = 0; i < that->alternatives().length(); i++) { + RegExpNode* node = that->alternatives()[i].node(); + if (node != that->loop_node()) { + EnsureAnalyzed(node); + if (has_failed()) return; + info->AddFromFollowing(node->info()); + } + } + + // 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()) + info->AddFromFollowing(that->loop_node()->info()); +} + +void +Analysis::VisitBackReference(BackReferenceNode* that) +{ + EnsureAnalyzed(that->on_success()); +} + +void +Analysis::VisitAssertion(AssertionNode* that) +{ + EnsureAnalyzed(that->on_success()); +} + +// ------------------------------------------------------------------- +// 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 RegExpAST.cpp. +// 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 +// +// backtrack code location: +// pop current position +// } +// +// +// * Actions nodes are generated as follows +// +// +// push backtrack code location +// +// backtrack code location: +// +// +// +// * Matching nodes are generated as follows: +// if input string matches at current position +// update current position +// +// else +// +// +// 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. + +/* static */ TextElement +TextElement::Atom(RegExpAtom* atom) +{ + return TextElement(ATOM, atom); +} + +/* static */ 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; + } + MOZ_CRASH("Bad text type"); +} + +class FrequencyCollator +{ + public: + FrequencyCollator() : total_samples_(0) { + for (int i = 0; i < RegExpMacroAssembler::kTableSize; i++) { + frequencies_[i] = CharacterFrequency(i); + } + } + + void CountCharacter(int character) { + int index = (character & RegExpMacroAssembler::kTableMask); + frequencies_[index].Increment(); + total_samples_++; + } + + // Does not measure in percent, but rather per-128 (the table size from the + // regexp macro assembler). + int Frequency(int in_character) { + MOZ_ASSERT((in_character & RegExpMacroAssembler::kTableMask) == in_character); + if (total_samples_ < 1) return 1; // Division by zero. + int freq_in_per128 = + (frequencies_[in_character].counter() * 128) / total_samples_; + return freq_in_per128; + } + + private: + class CharacterFrequency { + public: + CharacterFrequency() : counter_(0), character_(-1) { } + explicit CharacterFrequency(int character) + : counter_(0), character_(character) + {} + + void Increment() { counter_++; } + int counter() { return counter_; } + int character() { return character_; } + + private: + int counter_; + int character_; + }; + + private: + CharacterFrequency frequencies_[RegExpMacroAssembler::kTableSize]; + int total_samples_; +}; + +class irregexp::RegExpCompiler +{ + public: + RegExpCompiler(JSContext* cx, LifoAlloc* alloc, int capture_count, + bool ignore_case, bool is_ascii, bool match_only, bool unicode); + + int AllocateRegister() { + if (next_register_ >= RegExpMacroAssembler::kMaxRegister) { + reg_exp_too_big_ = true; + return next_register_; + } + return next_register_++; + } + + RegExpCode Assemble(JSContext* cx, + RegExpMacroAssembler* assembler, + RegExpNode* start, + int capture_count); + + inline void AddWork(RegExpNode* node) { + AutoEnterOOMUnsafeRegion oomUnsafe; + if (!work_list_.append(node)) + oomUnsafe.crash("AddWork"); + } + + static const int kImplementationOffset = 0; + static const int kNumberOfRegistersOffset = 0; + static const int kCodeOffset = 1; + + RegExpMacroAssembler* macro_assembler() { return macro_assembler_; } + EndNode* accept() { return accept_; } + + static const int kMaxRecursion = 100; + inline int recursion_depth() { return recursion_depth_; } + inline void IncrementRecursionDepth() { recursion_depth_++; } + inline void DecrementRecursionDepth() { recursion_depth_--; } + + void SetRegExpTooBig() { reg_exp_too_big_ = true; } + + inline bool ignore_case() { return ignore_case_; } + inline bool ascii() { return ascii_; } + inline bool unicode() { return unicode_; } + FrequencyCollator* frequency_collator() { return &frequency_collator_; } + + int current_expansion_factor() { return current_expansion_factor_; } + void set_current_expansion_factor(int value) { + current_expansion_factor_ = value; + } + + JSContext* cx() const { return cx_; } + LifoAlloc* alloc() const { return alloc_; } + + static const int kNoRegister = -1; + + private: + EndNode* accept_; + int next_register_; + Vector work_list_; + int recursion_depth_; + RegExpMacroAssembler* macro_assembler_; + bool ignore_case_; + bool ascii_; + bool match_only_; + bool unicode_; + bool reg_exp_too_big_; + int current_expansion_factor_; + FrequencyCollator frequency_collator_; + JSContext* cx_; + LifoAlloc* alloc_; +}; + +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(JSContext* cx, LifoAlloc* alloc, int capture_count, + bool ignore_case, bool ascii, bool match_only, bool unicode) + : next_register_(2 * (capture_count + 1)), + recursion_depth_(0), + ignore_case_(ignore_case), + ascii_(ascii), + match_only_(match_only), + unicode_(unicode), + reg_exp_too_big_(false), + current_expansion_factor_(1), + frequency_collator_(), + cx_(cx), + alloc_(alloc) +{ + accept_ = alloc->newInfallible(alloc, EndNode::ACCEPT); + MOZ_ASSERT(next_register_ - 1 <= RegExpMacroAssembler::kMaxRegister); +} + +RegExpCode +RegExpCompiler::Assemble(JSContext* cx, + RegExpMacroAssembler* assembler, + RegExpNode* start, + int capture_count) +{ + macro_assembler_ = assembler; + macro_assembler_->set_slow_safe(false); + + // The LifoAlloc used by the regexp compiler is infallible and is currently + // expected to crash on OOM. Thus we have to disable the assertions made to + // prevent us from allocating any new chunk in the LifoAlloc. This is needed + // because the jit::MacroAssembler turns these assertions on by default. + LifoAlloc::AutoFallibleScope fallibleAllocator(alloc()); + + jit::Label fail; + macro_assembler_->PushBacktrack(&fail); + Trace new_trace; + start->Emit(this, &new_trace); + macro_assembler_->BindBacktrack(&fail); + macro_assembler_->Fail(); + + while (!work_list_.empty()) + work_list_.popCopy()->Emit(this, &new_trace); + + RegExpCode code = macro_assembler_->GenerateCode(cx, match_only_); + if (code.empty()) + return RegExpCode(); + + if (reg_exp_too_big_) { + code.destroy(); + JS_ReportErrorASCII(cx, "regexp too big"); + return RegExpCode(); + } + + return code; +} + +template +static void +SampleChars(FrequencyCollator* collator, const CharT* chars, size_t length) +{ + // Sample some characters from the middle of the string. + static const int kSampleSize = 128; + + int chars_sampled = 0; + int half_way = (int(length) - kSampleSize) / 2; + for (size_t i = Max(0, half_way); + i < length && chars_sampled < kSampleSize; + i++, chars_sampled++) + { + collator->CountCharacter(chars[i]); + } +} + +static bool +IsNativeRegExpEnabled(JSContext* cx) +{ +#ifdef JS_CODEGEN_NONE + return false; +#else + return cx->options().nativeRegExp(); +#endif +} + +RegExpCode +irregexp::CompilePattern(JSContext* cx, RegExpShared* shared, RegExpCompileData* data, + HandleLinearString sample, bool is_global, bool ignore_case, + bool is_ascii, bool match_only, bool force_bytecode, bool sticky, + bool unicode) +{ + if ((data->capture_count + 1) * 2 - 1 > RegExpMacroAssembler::kMaxRegister) { + JS_ReportErrorASCII(cx, "regexp too big"); + return RegExpCode(); + } + + LifoAlloc& alloc = cx->tempLifoAlloc(); + RegExpCompiler compiler(cx, &alloc, data->capture_count, ignore_case, is_ascii, match_only, + unicode); + + // Sample some characters from the middle of the string. + if (sample->hasLatin1Chars()) { + JS::AutoCheckCannotGC nogc; + SampleChars(compiler.frequency_collator(), sample->latin1Chars(nogc), sample->length()); + } else { + JS::AutoCheckCannotGC nogc; + SampleChars(compiler.frequency_collator(), sample->twoByteChars(nogc), sample->length()); + } + + // Wrap the body of the regexp in capture #0. + RegExpNode* captured_body = RegExpCapture::ToNode(data->tree, + 0, + &compiler, + compiler.accept()); + RegExpNode* node = captured_body; + bool is_end_anchored = data->tree->IsAnchoredAtEnd(); + bool is_start_anchored = sticky || data->tree->IsAnchoredAtStart(); + int max_length = data->tree->max_match(); + if (!is_start_anchored) { + // Add a .*? at the beginning, outside the body capture, unless + // this expression is anchored at the beginning. + RegExpNode* loop_node = + RegExpQuantifier::ToNode(0, + RegExpTree::kInfinity, + false, + alloc.newInfallible('*'), + &compiler, + captured_body, + data->contains_anchor); + + if (data->contains_anchor) { + // Unroll loop once, to take care of the case that might start + // at the start of input. + ChoiceNode* first_step_node = alloc.newInfallible(&alloc, 2); + RegExpNode* char_class = + alloc.newInfallible(alloc.newInfallible('*'), loop_node); + first_step_node->AddAlternative(GuardedAlternative(captured_body)); + first_step_node->AddAlternative(GuardedAlternative(char_class)); + node = first_step_node; + } else { + node = loop_node; + } + } + if (is_ascii) { + node = node->FilterASCII(RegExpCompiler::kMaxRecursion, ignore_case, unicode); + // Do it again to propagate the new nodes to places where they were not + // put because they had not been calculated yet. + if (node != nullptr) { + node = node->FilterASCII(RegExpCompiler::kMaxRecursion, ignore_case, unicode); + } + } + + if (node == nullptr) + node = alloc.newInfallible(&alloc, EndNode::BACKTRACK); + + Analysis analysis(cx, ignore_case, is_ascii, unicode); + analysis.EnsureAnalyzed(node); + if (analysis.has_failed()) { + JS_ReportErrorASCII(cx, "%s", analysis.errorMessage()); + return RegExpCode(); + } + + Maybe ctx; + Maybe native_assembler; + Maybe interpreted_assembler; + + RegExpMacroAssembler* assembler; + if (IsNativeRegExpEnabled(cx) && + !force_bytecode && + jit::CanLikelyAllocateMoreExecutableMemory() && + shared->getSource()->length() < 32 * 1024) + { + NativeRegExpMacroAssembler::Mode mode = + is_ascii ? NativeRegExpMacroAssembler::ASCII + : NativeRegExpMacroAssembler::CHAR16; + + ctx.emplace(cx, (jit::TempAllocator*) nullptr); + native_assembler.emplace(&alloc, shared, cx->runtime(), mode, (data->capture_count + 1) * 2); + assembler = native_assembler.ptr(); + } else { + interpreted_assembler.emplace(&alloc, shared, (data->capture_count + 1) * 2); + assembler = interpreted_assembler.ptr(); + } + + // Inserted here, instead of in Assembler, because it depends on information + // in the AST that isn't replicated in the Node structure. + static const int kMaxBacksearchLimit = 1024; + if (is_end_anchored && + !is_start_anchored && + max_length < kMaxBacksearchLimit) { + assembler->SetCurrentPositionFromEnd(max_length); + } + + if (is_global) { + assembler->set_global_mode((data->tree->min_match() > 0) + ? RegExpMacroAssembler::GLOBAL_NO_ZERO_LENGTH_CHECK + : RegExpMacroAssembler::GLOBAL); + } + + return compiler.Assemble(cx, assembler, node, data->capture_count); +} + +template +RegExpRunStatus +irregexp::ExecuteCode(JSContext* cx, jit::JitCode* codeBlock, const CharT* chars, size_t start, + size_t length, MatchPairs* matches, size_t* endIndex) +{ + typedef void (*RegExpCodeSignature)(InputOutputData*); + + InputOutputData data(chars, chars + length, start, matches, endIndex); + + RegExpCodeSignature function = reinterpret_cast(codeBlock->raw()); + + { + JS::AutoSuppressGCAnalysis nogc; + CALL_GENERATED_1(function, &data); + } + + return (RegExpRunStatus) data.result; +} + +template RegExpRunStatus +irregexp::ExecuteCode(JSContext* cx, jit::JitCode* codeBlock, const Latin1Char* chars, size_t start, + size_t length, MatchPairs* matches, size_t* endIndex); + +template RegExpRunStatus +irregexp::ExecuteCode(JSContext* cx, jit::JitCode* codeBlock, const char16_t* chars, size_t start, + size_t length, MatchPairs* matches, size_t* endIndex); + +// ------------------------------------------------------------------- +// Tree to graph conversion + +RegExpNode* +RegExpAtom::ToNode(RegExpCompiler* compiler, RegExpNode* on_success) +{ + TextElementVector* elms = + compiler->alloc()->newInfallible(*compiler->alloc()); + elms->append(TextElement::Atom(this)); + return compiler->alloc()->newInfallible(elms, on_success); +} + +RegExpNode* +RegExpText::ToNode(RegExpCompiler* compiler, RegExpNode* on_success) +{ + return compiler->alloc()->newInfallible(&elements_, on_success); +} + +RegExpNode* +RegExpCharacterClass::ToNode(RegExpCompiler* compiler, RegExpNode* on_success) +{ + return compiler->alloc()->newInfallible(this, on_success); +} + +RegExpNode* +RegExpDisjunction::ToNode(RegExpCompiler* compiler, RegExpNode* on_success) +{ + const RegExpTreeVector& alternatives = this->alternatives(); + size_t length = alternatives.length(); + ChoiceNode* result = compiler->alloc()->newInfallible(compiler->alloc(), length); + for (size_t i = 0; i < length; i++) { + GuardedAlternative alternative(alternatives[i]->ToNode(compiler, on_success)); + result->AddAlternative(alternative); + } + return result; +} + +RegExpNode* +RegExpQuantifier::ToNode(RegExpCompiler* compiler, RegExpNode* on_success) +{ + return ToNode(min(), + max(), + is_greedy(), + body(), + compiler, + on_success); +} + +// Scoped object to keep track of how much we unroll quantifier loops in the +// regexp graph generator. +class RegExpExpansionLimiter +{ + public: + static const int kMaxExpansionFactor = 6; + RegExpExpansionLimiter(RegExpCompiler* compiler, int factor) + : compiler_(compiler), + saved_expansion_factor_(compiler->current_expansion_factor()), + ok_to_expand_(saved_expansion_factor_ <= kMaxExpansionFactor) + { + MOZ_ASSERT(factor > 0); + if (ok_to_expand_) { + if (factor > kMaxExpansionFactor) { + // Avoid integer overflow of the current expansion factor. + ok_to_expand_ = false; + compiler->set_current_expansion_factor(kMaxExpansionFactor + 1); + } else { + int new_factor = saved_expansion_factor_ * factor; + ok_to_expand_ = (new_factor <= kMaxExpansionFactor); + compiler->set_current_expansion_factor(new_factor); + } + } + } + + ~RegExpExpansionLimiter() { + compiler_->set_current_expansion_factor(saved_expansion_factor_); + } + + bool ok_to_expand() { return ok_to_expand_; } + + private: + RegExpCompiler* compiler_; + int saved_expansion_factor_; + bool ok_to_expand_; +}; + +/* static */ RegExpNode* +RegExpQuantifier::ToNode(int min, + int max, + bool is_greedy, + RegExpTree* body, + RegExpCompiler* compiler, + RegExpNode* on_success, + bool not_at_start /* = false */) +{ + // x{f, t} becomes this: + // + // (r++)<-. + // | ` + // | (x) + // v ^ + // (r=0)-->(?)---/ [if r < t] + // | + // [if r >= f] \----> ... + // + + // 15.10.2.5 RepeatMatcher algorithm. + // The parser has already eliminated the case where max is 0. In the case + // where max_match is zero the parser has removed the quantifier if min was + // > 0 and removed the atom if min was 0. See AddQuantifierToAtom. + + // If we know that we cannot match zero length then things are a little + // simpler since we don't need to make the special zero length match check + // from step 2.1. If the min and max are small we can unroll a little in + // this case. + static const int kMaxUnrolledMinMatches = 3; // Unroll (foo)+ and (foo){3,} + static const int kMaxUnrolledMaxMatches = 3; // Unroll (foo)? and (foo){x,3} + + if (max == 0) + return on_success; // This can happen due to recursion. + + bool body_can_be_empty = (body->min_match() == 0); + int body_start_reg = RegExpCompiler::kNoRegister; + Interval capture_registers = body->CaptureRegisters(); + bool needs_capture_clearing = !capture_registers.is_empty(); + LifoAlloc* alloc = compiler->alloc(); + + if (body_can_be_empty) { + body_start_reg = compiler->AllocateRegister(); + } else if (!needs_capture_clearing) { + // Only unroll if there are no captures and the body can't be + // empty. + { + RegExpExpansionLimiter limiter(compiler, min + ((max != min) ? 1 : 0)); + if (min > 0 && min <= kMaxUnrolledMinMatches && limiter.ok_to_expand()) { + int new_max = (max == kInfinity) ? max : max - min; + // Recurse once to get the loop or optional matches after the fixed + // ones. + RegExpNode* answer = ToNode(0, new_max, is_greedy, body, compiler, on_success, true); + // Unroll the forced matches from 0 to min. This can cause chains of + // TextNodes (which the parser does not generate). These should be + // combined if it turns out they hinder good code generation. + for (int i = 0; i < min; i++) + answer = body->ToNode(compiler, answer); + return answer; + } + } + if (max <= kMaxUnrolledMaxMatches && min == 0) { + MOZ_ASSERT(max > 0); // Due to the 'if' above. + RegExpExpansionLimiter limiter(compiler, max); + if (limiter.ok_to_expand()) { + // Unroll the optional matches up to max. + RegExpNode* answer = on_success; + for (int i = 0; i < max; i++) { + ChoiceNode* alternation = alloc->newInfallible(alloc, 2); + if (is_greedy) { + alternation->AddAlternative(GuardedAlternative(body->ToNode(compiler, answer))); + alternation->AddAlternative(GuardedAlternative(on_success)); + } else { + alternation->AddAlternative(GuardedAlternative(on_success)); + alternation->AddAlternative(GuardedAlternative(body->ToNode(compiler, answer))); + } + answer = alternation; + if (not_at_start) alternation->set_not_at_start(); + } + return answer; + } + } + } + bool has_min = min > 0; + bool has_max = max < RegExpTree::kInfinity; + bool needs_counter = has_min || has_max; + int reg_ctr = needs_counter + ? compiler->AllocateRegister() + : RegExpCompiler::kNoRegister; + LoopChoiceNode* center = alloc->newInfallible(alloc, body->min_match() == 0); + if (not_at_start) + center->set_not_at_start(); + RegExpNode* loop_return = needs_counter + ? static_cast(ActionNode::IncrementRegister(reg_ctr, center)) + : static_cast(center); + if (body_can_be_empty) { + // If the body can be empty we need to check if it was and then + // backtrack. + loop_return = ActionNode::EmptyMatchCheck(body_start_reg, + reg_ctr, + min, + loop_return); + } + RegExpNode* body_node = body->ToNode(compiler, loop_return); + if (body_can_be_empty) { + // If the body can be empty we need to store the start position + // so we can bail out if it was empty. + body_node = ActionNode::StorePosition(body_start_reg, false, body_node); + } + if (needs_capture_clearing) { + // Before entering the body of this loop we need to clear captures. + body_node = ActionNode::ClearCaptures(capture_registers, body_node); + } + GuardedAlternative body_alt(body_node); + if (has_max) { + Guard* body_guard = alloc->newInfallible(reg_ctr, Guard::LT, max); + body_alt.AddGuard(alloc, body_guard); + } + GuardedAlternative rest_alt(on_success); + if (has_min) { + Guard* rest_guard = alloc->newInfallible(reg_ctr, Guard::GEQ, min); + rest_alt.AddGuard(alloc, rest_guard); + } + if (is_greedy) { + center->AddLoopAlternative(body_alt); + center->AddContinueAlternative(rest_alt); + } else { + center->AddContinueAlternative(rest_alt); + center->AddLoopAlternative(body_alt); + } + if (needs_counter) + return ActionNode::SetRegister(reg_ctr, 0, center); + return center; +} + +RegExpNode* +RegExpAssertion::ToNode(RegExpCompiler* compiler, + RegExpNode* on_success) +{ + NodeInfo info; + LifoAlloc* alloc = compiler->alloc(); + + switch (assertion_type()) { + case START_OF_LINE: + return AssertionNode::AfterNewline(on_success); + case START_OF_INPUT: + return AssertionNode::AtStart(on_success); + case BOUNDARY: + return AssertionNode::AtBoundary(on_success); + case NON_BOUNDARY: + return AssertionNode::AtNonBoundary(on_success); + case END_OF_INPUT: + return AssertionNode::AtEnd(on_success); + case END_OF_LINE: { + // Compile $ in multiline regexps as an alternation with a positive + // lookahead in one side and an end-of-input on the other side. + // We need two registers for the lookahead. + int stack_pointer_register = compiler->AllocateRegister(); + int position_register = compiler->AllocateRegister(); + // The ChoiceNode to distinguish between a newline and end-of-input. + ChoiceNode* result = alloc->newInfallible(alloc, 2); + // Create a newline atom. + CharacterRangeVector* newline_ranges = alloc->newInfallible(*alloc); + CharacterRange::AddClassEscape(alloc, 'n', newline_ranges); + RegExpCharacterClass* newline_atom = alloc->newInfallible('n'); + TextNode* newline_matcher = + alloc->newInfallible(newline_atom, + ActionNode::PositiveSubmatchSuccess(stack_pointer_register, + position_register, + 0, // No captures inside. + -1, // Ignored if no captures. + on_success)); + // Create an end-of-input matcher. + RegExpNode* end_of_line = + ActionNode::BeginSubmatch(stack_pointer_register, position_register, newline_matcher); + + // Add the two alternatives to the ChoiceNode. + GuardedAlternative eol_alternative(end_of_line); + result->AddAlternative(eol_alternative); + GuardedAlternative end_alternative(AssertionNode::AtEnd(on_success)); + result->AddAlternative(end_alternative); + return result; + } + case NOT_AFTER_LEAD_SURROGATE: + return AssertionNode::NotAfterLeadSurrogate(on_success); + case NOT_IN_SURROGATE_PAIR: + return AssertionNode::NotInSurrogatePair(on_success); + default: + MOZ_CRASH("Bad assertion type"); + } + return on_success; +} + +RegExpNode* +RegExpBackReference::ToNode(RegExpCompiler* compiler, RegExpNode* on_success) +{ + return compiler->alloc()->newInfallible(RegExpCapture::StartRegister(index()), + RegExpCapture::EndRegister(index()), + on_success); +} + +RegExpNode* +RegExpEmpty::ToNode(RegExpCompiler* compiler, RegExpNode* on_success) +{ + return on_success; +} + +RegExpNode* +RegExpLookahead::ToNode(RegExpCompiler* compiler, RegExpNode* on_success) +{ + int stack_pointer_register = compiler->AllocateRegister(); + int position_register = compiler->AllocateRegister(); + + const int registers_per_capture = 2; + const int register_of_first_capture = 2; + int register_count = capture_count_ * registers_per_capture; + int register_start = + register_of_first_capture + capture_from_ * registers_per_capture; + + if (is_positive()) { + RegExpNode* bodyNode = + body()->ToNode(compiler, + ActionNode::PositiveSubmatchSuccess(stack_pointer_register, + position_register, + register_count, + register_start, + on_success)); + return ActionNode::BeginSubmatch(stack_pointer_register, + position_register, + bodyNode); + } + + // We use a ChoiceNode for a negative lookahead because it has most of + // the characteristics we need. It has the body of the lookahead as its + // first alternative and the expression after the lookahead of the second + // alternative. If the first alternative succeeds then the + // NegativeSubmatchSuccess will unwind the stack including everything the + // choice node set up and backtrack. If the first alternative fails then + // the second alternative is tried, which is exactly the desired result + // for a negative lookahead. The NegativeLookaheadChoiceNode is a special + // ChoiceNode that knows to ignore the first exit when calculating quick + // checks. + LifoAlloc* alloc = compiler->alloc(); + + RegExpNode* success = + alloc->newInfallible(alloc, + stack_pointer_register, + position_register, + register_count, + register_start); + GuardedAlternative body_alt(body()->ToNode(compiler, success)); + + ChoiceNode* choice_node = + alloc->newInfallible(alloc, body_alt, GuardedAlternative(on_success)); + + return ActionNode::BeginSubmatch(stack_pointer_register, + position_register, + choice_node); +} + +RegExpNode* +RegExpCapture::ToNode(RegExpCompiler* compiler, RegExpNode* on_success) +{ + return ToNode(body(), index(), compiler, on_success); +} + +/* static */ RegExpNode* +RegExpCapture::ToNode(RegExpTree* body, + int index, + RegExpCompiler* compiler, + RegExpNode* on_success) +{ + int start_reg = RegExpCapture::StartRegister(index); + int end_reg = RegExpCapture::EndRegister(index); + RegExpNode* store_end = ActionNode::StorePosition(end_reg, true, on_success); + RegExpNode* body_node = body->ToNode(compiler, store_end); + return ActionNode::StorePosition(start_reg, true, body_node); +} + +RegExpNode* +RegExpAlternative::ToNode(RegExpCompiler* compiler, RegExpNode* on_success) +{ + const RegExpTreeVector& children = nodes(); + RegExpNode* current = on_success; + for (int i = children.length() - 1; i >= 0; i--) + current = children[i]->ToNode(compiler, current); + return current; +} + +// ------------------------------------------------------------------- +// BoyerMooreLookahead + +ContainedInLattice +irregexp::AddRange(ContainedInLattice containment, + const int* ranges, + int ranges_length, + Interval new_range) +{ + MOZ_ASSERT((ranges_length & 1) == 1); + MOZ_ASSERT(ranges[ranges_length - 1] == kMaxUtf16CodeUnit + 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; +} + +void +BoyerMoorePositionInfo::Set(int character) +{ + SetInterval(Interval(character, character)); +} + +void +BoyerMoorePositionInfo::SetInterval(const Interval& interval) +{ + s_ = AddRange(s_, kSpaceRanges, kSpaceRangeCount, interval); + w_ = AddRange(w_, kWordRanges, kWordRangeCount, interval); + d_ = AddRange(d_, kDigitRanges, kDigitRangeCount, interval); + surrogate_ = + AddRange(surrogate_, kSurrogateRanges, kSurrogateRangeCount, interval); + if (interval.to() - interval.from() >= kMapSize - 1) { + if (map_count_ != kMapSize) { + map_count_ = kMapSize; + for (int i = 0; i < kMapSize; i++) + map_[i] = true; + } + return; + } + for (int i = interval.from(); i <= interval.to(); i++) { + int mod_character = (i & kMask); + if (!map_[mod_character]) { + map_count_++; + map_[mod_character] = true; + } + if (map_count_ == kMapSize) + return; + } +} + +void +BoyerMoorePositionInfo::SetAll() +{ + s_ = w_ = d_ = kLatticeUnknown; + if (map_count_ != kMapSize) { + map_count_ = kMapSize; + for (int i = 0; i < kMapSize; i++) + map_[i] = true; + } +} + +BoyerMooreLookahead::BoyerMooreLookahead(LifoAlloc* alloc, size_t length, RegExpCompiler* compiler) + : length_(length), compiler_(compiler), bitmaps_(*alloc) +{ + max_char_ = MaximumCharacter(compiler->ascii()); + + bitmaps_.reserve(length); + for (size_t i = 0; i < length; i++) + bitmaps_.append(alloc->newInfallible(alloc)); +} + +// 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; + bool union_map[kSize]; + for (int j = 0; j < kSize; j++) union_map[j] = false; + while (i < length_ && Count(i) <= max_number_of_chars) { + BoyerMoorePositionInfo* map = bitmaps_[i]; + for (int j = 0; j < kSize; j++) union_map[j] |= map->at(j); + i++; + } + int frequency = 0; + for (int j = 0; j < kSize; j++) { + if (union_map[j]) { + // 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; + } + } + // 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_->ascii() ? 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, + uint8_t* boolean_skip_table) +{ + const int kSize = RegExpMacroAssembler::kTableSize; + + const int kSkipArrayEntry = 0; + const int kDontSkipArrayEntry = 1; + + for (int i = 0; i < kSize; i++) + boolean_skip_table[i] = kSkipArrayEntry; + int skip = max_lookahead + 1 - min_lookahead; + + for (int i = max_lookahead; i >= min_lookahead; i--) { + BoyerMoorePositionInfo* map = bitmaps_[i]; + for (int j = 0; j < kSize; j++) { + if (map->at(j)) + boolean_skip_table[j] = kDontSkipArrayEntry; + } + } + + return skip; +} + +// See comment on the implementation of GetSkipTable. +bool +BoyerMooreLookahead::EmitSkipInstructions(RegExpMacroAssembler* masm) +{ + const int kSize = RegExpMacroAssembler::kTableSize; + + int min_lookahead = 0; + int max_lookahead = 0; + + if (!FindWorthwhileInterval(&min_lookahead, &max_lookahead)) + return false; + + bool found_single_character = false; + int single_character = 0; + for (int i = max_lookahead; i >= min_lookahead; i--) { + BoyerMoorePositionInfo* map = bitmaps_[i]; + if (map->map_count() > 1 || + (found_single_character && map->map_count() != 0)) { + found_single_character = false; + break; + } + for (int j = 0; j < kSize; j++) { + if (map->at(j)) { + found_single_character = true; + single_character = j; + break; + } + } + } + + 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 false; + } + + if (found_single_character) { + jit::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->JumpOrBacktrack(&again); + masm->Bind(&cont); + return true; + } + + uint8_t* boolean_skip_table; + { + AutoEnterOOMUnsafeRegion oomUnsafe; + boolean_skip_table = static_cast(js_malloc(kSize)); + if (!boolean_skip_table || !masm->shared->addTable(boolean_skip_table)) + oomUnsafe.crash("Table malloc"); + } + + int skip_distance = GetSkipTable(min_lookahead, max_lookahead, boolean_skip_table); + MOZ_ASSERT(skip_distance != 0); + + jit::Label cont, again; + masm->Bind(&again); + masm->LoadCurrentCharacter(max_lookahead, &cont, true); + masm->CheckBitInTable(boolean_skip_table, &cont); + masm->AdvanceCurrentPosition(skip_distance); + masm->JumpOrBacktrack(&again); + masm->Bind(&cont); + + return true; +} + +bool +BoyerMooreLookahead::CheckOverRecursed() +{ + JS_CHECK_RECURSION(compiler()->cx(), compiler()->SetRegExpTooBig(); return false); + return true; +} + +// ------------------------------------------------------------------- +// Trace + +bool Trace::DeferredAction::Mentions(int that) +{ + if (action_type() == ActionNode::CLEAR_CAPTURES) { + Interval range = static_cast(this)->range(); + return range.Contains(that); + } + 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) +{ + MOZ_ASSERT(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(action)->cp_offset(); + return true; + } + return false; + } + } + return false; +} + +int +Trace::FindAffectedRegisters(LifoAlloc* alloc, OutSet* affected_registers) +{ + 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(action)->range(); + for (int i = range.from(); i <= range.to(); i++) + affected_registers->Set(alloc, i); + if (range.to() > max_register) max_register = range.to(); + } else { + affected_registers->Set(alloc, action->reg()); + if (action->reg() > max_register) max_register = action->reg(); + } + } + return max_register; +} + +void +Trace::RestoreAffectedRegisters(RegExpMacroAssembler* assembler, + int max_register, + OutSet& registers_to_pop, + OutSet& registers_to_clear) +{ + 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); + } + } +} + +enum DeferredActionUndoType { + DEFER_IGNORE, + DEFER_RESTORE, + DEFER_CLEAR +}; + +void +Trace::PerformDeferredActions(LifoAlloc* alloc, + RegExpMacroAssembler* assembler, + int max_register, + OutSet& affected_registers, + OutSet* registers_to_pop, + OutSet* registers_to_clear) +{ + // 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). + DeferredActionUndoType undo_action = DEFER_IGNORE; + + int value = 0; + bool absolute = false; + bool clear = false; + int store_position = -1; + // 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()) { + case ActionNode::SET_REGISTER: { + Trace::DeferredSetRegister* psr = + static_cast(action); + if (!absolute) { + value += psr->value(); + absolute = true; + } + // SET_REGISTER is currently only used for newly introduced loop + // counters. They can have a significant previous value if they + // occour 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 = DEFER_RESTORE; + MOZ_ASSERT(store_position == -1); + MOZ_ASSERT(!clear); + break; + } + case ActionNode::INCREMENT_REGISTER: + if (!absolute) { + value++; + } + MOZ_ASSERT(store_position == -1); + MOZ_ASSERT(!clear); + undo_action = DEFER_RESTORE; + break; + case ActionNode::STORE_POSITION: { + Trace::DeferredCapture* pc = + static_cast(action); + if (!clear && store_position == -1) { + 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 = DEFER_IGNORE; + } else { + undo_action = pc->is_capture() ? DEFER_CLEAR : DEFER_RESTORE; + } + MOZ_ASSERT(!absolute); + MOZ_ASSERT(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 == -1) { + clear = true; + } + undo_action = DEFER_RESTORE; + MOZ_ASSERT(!absolute); + MOZ_ASSERT(value == 0); + break; + } + default: + MOZ_CRASH("Bad action"); + } + } + } + // Prepare for the undo-action (e.g., push if it's going to be popped). + if (undo_action == DEFER_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(alloc, reg); + } else if (undo_action == DEFER_CLEAR) { + registers_to_clear->Set(alloc, reg); + } + // Perform the chronologically last action (or accumulated increment) + // for the register. + if (store_position != -1) { + 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(); + + MOZ_ASSERT(!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. + OutSet affected_registers; + + 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(compiler->alloc(), &affected_registers); + OutSet registers_to_pop; + OutSet registers_to_clear; + PerformDeferredActions(compiler->alloc(), + assembler, + max_register, + affected_registers, + ®isters_to_pop, + ®isters_to_clear); + if (cp_offset_ != 0) + assembler->AdvanceCurrentPosition(cp_offset_); + + // Create a new trivial state and generate the node with that. + jit::Label undo; + assembler->PushBacktrack(&undo); + Trace new_state; + successor->Emit(compiler, &new_state); + + // On backtrack we need to restore state. + assembler->BindBacktrack(&undo); + RestoreAffectedRegisters(assembler, + max_register, + registers_to_pop, + registers_to_clear); + if (backtrack() == nullptr) { + assembler->Backtrack(); + } else { + assembler->PopCurrentPosition(); + assembler->JumpOrBacktrack(backtrack()); + } +} + +void +Trace::InvalidateCurrentCharacter() +{ + characters_preloaded_ = 0; +} + +void +Trace::AdvanceCurrentPositionInTrace(int by, RegExpCompiler* compiler) +{ + MOZ_ASSERT(by > 0); + // 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->ascii()); + 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 +OutSet::Set(LifoAlloc* alloc, unsigned value) +{ + if (value < kFirstLimit) { + first_ |= (1 << value); + } else { + if (remaining_ == nullptr) + remaining_ = alloc->newInfallible(*alloc); + + for (size_t i = 0; i < remaining().length(); i++) { + if (remaining()[i] == value) + return; + } + remaining().append(value); + } +} + +bool +OutSet::Get(unsigned value) +{ + if (value < kFirstLimit) + return (first_ & (1 << value)) != 0; + if (remaining_ == nullptr) + return false; + for (size_t i = 0; i < remaining().length(); i++) { + if (remaining()[i] == value) + return true; + } + return false; +} + +// ------------------------------------------------------------------- +// Graph emitting + +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()->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->ReadBacktrackStackPointerFromRegister(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()->bound()) { + assembler->Bind(label()); + } + switch (action_) { + case ACCEPT: + assembler->Succeed(); + return; + case BACKTRACK: + assembler->JumpOrBacktrack(trace->backtrack()); + return; + case NEGATIVE_SUBMATCH_SUCCESS: + // This case is handled in a different virtual method. + MOZ_CRASH("Bad action: NEGATIVE_SUBMATCH_SUCCESS"); + } + MOZ_CRASH("Bad action"); +} + +// Emit the code to check for a ^ in multiline mode (1-character lookbehind +// that matches newline or the start of input). +static void +EmitHat(RegExpCompiler* compiler, RegExpNode* on_success, Trace* trace) +{ + RegExpMacroAssembler* assembler = compiler->macro_assembler(); + + // We will be loading the previous character into the current character + // register. + Trace new_trace(*trace); + new_trace.InvalidateCurrentCharacter(); + + jit::Label ok; + if (new_trace.cp_offset() == 0) { + // The start of input counts as a newline in this context, so skip to + // ok if we are at the start. + assembler->CheckAtStart(&ok); + } + + // We already checked that we are not at the start of input so it must be + // OK to load the previous character. + assembler->LoadCurrentCharacter(new_trace.cp_offset() -1, new_trace.backtrack(), false); + + if (!assembler->CheckSpecialCharacterClass('n', new_trace.backtrack())) { + // Newline means \n, \r, 0x2028 or 0x2029. + if (!compiler->ascii()) + assembler->CheckCharacterAfterAnd(0x2028, 0xfffe, &ok); + assembler->CheckCharacter('\n', &ok); + assembler->CheckNotCharacter('\r', new_trace.backtrack()); + } + assembler->Bind(&ok); + on_success->Emit(compiler, &new_trace); +} + +// Assert that the next character cannot be a part of a surrogate pair. +static void +EmitNotAfterLeadSurrogate(RegExpCompiler* compiler, RegExpNode* on_success, Trace* trace) +{ + RegExpMacroAssembler* assembler = compiler->macro_assembler(); + + // We will be loading the previous character into the current character + // register. + Trace new_trace(*trace); + new_trace.InvalidateCurrentCharacter(); + + jit::Label ok; + if (new_trace.cp_offset() == 0) + assembler->CheckAtStart(&ok); + + // We already checked that we are not at the start of input so it must be + // OK to load the previous character. + assembler->LoadCurrentCharacter(new_trace.cp_offset() - 1, new_trace.backtrack(), false); + assembler->CheckCharacterInRange(unicode::LeadSurrogateMin, unicode::LeadSurrogateMax, + new_trace.backtrack()); + + assembler->Bind(&ok); + on_success->Emit(compiler, &new_trace); +} + +// Assert that the next character is not a trail surrogate that has a +// corresponding lead surrogate. +static void +EmitNotInSurrogatePair(RegExpCompiler* compiler, RegExpNode* on_success, Trace* trace) +{ + RegExpMacroAssembler* assembler = compiler->macro_assembler(); + + jit::Label ok; + assembler->CheckPosition(trace->cp_offset(), &ok); + + // We will be loading the next and previous characters into the current + // character register. + Trace new_trace(*trace); + new_trace.InvalidateCurrentCharacter(); + + if (new_trace.cp_offset() == 0) + assembler->CheckAtStart(&ok); + + // First check if next character is a trail surrogate. + assembler->LoadCurrentCharacter(new_trace.cp_offset(), new_trace.backtrack(), false); + assembler->CheckCharacterNotInRange(unicode::TrailSurrogateMin, unicode::TrailSurrogateMax, + &ok); + + // Next check if previous character is a lead surrogate. + // We already checked that we are not at the start of input so it must be + // OK to load the previous character. + assembler->LoadCurrentCharacter(new_trace.cp_offset() - 1, new_trace.backtrack(), false); + assembler->CheckCharacterInRange(unicode::LeadSurrogateMin, unicode::LeadSurrogateMax, + new_trace.backtrack()); + + assembler->Bind(&ok); + on_success->Emit(compiler, &new_trace); +} + +// Check for [0-9A-Z_a-z]. +static void +EmitWordCheck(RegExpMacroAssembler* assembler, + jit::Label* word, jit::Label* non_word, bool fall_through_on_word) +{ + 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 handle \b and \B (word-boundary or non-word-boundary). +void +AssertionNode::EmitBoundaryCheck(RegExpCompiler* compiler, Trace* trace) +{ + RegExpMacroAssembler* assembler = compiler->macro_assembler(); + 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) { + int eats_at_least = + Min(kMaxLookaheadForBoyerMoore, EatsAtLeast(kMaxLookaheadForBoyerMoore, + kRecursionBudget, + not_at_start)); + if (eats_at_least >= 1) { + BoyerMooreLookahead* bm = + alloc()->newInfallible(alloc(), eats_at_least, compiler); + FillInBMInfo(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) { + jit::Label before_non_word; + jit::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); + jit::Label ok; + BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord); + assembler->JumpOrBacktrack(&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 { + MOZ_ASSERT(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(); + + jit::Label fall_through, dummy; + + jit::Label* non_word = backtrack_if_previous == kIsNonWord ? new_trace.backtrack() : &fall_through; + jit::Label* word = backtrack_if_previous == kIsNonWord ? &fall_through : new_trace.backtrack(); + + if (new_trace.cp_offset() == 0) { + // The start of input counts as a non-word character, so the question is + // decided if we are at the start. + assembler->CheckAtStart(non_word); + } + // We already checked that we are not at the start of input so it must be + // OK to load the previous character. + assembler->LoadCurrentCharacter(new_trace.cp_offset() - 1, &dummy, false); + 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: { + jit::Label ok; + assembler->CheckPosition(trace->cp_offset(), &ok); + assembler->JumpOrBacktrack(trace->backtrack()); + assembler->Bind(&ok); + break; + } + case AT_START: { + if (trace->at_start() == Trace::FALSE_VALUE) { + assembler->JumpOrBacktrack(trace->backtrack()); + return; + } + if (trace->at_start() == Trace::UNKNOWN) { + assembler->CheckNotAtStart(trace->backtrack()); + Trace at_start_trace = *trace; + at_start_trace.set_at_start(true); + 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; + } + case NOT_AFTER_LEAD_SURROGATE: + EmitNotAfterLeadSurrogate(compiler, on_success(), trace); + return; + case NOT_IN_SURROGATE_PAIR: + EmitNotInSurrogatePair(compiler, on_success(), 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; +} + +static void +EmitBoundaryTest(RegExpMacroAssembler* masm, + int border, + jit::Label* fall_through, + jit::Label* above_or_equal, + jit::Label* below) +{ + if (below != fall_through) { + masm->CheckCharacterLT(border, below); + if (above_or_equal != fall_through) + masm->JumpOrBacktrack(above_or_equal); + } else { + masm->CheckCharacterGT(border - 1, above_or_equal); + } +} + +static void +EmitDoubleBoundaryTest(RegExpMacroAssembler* masm, + int first, + int last, + jit::Label* fall_through, + jit::Label* in_range, + jit::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->JumpOrBacktrack(out_of_range); + } +} + +typedef InfallibleVector RangeBoundaryVector; + +// 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, + RangeBoundaryVector& ranges, + int start_index, + int end_index, + int min_char, + jit::Label* fall_through, + jit::Label* even_label, + jit::Label* odd_label) +{ + static const int kSize = RegExpMacroAssembler::kTableSize; + static const int kMask = RegExpMacroAssembler::kTableMask; + + DebugOnly base = (min_char & ~kMask); + + // Assert that everything is on one kTableSize page. + for (int i = start_index; i <= end_index; i++) + MOZ_ASSERT((ranges[i] & ~kMask) == base); + MOZ_ASSERT(start_index == 0 || (ranges[start_index - 1] & ~kMask) <= base); + + char templ[kSize]; + jit::Label* on_bit_set; + jit::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[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[i] & kMask); j < (ranges[i + 1] & kMask); j++) { + templ[j] = bit; + } + bit ^= 1; + } + for (int i = j; i < kSize; i++) { + templ[i] = bit; + } + + // TODO(erikcorry): Cache these. + uint8_t* ba; + { + AutoEnterOOMUnsafeRegion oomUnsafe; + ba = static_cast(js_malloc(kSize)); + if (!ba || !masm->shared->addTable(ba)) + oomUnsafe.crash("Table malloc"); + } + + for (int i = 0; i < kSize; i++) + ba[i] = templ[i]; + + masm->CheckBitInTable(ba, on_bit_set); + if (on_bit_clear != fall_through) + masm->JumpOrBacktrack(on_bit_clear); +} + +static void +CutOutRange(RegExpMacroAssembler* masm, + RangeBoundaryVector& ranges, + int start_index, + int end_index, + int cut_index, + jit::Label* even_label, + jit::Label* odd_label) +{ + bool odd = (((cut_index - start_index) & 1) == 1); + jit::Label* in_range_label = odd ? odd_label : even_label; + jit::Label dummy; + EmitDoubleBoundaryTest(masm, + ranges[cut_index], + ranges[cut_index + 1] - 1, + &dummy, + in_range_label, + &dummy); + MOZ_ASSERT(!dummy.used()); + // 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[j] = ranges[j - 1]; + for (int j = cut_index + 1; j < end_index; j++) + ranges[j] = ranges[j + 1]; +} + +// Unicode case. Split the search space into kSize spaces that are handled +// with recursion. +static void +SplitSearchSpace(RangeBoundaryVector& 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[start_index]; + int last = ranges[end_index] - 1; + + *new_start_index = start_index; + *border = (ranges[start_index] & ~kMask) + kSize; + while (*new_start_index < end_index) { + if (ranges[*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-ASCII + // 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 ASCII + // range with a single not-taken branch, speeding up this important + // character range (even non-ASCII charset-based text has spaces and + // punctuation). + if (*border - 1 > kMaxOneByteCharCode && // ASCII case. + end_index - start_index > (*new_start_index - start_index) * 2 && + last - first > kSize * 2 && + binary_chop_index > *new_start_index && + ranges[binary_chop_index] >= first + 2 * kSize) + { + int scan_forward_for_section_border = binary_chop_index;; + int new_border = (ranges[binary_chop_index] | kMask) + 1; + + while (scan_forward_for_section_border < end_index) { + if (ranges[scan_forward_for_section_border] > new_border) { + *new_start_index = scan_forward_for_section_border; + *border = new_border; + break; + } + scan_forward_for_section_border++; + } + } + + MOZ_ASSERT(*new_start_index > start_index); + *new_end_index = *new_start_index - 1; + if (ranges[*new_end_index] == *border) + (*new_end_index)--; + if (*border >= ranges[end_index]) { + *border = ranges[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, + RangeBoundaryVector& ranges, + int start_index, + int end_index, + char16_t min_char, + char16_t max_char, + jit::Label* fall_through, + jit::Label* even_label, + jit::Label* odd_label) +{ + int first = ranges[start_index]; + int last = ranges[end_index] - 1; + + MOZ_ASSERT(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[i] == ranges[i + 1] - 1) { + cut = i; + break; + } + } + if (cut == kNoCutIndex) cut = start_index; + CutOutRange(masm, ranges, start_index, end_index, cut, even_label, odd_label); + MOZ_ASSERT(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); + + jit::Label handle_rest; + jit::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; + MOZ_ASSERT(new_end_index == end_index - 1); + } + + MOZ_ASSERT(start_index <= new_end_index); + MOZ_ASSERT(new_start_index <= end_index); + MOZ_ASSERT(start_index < new_start_index); + MOZ_ASSERT(new_end_index < end_index); + MOZ_ASSERT(new_end_index + 1 == new_start_index || + (new_end_index + 2 == new_start_index && + border == ranges[new_end_index + 1])); + MOZ_ASSERT(min_char < border - 1); + MOZ_ASSERT(border < max_char); + MOZ_ASSERT(ranges[new_end_index] < border); + MOZ_ASSERT(border < ranges[new_start_index] || + (border == ranges[new_start_index] && + new_start_index == end_index && + new_end_index == end_index - 1 && + border == last + 1)); + MOZ_ASSERT(new_start_index == 0 || border >= ranges[new_start_index - 1]); + + masm->CheckCharacterGT(border - 1, above); + jit::Label dummy; + GenerateBranches(masm, + ranges, + start_index, + new_end_index, + min_char, + border - 1, + &dummy, + even_label, + odd_label); + if (handle_rest.used()) { + 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(LifoAlloc* alloc, + RegExpMacroAssembler* macro_assembler, + RegExpCharacterClass* cc, + bool ascii, + jit::Label* on_failure, + int cp_offset, + bool check_offset, + bool preloaded) +{ + CharacterRangeVector& ranges = cc->ranges(alloc); + if (!CharacterRange::IsCanonical(ranges)) { + CharacterRange::Canonicalize(ranges); + } + + int max_char = MaximumCharacter(ascii); + int range_count = ranges.length(); + + int last_valid_range = range_count - 1; + while (last_valid_range >= 0) { + CharacterRange& range = ranges[last_valid_range]; + if (range.from() <= max_char) { + break; + } + last_valid_range--; + } + + if (last_valid_range < 0) { + if (!cc->is_negated()) { + macro_assembler->JumpOrBacktrack(on_failure); + } + if (check_offset) { + macro_assembler->CheckPosition(cp_offset, on_failure); + } + return; + } + + if (last_valid_range == 0 && + ranges[0].IsEverything(max_char)) { + if (cc->is_negated()) { + macro_assembler->JumpOrBacktrack(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 (last_valid_range == 0 && + !cc->is_negated() && + ranges[0].IsEverything(max_char)) { + // 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(alloc) && + 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). + RangeBoundaryVector* range_boundaries = + alloc->newInfallible(*alloc); + + bool zeroth_entry_is_failure = !cc->is_negated(); + + range_boundaries->reserve(last_valid_range); + for (int i = 0; i <= last_valid_range; i++) { + CharacterRange& range = ranges[i]; + if (range.from() == 0) { + MOZ_ASSERT(i == 0); + zeroth_entry_is_failure = !zeroth_entry_is_failure; + } else { + range_boundaries->append(range.from()); + } + range_boundaries->append(range.to() + 1); + } + int end_index = range_boundaries->length() - 1; + if ((*range_boundaries)[end_index] > max_char) + end_index--; + + jit::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); +} + +typedef bool EmitCharacterFunction(RegExpCompiler* compiler, + char16_t c, + jit::Label* on_failure, + int cp_offset, + bool check, + bool preloaded); + +static inline bool +EmitSimpleCharacter(RegExpCompiler* compiler, + char16_t c, + jit::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(RegExpCompiler* compiler, + char16_t c, + jit::Label* on_failure, + int cp_offset, + bool check, + bool preloaded) +{ + RegExpMacroAssembler* macro_assembler = compiler->macro_assembler(); + bool ascii = compiler->ascii(); + char16_t chars[kEcma262UnCanonicalizeMaxWidth]; + int length = GetCaseIndependentLetters(c, ascii, compiler->unicode(), chars); + if (length < 1) { + // This can't match. Must be an ASCII subject and a non-ASCII character. + // We do not need to do anything since the ASCII 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 (ascii && c > kMaxOneByteCharCode) { + // 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 ascii, + char16_t c1, + char16_t c2, + jit::Label* on_failure) +{ + char16_t char_mask = MaximumCharacter(ascii); + + MOZ_ASSERT(c1 != c2); + if (c1 > c2) { + char16_t tmp = c1; + c1 = c2; + c2 = tmp; + } + + char16_t 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. + char16_t mask = char_mask ^ exor; + macro_assembler->CheckNotCharacterAfterAnd(c1, mask, on_failure); + return true; + } + + char16_t 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. + char16_t mask = char_mask ^ diff; + macro_assembler->CheckNotCharacterAfterMinusAnd(c1 - diff, + diff, + mask, + on_failure); + return true; + } + return false; +} + +// Only emits letters (things that have case). Only used for case independent +// matches. +static inline bool +EmitAtomLetter(RegExpCompiler* compiler, + char16_t c, + jit::Label* on_failure, + int cp_offset, + bool check, + bool preloaded) +{ + RegExpMacroAssembler* macro_assembler = compiler->macro_assembler(); + bool ascii = compiler->ascii(); + char16_t chars[kEcma262UnCanonicalizeMaxWidth]; + int length = GetCaseIndependentLetters(c, ascii, compiler->unicode(), chars); + 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); + jit::Label ok; + MOZ_ASSERT(kEcma262UnCanonicalizeMaxWidth == 4); + switch (length) { + case 2: { + if (ShortCutEmitCharacterPair(macro_assembler, + ascii, + 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); + MOZ_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: + MOZ_CRASH("Bad length"); + } + return true; +} + +// 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(); + bool ascii = compiler->ascii(); + jit::Label* backtrack = trace->backtrack(); + QuickCheckDetails* quick_check = trace->quick_check_performed(); + int element_count = elements().length(); + for (int i = preloaded ? 0 : element_count - 1; i >= 0; i--) { + TextElement elm = elements()[i]; + int cp_offset = trace->cp_offset() + elm.cp_offset(); + if (elm.text_type() == TextElement::ATOM) { + const CharacterVector& 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; + switch (pass) { + case NON_ASCII_MATCH: + MOZ_ASSERT(ascii); + if (quarks[j] > kMaxOneByteCharCode) { + assembler->JumpOrBacktrack(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 bound_checked = emit_function(compiler, + quarks[j], + backtrack, + cp_offset + j, + *checked_up_to < cp_offset + j, + preloaded); + if (bound_checked) UpdateBoundsCheck(cp_offset + j, checked_up_to); + } + } + } else { + MOZ_ASSERT(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(); + EmitCharClass(alloc(), + assembler, + cc, + ascii, + backtrack, + cp_offset, + *checked_up_to < cp_offset, + preloaded); + UpdateBoundsCheck(cp_offset, checked_up_to); + } + } + } +} + +int +TextNode::Length() +{ + TextElement elm = elements()[elements().length() - 1]; + MOZ_ASSERT(elm.cp_offset() >= 0); + return elm.cp_offset() + elm.length(); +} + +bool +TextNode::SkipPass(int int_pass, bool ignore_case) +{ + TextEmitPassType pass = static_cast(int_pass); + if (ignore_case) + return pass == SIMPLE_CHARACTER_MATCH; + return pass == NON_LETTER_CHARACTER_MATCH || pass == CASE_CHARACTER_MATCH; +} + +// 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; + MOZ_ASSERT(limit_result == CONTINUE); + + if (trace->cp_offset() + Length() > RegExpMacroAssembler::kMaxCPOffset) { + compiler->SetRegExpTooBig(); + return; + } + + if (compiler->ascii()) { + int dummy = 0; + TextEmitPass(compiler, NON_ASCII_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++) { + if (!SkipPass(pass, compiler->ignore_case())) { + TextEmitPass(compiler, + static_cast(pass), + true, + trace, + false, + &bound_checked_to); + } + } + first_elt_done = true; + } + + for (int pass = kFirstRealPass; pass <= kLastPass; pass++) { + if (!SkipPass(pass, compiler->ignore_case())) { + TextEmitPass(compiler, + static_cast(pass), + false, + trace, + first_elt_done, + &bound_checked_to); + } + } + + Trace successor_trace(*trace); + successor_trace.set_at_start(false); + successor_trace.AdvanceCurrentPositionInTrace(Length(), compiler); + RecursionCheck rc(compiler); + on_success()->Emit(compiler, &successor_trace); +} + +void +LoopChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) +{ + RegExpMacroAssembler* macro_assembler = compiler->macro_assembler(); + if (trace->stop_node() == this) { + int text_length = + GreedyLoopTextLengthForAlternative(&alternatives()[0]); + MOZ_ASSERT(text_length != kNodeIsTooComplexForGreedyLoops); + // Update the counter-based backtracking info on the stack. This is an + // optimization for greedy loops (see below). + MOZ_ASSERT(trace->cp_offset() == text_length); + macro_assembler->AdvanceCurrentPosition(text_length); + macro_assembler->JumpOrBacktrack(trace->loop_label()); + return; + } + MOZ_ASSERT(trace->stop_node() == nullptr); + if (!trace->is_trivial()) { + trace->Flush(compiler, this); + return; + } + ChoiceNode::Emit(compiler, trace); +} + +/* 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 JumpOrBacktrack 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 reverse our expectation and expect to match rather + * than fail. Therefore we want the loop code to look like this (U is the + * unwind code that steps back in the greedy loop). The following alternatives + * look the same as above. + * _____ + * / \ + * V | + * ----------> S1 | + * /| | + * / |S | + * F/ \_____/ + * / + * |<----------- + * | \ + * V \ + * Q2 ---> S2 \ + * | S / | + * F| / | + * | F/ | + * | / | + * | R | + * | / | + * F VL | + * <------U | + * back |S | + * \______________/ + */ + +// This class is used when generating the alternatives in a choice node. It +// records the way the alternative is being code generated. +class irregexp::AlternativeGeneration +{ + public: + AlternativeGeneration() + : possible_success(), + expects_preload(false), + after(), + quick_check_details() + {} + + jit::Label possible_success; + bool expects_preload; + jit::Label after; + QuickCheckDetails quick_check_details; +}; + +void +ChoiceNode::GenerateGuard(RegExpMacroAssembler* macro_assembler, + Guard* guard, Trace* trace) +{ + switch (guard->op()) { + case Guard::LT: + MOZ_ASSERT(!trace->mentions_reg(guard->reg())); + macro_assembler->IfRegisterGE(guard->reg(), + guard->value(), + trace->backtrack()); + break; + case Guard::GEQ: + MOZ_ASSERT(!trace->mentions_reg(guard->reg())); + macro_assembler->IfRegisterLT(guard->reg(), + guard->value(), + trace->backtrack()); + break; + } +} + +int +ChoiceNode::CalculatePreloadCharacters(RegExpCompiler* compiler, int eats_at_least) +{ + int preload_characters = Min(4, eats_at_least); + if (compiler->macro_assembler()->CanReadUnaligned()) { + bool ascii = compiler->ascii(); + if (ascii) { + if (preload_characters > 4) + preload_characters = 4; + // 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; +} + +RegExpNode* +TextNode::GetSuccessorOfOmnivorousTextNode(RegExpCompiler* compiler) +{ + if (elements().length() != 1) + return nullptr; + + TextElement elm = elements()[0]; + if (elm.text_type() != TextElement::CHAR_CLASS) + return nullptr; + + RegExpCharacterClass* node = elm.char_class(); + CharacterRangeVector& ranges = node->ranges(alloc()); + + if (!CharacterRange::IsCanonical(ranges)) + CharacterRange::Canonicalize(ranges); + + if (node->is_negated()) + return ranges.length() == 0 ? on_success() : nullptr; + + if (ranges.length() != 1) + return nullptr; + + uint32_t max_char = MaximumCharacter(compiler->ascii()); + return ranges[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(node); + node = seq_node->on_success(); + } + return length; +} + +// 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(LifoAlloc* alloc, size_t count) + : alt_gens_(*alloc) + { + alt_gens_.reserve(count); + for (size_t i = 0; i < count && i < kAFew; i++) + alt_gens_.append(a_few_alt_gens_ + i); + for (size_t i = kAFew; i < count; i++) { + AutoEnterOOMUnsafeRegion oomUnsafe; + AlternativeGeneration* gen = js_new(); + if (!gen) + oomUnsafe.crash("AlternativeGenerationList js_new"); + alt_gens_.append(gen); + } + } + + ~AlternativeGenerationList() { + for (size_t i = kAFew; i < alt_gens_.length(); i++) { + js_delete(alt_gens_[i]); + alt_gens_[i] = nullptr; + } + } + + AlternativeGeneration* at(int i) { + return alt_gens_[i]; + } + + private: + static const size_t kAFew = 10; + InfallibleVector alt_gens_; + AlternativeGeneration a_few_alt_gens_[kAFew]; +}; + +void +ChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) +{ + RegExpMacroAssembler* macro_assembler = compiler->macro_assembler(); + size_t choice_count = alternatives().length(); +#ifdef DEBUG + for (size_t i = 0; i < choice_count - 1; i++) { + const GuardedAlternative& alternative = alternatives()[i]; + const GuardVector* guards = alternative.guards(); + if (guards) { + for (size_t j = 0; j < guards->length(); j++) + MOZ_ASSERT(!trace->mentions_reg((*guards)[j]->reg())); + } + } +#endif + + LimitResult limit_result = LimitVersions(compiler, trace); + if (limit_result == DONE) return; + MOZ_ASSERT(limit_result == CONTINUE); + + int new_flush_budget = trace->flush_budget() / choice_count; + if (trace->flush_budget() == 0 && trace->actions() != nullptr) { + trace->Flush(compiler, this); + return; + } + + RecursionCheck rc(compiler); + + Trace* current_trace = trace; + + int text_length = GreedyLoopTextLengthForAlternative(&alternatives()[0]); + bool greedy_loop = false; + jit::Label greedy_loop_label; + Trace counter_backtrack_trace; + counter_backtrack_trace.set_backtrack(&greedy_loop_label); + if (not_at_start()) counter_backtrack_trace.set_at_start(false); + + if (choice_count > 1 && text_length != kNodeIsTooComplexForGreedyLoops) { + // 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. + greedy_loop = true; + MOZ_ASSERT(trace->stop_node() == nullptr); + macro_assembler->PushCurrentPosition(); + current_trace = &counter_backtrack_trace; + jit::Label greedy_match_failed; + Trace greedy_match_trace; + if (not_at_start()) greedy_match_trace.set_at_start(false); + greedy_match_trace.set_backtrack(&greedy_match_failed); + jit::Label loop_label; + macro_assembler->Bind(&loop_label); + greedy_match_trace.set_stop_node(this); + greedy_match_trace.set_loop_label(&loop_label); + alternatives()[0].node()->Emit(compiler, &greedy_match_trace); + macro_assembler->Bind(&greedy_match_failed); + } + + jit::Label second_choice; // For use in greedy matches. + macro_assembler->Bind(&second_choice); + + size_t first_normal_choice = greedy_loop ? 1 : 0; + + bool not_at_start = current_trace->at_start() == Trace::FALSE_VALUE; + const int kEatsAtLeastNotYetInitialized = -1; + int eats_at_least = kEatsAtLeastNotYetInitialized; + + bool skip_was_emitted = false; + + if (!greedy_loop && choice_count == 2) { + GuardedAlternative alt1 = alternatives()[1]; + if (!alt1.guards() || alt1.guards()->length() == 0) { + RegExpNode* eats_anything_node = alt1.node(); + if (eats_anything_node->GetSuccessorOfOmnivorousTextNode(compiler) == this) { + // 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. + MOZ_ASSERT(trace->is_trivial()); // This is the case on LoopChoiceNodes. + BoyerMooreLookahead* lookahead = bm_info(not_at_start); + if (lookahead == nullptr) { + eats_at_least = Min(kMaxLookaheadForBoyerMoore, + EatsAtLeast(kMaxLookaheadForBoyerMoore, + kRecursionBudget, + not_at_start)); + if (eats_at_least >= 1) { + BoyerMooreLookahead* bm = + alloc()->newInfallible(alloc(), eats_at_least, compiler); + GuardedAlternative alt0 = alternatives()[0]; + alt0.node()->FillInBMInfo(0, kRecursionBudget, bm, not_at_start); + skip_was_emitted = bm->EmitSkipInstructions(macro_assembler); + } + } else { + skip_was_emitted = lookahead->EmitSkipInstructions(macro_assembler); + } + } + } + } + + if (eats_at_least == kEatsAtLeastNotYetInitialized) { + // Save some time by looking at most one machine word ahead. + eats_at_least = + EatsAtLeast(compiler->ascii() ? 4 : 2, kRecursionBudget, not_at_start); + } + int preload_characters = CalculatePreloadCharacters(compiler, eats_at_least); + + bool preload_is_current = !skip_was_emitted && + (current_trace->characters_preloaded() == preload_characters); + bool preload_has_checked_bounds = preload_is_current; + + AlternativeGenerationList alt_gens(alloc(), choice_count); + + // 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. + for (size_t i = first_normal_choice; i < choice_count; i++) { + GuardedAlternative alternative = alternatives()[i]; + AlternativeGeneration* alt_gen = alt_gens.at(i); + alt_gen->quick_check_details.set_characters(preload_characters); + const GuardVector* guards = alternative.guards(); + Trace new_trace(*current_trace); + new_trace.set_characters_preloaded(preload_is_current ? + preload_characters : + 0); + if (preload_has_checked_bounds) { + new_trace.set_bound_checked_up_to(preload_characters); + } + new_trace.quick_check_performed()->Clear(); + if (not_at_start_) new_trace.set_at_start(Trace::FALSE_VALUE); + alt_gen->expects_preload = preload_is_current; + bool generate_full_check_inline = false; + if (try_to_emit_quick_check_for_alternative(i) && + alternative.node()->EmitQuickCheck(compiler, + &new_trace, + preload_has_checked_bounds, + &alt_gen->possible_success, + &alt_gen->quick_check_details, + i < choice_count - 1)) { + // Quick check was generated for this choice. + preload_is_current = true; + preload_has_checked_bounds = true; + // On the last choice in the ChoiceNode we generated the quick + // check to fall through on possible success. So now we need to + // generate the full check inline. + if (i == choice_count - 1) { + macro_assembler->Bind(&alt_gen->possible_success); + new_trace.set_quick_check_performed(&alt_gen->quick_check_details); + new_trace.set_characters_preloaded(preload_characters); + new_trace.set_bound_checked_up_to(preload_characters); + generate_full_check_inline = true; + } + } else if (alt_gen->quick_check_details.cannot_match()) { + if (i == choice_count - 1 && !greedy_loop) { + macro_assembler->JumpOrBacktrack(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_normal_choice) { + alt_gen->expects_preload = false; + new_trace.InvalidateCurrentCharacter(); + } + if (i < choice_count - 1) { + new_trace.set_backtrack(&alt_gen->after); + } + generate_full_check_inline = true; + } + if (generate_full_check_inline) { + if (new_trace.actions() != nullptr) + new_trace.set_flush_budget(new_flush_budget); + if (guards) { + for (size_t j = 0; j < guards->length(); j++) + GenerateGuard(macro_assembler, (*guards)[j], &new_trace); + } + alternative.node()->Emit(compiler, &new_trace); + preload_is_current = false; + } + macro_assembler->Bind(&alt_gen->after); + } + if (greedy_loop) { + macro_assembler->Bind(&greedy_loop_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->JumpOrBacktrack(&second_choice); + } + + // 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. + for (size_t i = first_normal_choice; i < choice_count - 1; i++) { + AlternativeGeneration* alt_gen = alt_gens.at(i); + Trace new_trace(*current_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); + } + EmitOutOfLineContinuation(compiler, + &new_trace, + alternatives()[i], + alt_gen, + preload_characters, + alt_gens.at(i + 1)->expects_preload); + } +} + +void +ChoiceNode::EmitOutOfLineContinuation(RegExpCompiler* compiler, + Trace* trace, + GuardedAlternative alternative, + AlternativeGeneration* alt_gen, + int preload_characters, + bool next_expects_preload) +{ + if (!alt_gen->possible_success.used()) + 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); + const GuardVector* guards = alternative.guards(); + if (next_expects_preload) { + jit::Label reload_current_char; + out_of_line_trace.set_backtrack(&reload_current_char); + if (guards) { + for (size_t j = 0; j < guards->length(); j++) + GenerateGuard(macro_assembler, (*guards)[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->JumpOrBacktrack(&(alt_gen->after)); + } else { + out_of_line_trace.set_backtrack(&(alt_gen->after)); + if (guards) { + for (size_t j = 0; j < guards->length(); j++) + GenerateGuard(macro_assembler, (*guards)[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; + MOZ_ASSERT(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; + } + case SET_REGISTER: { + Trace::DeferredSetRegister + new_set(data_.u_store_register.reg, data_.u_store_register.value); + 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->WriteBacktrackStackPointerToRegister(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->JumpOrBacktrack(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 { + jit::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->ReadBacktrackStackPointerFromRegister(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; + jit::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); + + MOZ_ASSERT(trace->backtrack() == nullptr); + assembler->Backtrack(); + return; + } + default: + MOZ_CRASH("Bad action"); + } +} + +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; + MOZ_ASSERT(limit_result == CONTINUE); + + RecursionCheck rc(compiler); + + MOZ_ASSERT(start_reg_ + 1 == end_reg_); + if (compiler->ignore_case()) { + assembler->CheckNotBackReferenceIgnoreCase(start_reg_, + trace->backtrack(), + compiler->unicode()); + } else { + assembler->CheckNotBackReference(start_reg_, trace->backtrack()); + } + on_success()->Emit(compiler, trace); +} + +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()->bound()) { + // We are being asked to generate a generic version, but that's already + // been done so just go to it. + macro_assembler->JumpOrBacktrack(label()); + return DONE; + } + if (compiler->recursion_depth() >= RegExpCompiler::kMaxRecursion) { + // To avoid too deep recursion we push the node to the work queue and just + // generate a goto here. + compiler->AddWork(this); + macro_assembler->JumpOrBacktrack(label()); + 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 (trace_count_ < kMaxCopiesCodeGenerated && + compiler->recursion_depth() <= RegExpCompiler::kMaxRecursion) { + 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. + trace->Flush(compiler, this); + return DONE; +} + +bool +RegExpNode::EmitQuickCheck(RegExpCompiler* compiler, + Trace* trace, + bool preload_has_checked_bounds, + jit::Label* on_possible_success, + QuickCheckDetails* details, + bool fall_through_on_failure) +{ + 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->ascii())) return false; + MOZ_ASSERT(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()) { + assembler->LoadCurrentCharacter(trace->cp_offset(), + trace->backtrack(), + !preload_has_checked_bounds, + details->characters()); + } + + 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 = MaximumCharacter(compiler->ascii()); + if ((mask & char_mask) == char_mask) need_mask = false; + mask &= char_mask; + } else { + // For 2-character preloads in ASCII mode or 1-character preloads in + // TWO_BYTE mode we also use a 16 bit load with zero extend. + if (details->characters() == 2 && compiler->ascii()) { + if ((mask & 0xffff) == 0xffff) need_mask = false; + } else if (details->characters() == 1 && !compiler->ascii()) { + if ((mask & 0xffff) == 0xffff) need_mask = false; + } else { + if (mask == 0xffffffff) 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; +} + +bool +TextNode::FillInBMInfo(int initial_offset, + int budget, + BoyerMooreLookahead* bm, + bool not_at_start) +{ + if (!bm->CheckOverRecursed()) + return false; + + if (initial_offset >= bm->length()) + return true; + + int offset = initial_offset; + int max_char = bm->max_char(); + for (size_t i = 0; i < elements().length(); i++) { + if (offset >= bm->length()) { + if (initial_offset == 0) + set_bm_info(not_at_start, bm); + return true; + } + TextElement text = elements()[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 true; + } + char16_t character = atom->data()[j]; + if (bm->compiler()->ignore_case()) { + char16_t chars[kEcma262UnCanonicalizeMaxWidth]; + int length = GetCaseIndependentLetters(character, + bm->max_char() == kMaxOneByteCharCode, + bm->compiler()->unicode(), + chars); + for (int j = 0; j < length; j++) + bm->Set(offset, chars[j]); + } else { + if (character <= max_char) bm->Set(offset, character); + } + } + } else { + MOZ_ASSERT(TextElement::CHAR_CLASS == text.text_type()); + RegExpCharacterClass* char_class = text.char_class(); + const CharacterRangeVector& ranges = char_class->ranges(alloc()); + if (char_class->is_negated()) { + bm->SetAll(offset); + } else { + for (size_t k = 0; k < ranges.length(); k++) { + const CharacterRange& range = ranges[k]; + if (range.from() > max_char) + continue; + int to = Min(max_char, static_cast(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 true; + } + if (!on_success()->FillInBMInfo(offset, + budget - 1, + bm, + true)) // Not at start after a text node. + return false; + if (initial_offset == 0) + set_bm_info(not_at_start, bm); + return true; +} + +// ------------------------------------------------------------------- +// QuickCheckDetails + +// 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; +} + +// 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) +{ + MOZ_ASSERT(characters_filled_in < details->characters()); + int characters = details->characters(); + int char_mask = MaximumCharacter(compiler->ascii()); + + for (size_t k = 0; k < elements().length(); k++) { + TextElement elm = elements()[k]; + if (elm.text_type() == TextElement::ATOM) { + const CharacterVector& quarks = elm.atom()->data(); + for (size_t i = 0; i < (size_t) characters && i < quarks.length(); i++) { + QuickCheckDetails::Position* pos = + details->positions(characters_filled_in); + char16_t c = quarks[i]; + if (c > char_mask) { + // If we expect a non-ASCII character from an ASCII string, + // there is no way we can match. Not even case independent + // matching can turn an ASCII character into non-ASCII or + // vice versa. + details->set_cannot_match(); + pos->determines_perfectly = false; + return; + } + if (compiler->ignore_case()) { + char16_t chars[kEcma262UnCanonicalizeMaxWidth]; + size_t length = GetCaseIndependentLetters(c, compiler->ascii(), + compiler->unicode(), chars); + MOZ_ASSERT(length != 0); // Can only happen if c > char_mask (see above). + 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; + pos->value = c; + pos->determines_perfectly = true; + } else { + uint32_t common_bits = char_mask; + uint32_t bits = chars[0]; + for (size_t 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. + pos->mask = char_mask; + pos->value = c; + pos->determines_perfectly = true; + } + characters_filled_in++; + MOZ_ASSERT(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(); + const CharacterRangeVector& ranges = tree->ranges(alloc()); + 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 { + size_t first_range = 0; + while (ranges[first_range].from() > char_mask) { + first_range++; + if (first_range == ranges.length()) { + details->set_cannot_match(); + pos->determines_perfectly = false; + return; + } + } + CharacterRange range = ranges[first_range]; + char16_t from = range.from(); + char16_t 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 (size_t i = first_range + 1; i < ranges.length(); i++) { + CharacterRange range = ranges[i]; + char16_t from = range.from(); + char16_t 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++; + MOZ_ASSERT(characters_filled_in <= details->characters()); + if (characters_filled_in == details->characters()) { + return; + } + } + } + MOZ_ASSERT(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 ascii) +{ + MOZ_ASSERT(by >= 0); + if (by >= characters_) { + Clear(); + return; + } + 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. +} + +bool +QuickCheckDetails::Rationalize(bool is_ascii) +{ + bool found_useful_op = false; + uint32_t char_mask = MaximumCharacter(is_ascii); + + mask_ = 0; + value_ = 0; + int char_shift = 0; + for (int i = 0; i < characters_; i++) { + Position* pos = &positions_[i]; + if ((pos->mask & kMaxOneByteCharCode) != 0) + found_useful_op = true; + mask_ |= (pos->mask & char_mask) << char_shift; + value_ |= (pos->value & char_mask) << char_shift; + char_shift += is_ascii ? 8 : 16; + } + return found_useful_op; +} + +void QuickCheckDetails::Merge(QuickCheckDetails* other, int from_index) +{ + MOZ_ASSERT(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; + char16_t differing_bits = (pos->value ^ other_pos->value); + pos->mask &= ~differing_bits; + pos->value &= pos->mask; + } +} diff --git a/js/src/irregexp/RegExpEngine.h b/js/src/irregexp/RegExpEngine.h new file mode 100644 index 000000000..78c784aaf --- /dev/null +++ b/js/src/irregexp/RegExpEngine.h @@ -0,0 +1,1546 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- + * vim: set ts=8 sts=4 et sw=4 tw=99: */ + +// Copyright 2012 the V8 project authors. All rights reserved. +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived +// from this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#ifndef V8_JSREGEXP_H_ +#define V8_JSREGEXP_H_ + +#include "jscntxt.h" + +#include "ds/SplayTree.h" +#include "jit/Label.h" +#include "vm/RegExpObject.h" + +namespace js { + +class MatchPairs; +class RegExpShared; + +namespace jit { + class Label; + class JitCode; +} + +namespace irregexp { + +class RegExpTree; +class RegExpMacroAssembler; + +struct RegExpCompileData +{ + RegExpCompileData() + : tree(nullptr), + simple(true), + contains_anchor(false), + capture_count(0) + {} + + RegExpTree* tree; + bool simple; + bool contains_anchor; + int capture_count; +}; + +struct RegExpCode +{ + jit::JitCode* jitCode; + uint8_t* byteCode; + + RegExpCode() + : jitCode(nullptr), byteCode(nullptr) + {} + + bool empty() { + return !jitCode && !byteCode; + } + + void destroy() { + js_free(byteCode); + } +}; + +RegExpCode +CompilePattern(JSContext* cx, RegExpShared* shared, RegExpCompileData* data, + HandleLinearString sample, bool is_global, bool ignore_case, + bool is_ascii, bool match_only, bool force_bytecode, bool sticky, + bool unicode); + +// Note: this may return RegExpRunStatus_Error if an interrupt was requested +// while the code was executing. +template +RegExpRunStatus +ExecuteCode(JSContext* cx, jit::JitCode* codeBlock, const CharT* chars, size_t start, + size_t length, MatchPairs* matches, size_t* endIndex); + +template +RegExpRunStatus +InterpretCode(JSContext* cx, const uint8_t* byteCode, const CharT* chars, size_t start, + size_t length, MatchPairs* matches, size_t* endIndex); + +#define FOR_EACH_NODE_TYPE(VISIT) \ + VISIT(End) \ + VISIT(Action) \ + VISIT(Choice) \ + VISIT(BackReference) \ + VISIT(Assertion) \ + VISIT(Text) + +#define FOR_EACH_REG_EXP_TREE_TYPE(VISIT) \ + VISIT(Disjunction) \ + VISIT(Alternative) \ + VISIT(Assertion) \ + VISIT(CharacterClass) \ + VISIT(Atom) \ + VISIT(Quantifier) \ + VISIT(Capture) \ + VISIT(Lookahead) \ + VISIT(BackReference) \ + VISIT(Empty) \ + VISIT(Text) + +#define FORWARD_DECLARE(Name) class RegExp##Name; +FOR_EACH_REG_EXP_TREE_TYPE(FORWARD_DECLARE) +#undef FORWARD_DECLARE + +// InfallibleVector is like Vector, but all its methods are infallible (they +// crash on OOM). We use this class instead of Vector to avoid a ton of +// MOZ_MUST_USE warnings in irregexp code (imported from V8). +template +class InfallibleVector +{ + Vector> vector_; + + InfallibleVector(const InfallibleVector&) = delete; + void operator=(const InfallibleVector&) = delete; + + public: + explicit InfallibleVector(const LifoAllocPolicy& alloc) : vector_(alloc) {} + + void append(const T& t) { MOZ_ALWAYS_TRUE(vector_.append(t)); } + void append(const T* begin, size_t length) { MOZ_ALWAYS_TRUE(vector_.append(begin, length)); } + + void clear() { vector_.clear(); } + void popBack() { vector_.popBack(); } + void reserve(size_t n) { MOZ_ALWAYS_TRUE(vector_.reserve(n)); } + + size_t length() const { return vector_.length(); } + T popCopy() { return vector_.popCopy(); } + + T* begin() { return vector_.begin(); } + const T* begin() const { return vector_.begin(); } + + T& operator[](size_t index) { return vector_[index]; } + const T& operator[](size_t index) const { return vector_[index]; } + + InfallibleVector& operator=(InfallibleVector&& rhs) { vector_ = Move(rhs.vector_); return *this; } +}; + +class CharacterRange; +typedef InfallibleVector CharacterRangeVector; + +// Represents code units in the range from from_ to to_, both ends are +// inclusive. +class CharacterRange +{ + public: + CharacterRange() + : from_(0), to_(0) + {} + + CharacterRange(char16_t from, char16_t to) + : from_(from), to_(to) + {} + + static void AddClassEscape(LifoAlloc* alloc, char16_t type, CharacterRangeVector* ranges); + static void AddClassEscapeUnicode(LifoAlloc* alloc, char16_t type, + CharacterRangeVector* ranges, bool ignoreCase); + + static inline CharacterRange Singleton(char16_t value) { + return CharacterRange(value, value); + } + static inline CharacterRange Range(char16_t from, char16_t to) { + MOZ_ASSERT(from <= to); + return CharacterRange(from, to); + } + static inline CharacterRange Everything() { + return CharacterRange(0, 0xFFFF); + } + bool Contains(char16_t i) { return from_ <= i && i <= to_; } + char16_t from() const { return from_; } + void set_from(char16_t value) { from_ = value; } + char16_t to() const { return to_; } + void set_to(char16_t value) { to_ = value; } + bool is_valid() { return from_ <= to_; } + bool IsEverything(char16_t max) { return from_ == 0 && to_ >= max; } + bool IsSingleton() { return (from_ == to_); } + void AddCaseEquivalents(bool is_ascii, bool unicode, CharacterRangeVector* ranges); + + static void Split(const LifoAlloc* alloc, + CharacterRangeVector base, + const Vector& overlay, + CharacterRangeVector* included, + CharacterRangeVector* excluded); + + // Whether a range list is in canonical form: Ranges ordered by from value, + // and ranges non-overlapping and non-adjacent. + static bool IsCanonical(const CharacterRangeVector& ranges); + + // Convert range list to canonical form. The characters covered by the ranges + // will still be the same, but no character is in more than one range, and + // adjacent ranges are merged. The resulting list may be shorter than the + // original, but cannot be longer. + static void Canonicalize(CharacterRangeVector& ranges); + + // Negate the contents of a character range in canonical form. + static void Negate(const LifoAlloc* alloc, + CharacterRangeVector src, + CharacterRangeVector* dst); + + static const int kStartMarker = (1 << 24); + static const int kPayloadMask = (1 << 24) - 1; + + private: + char16_t from_; + char16_t to_; +}; + +// A set of unsigned integers that behaves especially well on small +// integers (< 32). +class OutSet +{ + public: + OutSet() + : first_(0), remaining_(nullptr), successors_(nullptr) + {} + + OutSet* Extend(LifoAlloc* alloc, unsigned value); + bool Get(unsigned value); + static const unsigned kFirstLimit = 32; + + private: + typedef InfallibleVector OutSetVector; + typedef InfallibleVector RemainingVector; + + // Destructively set a value in this set. In most cases you want + // to use Extend instead to ensure that only one instance exists + // that contains the same values. + void Set(LifoAlloc* alloc, unsigned value); + + // The successors are a list of sets that contain the same values + // as this set and the one more value that is not present in this + // set. + OutSetVector* successors() { return successors_; } + + OutSet(uint32_t first, RemainingVector* remaining) + : first_(first), remaining_(remaining), successors_(nullptr) + {} + + RemainingVector& remaining() { return *remaining_; } + + uint32_t first_; + RemainingVector* remaining_; + OutSetVector* successors_; + friend class Trace; +}; + +// A mapping from integers, specified as ranges, to a set of integers. +// Used for mapping character ranges to choices. +class DispatchTable +{ + public: + explicit DispatchTable(LifoAlloc* alloc) + {} + + class Entry { + public: + Entry() + : from_(0), to_(0), out_set_(nullptr) + {} + + Entry(char16_t from, char16_t to, OutSet* out_set) + : from_(from), to_(to), out_set_(out_set) + {} + + char16_t from() { return from_; } + char16_t to() { return to_; } + void set_to(char16_t value) { to_ = value; } + void AddValue(LifoAlloc* alloc, int value) { + out_set_ = out_set_->Extend(alloc, value); + } + OutSet* out_set() { return out_set_; } + private: + char16_t from_; + char16_t to_; + OutSet* out_set_; + }; + + void AddRange(LifoAlloc* alloc, CharacterRange range, int value); + OutSet* Get(char16_t value); + void Dump(); + + private: + // There can't be a static empty set since it allocates its + // successors in a LifoAlloc and caches them. + OutSet* empty() { return &empty_; } + OutSet empty_; +}; + +class TextElement +{ + public: + enum TextType { + ATOM, + CHAR_CLASS + }; + + static TextElement Atom(RegExpAtom* atom); + static TextElement CharClass(RegExpCharacterClass* char_class); + + int cp_offset() const { return cp_offset_; } + void set_cp_offset(int cp_offset) { cp_offset_ = cp_offset; } + int length() const; + + TextType text_type() const { return text_type_; } + + RegExpTree* tree() const { return tree_; } + + RegExpAtom* atom() const { + MOZ_ASSERT(text_type() == ATOM); + return reinterpret_cast(tree()); + } + + RegExpCharacterClass* char_class() const { + MOZ_ASSERT(text_type() == CHAR_CLASS); + return reinterpret_cast(tree()); + } + + private: + TextElement(TextType text_type, RegExpTree* tree) + : cp_offset_(-1), text_type_(text_type), tree_(tree) + {} + + int cp_offset_; + TextType text_type_; + RegExpTree* tree_; +}; + +typedef InfallibleVector TextElementVector; + +class NodeVisitor; +class RegExpCompiler; +class Trace; +class BoyerMooreLookahead; + +struct NodeInfo +{ + NodeInfo() + : being_analyzed(false), + been_analyzed(false), + follows_word_interest(false), + follows_newline_interest(false), + follows_start_interest(false), + at_end(false), + visited(false), + replacement_calculated(false) + {} + + // Returns true if the interests and assumptions of this node + // matches the given one. + bool Matches(NodeInfo* that) { + return (at_end == that->at_end) && + (follows_word_interest == that->follows_word_interest) && + (follows_newline_interest == that->follows_newline_interest) && + (follows_start_interest == that->follows_start_interest); + } + + // Updates the interests of this node given the interests of the + // node preceding it. + void AddFromPreceding(NodeInfo* that) { + at_end |= that->at_end; + follows_word_interest |= that->follows_word_interest; + follows_newline_interest |= that->follows_newline_interest; + follows_start_interest |= that->follows_start_interest; + } + + bool HasLookbehind() { + return follows_word_interest || + follows_newline_interest || + follows_start_interest; + } + + // Sets the interests of this node to include the interests of the + // following node. + void AddFromFollowing(NodeInfo* that) { + follows_word_interest |= that->follows_word_interest; + follows_newline_interest |= that->follows_newline_interest; + follows_start_interest |= that->follows_start_interest; + } + + void ResetCompilationState() { + being_analyzed = false; + been_analyzed = false; + } + + bool being_analyzed: 1; + bool been_analyzed: 1; + + // These bits are set of this node has to know what the preceding + // character was. + bool follows_word_interest: 1; + bool follows_newline_interest: 1; + bool follows_start_interest: 1; + + bool at_end: 1; + bool visited: 1; + bool replacement_calculated: 1; +}; + +// Details of a quick mask-compare check that can look ahead in the +// input stream. +class QuickCheckDetails +{ + public: + QuickCheckDetails() + : characters_(0), + mask_(0), + value_(0), + cannot_match_(false) + {} + + explicit QuickCheckDetails(int characters) + : characters_(characters), + mask_(0), + value_(0), + cannot_match_(false) + {} + + bool Rationalize(bool ascii); + + // Merge in the information from another branch of an alternation. + void Merge(QuickCheckDetails* other, int from_index); + + // Advance the current position by some amount. + void Advance(int by, bool ascii); + + void Clear(); + + bool cannot_match() { return cannot_match_; } + void set_cannot_match() { cannot_match_ = true; } + + int characters() { return characters_; } + void set_characters(int characters) { characters_ = characters; } + + struct Position { + Position() : mask(0), value(0), determines_perfectly(false) { } + char16_t mask; + char16_t value; + bool determines_perfectly; + }; + + Position* positions(int index) { + MOZ_ASSERT(index >= 0); + MOZ_ASSERT(index < characters_); + return positions_ + index; + } + + uint32_t mask() { return mask_; } + uint32_t value() { return value_; } + + private: + // How many characters do we have quick check information from. This is + // the same for all branches of a choice node. + int characters_; + Position positions_[4]; + + // These values are the condensate of the above array after Rationalize(). + uint32_t mask_; + uint32_t value_; + + // If set to true, there is no way this quick check can match at all. + // E.g., if it requires to be at the start of the input, and isn't. + bool cannot_match_; +}; + +class RegExpNode +{ + public: + explicit RegExpNode(LifoAlloc* alloc); + virtual ~RegExpNode() {} + virtual void Accept(NodeVisitor* visitor) = 0; + + // Generates a goto to this node or actually generates the code at this point. + virtual void Emit(RegExpCompiler* compiler, Trace* trace) = 0; + + // How many characters must this node consume at a minimum in order to + // succeed. If we have found at least 'still_to_find' characters that + // must be consumed there is no need to ask any following nodes whether + // they are sure to eat any more characters. The not_at_start argument is + // used to indicate that we know we are not at the start of the input. In + // this case anchored branches will always fail and can be ignored when + // determining how many characters are consumed on success. + virtual int EatsAtLeast(int still_to_find, int budget, bool not_at_start) = 0; + + // Emits some quick code that checks whether the preloaded characters match. + // Falls through on certain failure, jumps to the label on possible success. + // If the node cannot make a quick check it does nothing and returns false. + bool EmitQuickCheck(RegExpCompiler* compiler, + Trace* trace, + bool preload_has_checked_bounds, + jit::Label* on_possible_success, + QuickCheckDetails* details_return, + bool fall_through_on_failure); + + // For a given number of characters this returns a mask and a value. The + // next n characters are anded with the mask and compared with the value. + // A comparison failure indicates the node cannot match the next n characters. + // A comparison success indicates the node may match. + virtual void GetQuickCheckDetails(QuickCheckDetails* details, + RegExpCompiler* compiler, + int characters_filled_in, + bool not_at_start) = 0; + + static const int kNodeIsTooComplexForGreedyLoops = -1; + + virtual int GreedyLoopTextLength() { return kNodeIsTooComplexForGreedyLoops; } + + // Only returns the successor for a text node of length 1 that matches any + // character and that has no guards on it. + virtual RegExpNode* GetSuccessorOfOmnivorousTextNode(RegExpCompiler* compiler) { + return nullptr; + } + + static const int kRecursionBudget = 200; + + // Collects information on the possible code units (mod 128) that can match if + // we look forward. This is used for a Boyer-Moore-like string searching + // implementation. TODO(erikcorry): This should share more code with + // EatsAtLeast, GetQuickCheckDetails. The budget argument is used to limit + // the number of nodes we are willing to look at in order to create this data. + virtual bool FillInBMInfo(int offset, + int budget, + BoyerMooreLookahead* bm, + bool not_at_start) { + MOZ_CRASH("Bad call"); + } + + // If we know that the input is ASCII then there are some nodes that can + // never match. This method returns a node that can be substituted for + // itself, or nullptr if the node can never match. + virtual RegExpNode* FilterASCII(int depth, bool ignore_case, bool unicode) { return this; } + + // Helper for FilterASCII. + RegExpNode* replacement() { + MOZ_ASSERT(info()->replacement_calculated); + return replacement_; + } + RegExpNode* set_replacement(RegExpNode* replacement) { + info()->replacement_calculated = true; + replacement_ = replacement; + return replacement; // For convenience. + } + + // We want to avoid recalculating the lookahead info, so we store it on the + // node. Only info that is for this node is stored. We can tell that the + // info is for this node when offset == 0, so the information is calculated + // relative to this node. + void SaveBMInfo(BoyerMooreLookahead* bm, bool not_at_start, int offset) { + if (offset == 0) set_bm_info(not_at_start, bm); + } + + jit::Label* label() { return &label_; } + + // If non-generic code is generated for a node (i.e. the node is not at the + // start of the trace) then it cannot be reused. This variable sets a limit + // on how often we allow that to happen before we insist on starting a new + // trace and generating generic code for a node that can be reused by flushing + // the deferred actions in the current trace and generating a goto. + static const int kMaxCopiesCodeGenerated = 10; + + NodeInfo* info() { return &info_; } + + BoyerMooreLookahead* bm_info(bool not_at_start) { + return bm_info_[not_at_start ? 1 : 0]; + } + + LifoAlloc* alloc() const { return alloc_; } + + protected: + enum LimitResult { DONE, CONTINUE }; + RegExpNode* replacement_; + + LimitResult LimitVersions(RegExpCompiler* compiler, Trace* trace); + + void set_bm_info(bool not_at_start, BoyerMooreLookahead* bm) { + bm_info_[not_at_start ? 1 : 0] = bm; + } + + private: + static const int kFirstCharBudget = 10; + jit::Label label_; + NodeInfo info_; + + // This variable keeps track of how many times code has been generated for + // this node (in different traces). We don't keep track of where the + // generated code is located unless the code is generated at the start of + // a trace, in which case it is generic and can be reused by flushing the + // deferred operations in the current trace and generating a goto. + int trace_count_; + BoyerMooreLookahead* bm_info_[2]; + + LifoAlloc* alloc_; +}; + +// A simple closed interval. +class Interval +{ + public: + Interval() : from_(kNone), to_(kNone) { } + + Interval(int from, int to) : from_(from), to_(to) { } + + Interval Union(Interval that) { + if (that.from_ == kNone) + return *this; + else if (from_ == kNone) + return that; + else + return Interval(Min(from_, that.from_), Max(to_, that.to_)); + } + + bool Contains(int value) { + return (from_ <= value) && (value <= to_); + } + + bool is_empty() { return from_ == kNone; } + + int from() const { return from_; } + int to() const { return to_; } + + static Interval Empty() { return Interval(); } + static const int kNone = -1; + + private: + int from_; + int to_; +}; + +class SeqRegExpNode : public RegExpNode +{ + public: + explicit SeqRegExpNode(RegExpNode* on_success) + : RegExpNode(on_success->alloc()), on_success_(on_success) + {} + + RegExpNode* on_success() { return on_success_; } + void set_on_success(RegExpNode* node) { on_success_ = node; } + virtual RegExpNode* FilterASCII(int depth, bool ignore_case, bool unicode); + virtual bool FillInBMInfo(int offset, + int budget, + BoyerMooreLookahead* bm, + bool not_at_start); + + protected: + RegExpNode* FilterSuccessor(int depth, bool ignore_case, bool unicode); + + private: + RegExpNode* on_success_; +}; + +class ActionNode : public SeqRegExpNode +{ + public: + enum ActionType { + SET_REGISTER, + INCREMENT_REGISTER, + STORE_POSITION, + BEGIN_SUBMATCH, + POSITIVE_SUBMATCH_SUCCESS, + EMPTY_MATCH_CHECK, + CLEAR_CAPTURES + }; + + ActionNode(ActionType action_type, RegExpNode* on_success) + : SeqRegExpNode(on_success), + action_type_(action_type) + {} + + static ActionNode* SetRegister(int reg, int val, RegExpNode* on_success); + static ActionNode* IncrementRegister(int reg, RegExpNode* on_success); + static ActionNode* StorePosition(int reg, + bool is_capture, + RegExpNode* on_success); + static ActionNode* ClearCaptures(Interval range, RegExpNode* on_success); + static ActionNode* BeginSubmatch(int stack_pointer_reg, + int position_reg, + RegExpNode* on_success); + static ActionNode* PositiveSubmatchSuccess(int stack_pointer_reg, + int restore_reg, + int clear_capture_count, + int clear_capture_from, + RegExpNode* on_success); + static ActionNode* EmptyMatchCheck(int start_register, + int repetition_register, + int repetition_limit, + RegExpNode* on_success); + virtual void Accept(NodeVisitor* visitor); + virtual void Emit(RegExpCompiler* compiler, Trace* trace); + virtual int EatsAtLeast(int still_to_find, int budget, bool not_at_start); + virtual void GetQuickCheckDetails(QuickCheckDetails* details, + RegExpCompiler* compiler, + int filled_in, + bool not_at_start) { + return on_success()->GetQuickCheckDetails( + details, compiler, filled_in, not_at_start); + } + virtual bool FillInBMInfo(int offset, + int budget, + BoyerMooreLookahead* bm, + bool not_at_start); + ActionType action_type() { return action_type_; } + // TODO(erikcorry): We should allow some action nodes in greedy loops. + virtual int GreedyLoopTextLength() { return kNodeIsTooComplexForGreedyLoops; } + + private: + union { + struct { + int reg; + int value; + } u_store_register; + struct { + int reg; + } u_increment_register; + struct { + int reg; + bool is_capture; + } u_position_register; + struct { + int stack_pointer_register; + int current_position_register; + int clear_register_count; + int clear_register_from; + } u_submatch; + struct { + int start_register; + int repetition_register; + int repetition_limit; + } u_empty_match_check; + struct { + int range_from; + int range_to; + } u_clear_captures; + } data_; + ActionType action_type_; + friend class DotPrinter; +}; + +class TextNode : public SeqRegExpNode +{ + public: + TextNode(TextElementVector* elements, + RegExpNode* on_success) + : SeqRegExpNode(on_success), + elements_(elements) + {} + + TextNode(RegExpCharacterClass* that, + RegExpNode* on_success) + : SeqRegExpNode(on_success), + elements_(alloc()->newInfallible(*alloc())) + { + elements_->append(TextElement::CharClass(that)); + } + + virtual void Accept(NodeVisitor* visitor); + virtual void Emit(RegExpCompiler* compiler, Trace* trace); + virtual int EatsAtLeast(int still_to_find, int budget, bool not_at_start); + virtual void GetQuickCheckDetails(QuickCheckDetails* details, + RegExpCompiler* compiler, + int characters_filled_in, + bool not_at_start); + TextElementVector& elements() { return *elements_; } + void MakeCaseIndependent(bool is_ascii, bool unicode); + virtual int GreedyLoopTextLength(); + virtual RegExpNode* GetSuccessorOfOmnivorousTextNode( + RegExpCompiler* compiler); + virtual bool FillInBMInfo(int offset, + int budget, + BoyerMooreLookahead* bm, + bool not_at_start); + void CalculateOffsets(); + virtual RegExpNode* FilterASCII(int depth, bool ignore_case, bool unicode); + + private: + enum TextEmitPassType { + NON_ASCII_MATCH, // Check for characters that can't match. + SIMPLE_CHARACTER_MATCH, // Case-dependent single character check. + NON_LETTER_CHARACTER_MATCH, // Check characters that have no case equivs. + CASE_CHARACTER_MATCH, // Case-independent single character check. + CHARACTER_CLASS_MATCH // Character class. + }; + static bool SkipPass(int pass, bool ignore_case); + static const int kFirstRealPass = SIMPLE_CHARACTER_MATCH; + static const int kLastPass = CHARACTER_CLASS_MATCH; + void TextEmitPass(RegExpCompiler* compiler, + TextEmitPassType pass, + bool preloaded, + Trace* trace, + bool first_element_checked, + int* checked_up_to); + int Length(); + TextElementVector* elements_; +}; + +class AssertionNode : public SeqRegExpNode +{ + public: + enum AssertionType { + AT_END, + AT_START, + AT_BOUNDARY, + AT_NON_BOUNDARY, + AFTER_NEWLINE, + NOT_AFTER_LEAD_SURROGATE, + NOT_IN_SURROGATE_PAIR + }; + AssertionNode(AssertionType t, RegExpNode* on_success) + : SeqRegExpNode(on_success), assertion_type_(t) + {} + + static AssertionNode* AtEnd(RegExpNode* on_success) { + return on_success->alloc()->newInfallible(AT_END, on_success); + } + static AssertionNode* AtStart(RegExpNode* on_success) { + return on_success->alloc()->newInfallible(AT_START, on_success); + } + static AssertionNode* AtBoundary(RegExpNode* on_success) { + return on_success->alloc()->newInfallible(AT_BOUNDARY, on_success); + } + static AssertionNode* AtNonBoundary(RegExpNode* on_success) { + return on_success->alloc()->newInfallible(AT_NON_BOUNDARY, on_success); + } + static AssertionNode* AfterNewline(RegExpNode* on_success) { + return on_success->alloc()->newInfallible(AFTER_NEWLINE, on_success); + } + static AssertionNode* NotAfterLeadSurrogate(RegExpNode* on_success) { + return on_success->alloc()->newInfallible(NOT_AFTER_LEAD_SURROGATE, + on_success); + } + static AssertionNode* NotInSurrogatePair(RegExpNode* on_success) { + return on_success->alloc()->newInfallible(NOT_IN_SURROGATE_PAIR, + on_success); + } + virtual void Accept(NodeVisitor* visitor); + virtual void Emit(RegExpCompiler* compiler, Trace* trace); + virtual int EatsAtLeast(int still_to_find, int budget, bool not_at_start); + virtual void GetQuickCheckDetails(QuickCheckDetails* details, + RegExpCompiler* compiler, + int filled_in, + bool not_at_start); + virtual bool FillInBMInfo(int offset, + int budget, + BoyerMooreLookahead* bm, + bool not_at_start); + AssertionType assertion_type() { return assertion_type_; } + + private: + void EmitBoundaryCheck(RegExpCompiler* compiler, Trace* trace); + enum IfPrevious { kIsNonWord, kIsWord }; + void BacktrackIfPrevious(RegExpCompiler* compiler, + Trace* trace, + IfPrevious backtrack_if_previous); + AssertionType assertion_type_; +}; + +class BackReferenceNode : public SeqRegExpNode +{ + public: + BackReferenceNode(int start_reg, + int end_reg, + RegExpNode* on_success) + : SeqRegExpNode(on_success), + start_reg_(start_reg), + end_reg_(end_reg) + {} + + virtual void Accept(NodeVisitor* visitor); + int start_register() { return start_reg_; } + int end_register() { return end_reg_; } + virtual void Emit(RegExpCompiler* compiler, Trace* trace); + virtual int EatsAtLeast(int still_to_find, + int recursion_depth, + bool not_at_start); + virtual void GetQuickCheckDetails(QuickCheckDetails* details, + RegExpCompiler* compiler, + int characters_filled_in, + bool not_at_start) { + return; + } + virtual bool FillInBMInfo(int offset, + int budget, + BoyerMooreLookahead* bm, + bool not_at_start); + + private: + int start_reg_; + int end_reg_; +}; + +class EndNode : public RegExpNode +{ + public: + enum Action { ACCEPT, BACKTRACK, NEGATIVE_SUBMATCH_SUCCESS }; + + explicit EndNode(LifoAlloc* alloc, Action action) + : RegExpNode(alloc), action_(action) + {} + + virtual void Accept(NodeVisitor* visitor); + virtual void Emit(RegExpCompiler* compiler, Trace* trace); + virtual int EatsAtLeast(int still_to_find, + int recursion_depth, + bool not_at_start) { return 0; } + virtual void GetQuickCheckDetails(QuickCheckDetails* details, + RegExpCompiler* compiler, + int characters_filled_in, + bool not_at_start) + { + // Returning 0 from EatsAtLeast should ensure we never get here. + MOZ_CRASH("Bad call"); + } + virtual bool FillInBMInfo(int offset, + int budget, + BoyerMooreLookahead* bm, + bool not_at_start) { + // Returning 0 from EatsAtLeast should ensure we never get here. + MOZ_CRASH("Bad call"); + } + + private: + Action action_; +}; + +class NegativeSubmatchSuccess : public EndNode +{ + public: + NegativeSubmatchSuccess(LifoAlloc* alloc, + int stack_pointer_reg, + int position_reg, + int clear_capture_count, + int clear_capture_start) + : EndNode(alloc, NEGATIVE_SUBMATCH_SUCCESS), + stack_pointer_register_(stack_pointer_reg), + current_position_register_(position_reg), + clear_capture_count_(clear_capture_count), + clear_capture_start_(clear_capture_start) + {} + + virtual void Emit(RegExpCompiler* compiler, Trace* trace); + + private: + int stack_pointer_register_; + int current_position_register_; + int clear_capture_count_; + int clear_capture_start_; +}; + +class Guard +{ + public: + enum Relation { LT, GEQ }; + Guard(int reg, Relation op, int value) + : reg_(reg), + op_(op), + value_(value) + {} + + int reg() { return reg_; } + Relation op() { return op_; } + int value() { return value_; } + + private: + int reg_; + Relation op_; + int value_; +}; + +typedef InfallibleVector GuardVector; + +class GuardedAlternative +{ + public: + explicit GuardedAlternative(RegExpNode* node) + : node_(node), guards_(nullptr) + {} + + void AddGuard(LifoAlloc* alloc, Guard* guard); + RegExpNode* node() const { return node_; } + void set_node(RegExpNode* node) { node_ = node; } + const GuardVector* guards() const { return guards_; } + + private: + RegExpNode* node_; + GuardVector* guards_; +}; + +typedef InfallibleVector GuardedAlternativeVector; + +class AlternativeGeneration; + +class ChoiceNode : public RegExpNode +{ + public: + explicit ChoiceNode(LifoAlloc* alloc, int expected_size) + : RegExpNode(alloc), + alternatives_(*alloc), + table_(nullptr), + not_at_start_(false), + being_calculated_(false) + { + alternatives_.reserve(expected_size); + } + + virtual void Accept(NodeVisitor* visitor); + void AddAlternative(GuardedAlternative node) { + alternatives_.append(node); + } + + GuardedAlternativeVector& alternatives() { return alternatives_; } + DispatchTable* GetTable(bool ignore_case); + virtual void Emit(RegExpCompiler* compiler, Trace* trace); + virtual int EatsAtLeast(int still_to_find, int budget, bool not_at_start); + int EatsAtLeastHelper(int still_to_find, + int budget, + RegExpNode* ignore_this_node, + bool not_at_start); + virtual void GetQuickCheckDetails(QuickCheckDetails* details, + RegExpCompiler* compiler, + int characters_filled_in, + bool not_at_start); + virtual bool FillInBMInfo(int offset, + int budget, + BoyerMooreLookahead* bm, + bool not_at_start); + + bool being_calculated() { return being_calculated_; } + bool not_at_start() { return not_at_start_; } + void set_not_at_start() { not_at_start_ = true; } + void set_being_calculated(bool b) { being_calculated_ = b; } + virtual bool try_to_emit_quick_check_for_alternative(int i) { return true; } + virtual RegExpNode* FilterASCII(int depth, bool ignore_case, bool unicode); + + protected: + int GreedyLoopTextLengthForAlternative(GuardedAlternative* alternative); + GuardedAlternativeVector alternatives_; + + private: + friend class Analysis; + void GenerateGuard(RegExpMacroAssembler* macro_assembler, + Guard* guard, + Trace* trace); + int CalculatePreloadCharacters(RegExpCompiler* compiler, int eats_at_least); + void EmitOutOfLineContinuation(RegExpCompiler* compiler, + Trace* trace, + GuardedAlternative alternative, + AlternativeGeneration* alt_gen, + int preload_characters, + bool next_expects_preload); + DispatchTable* table_; + + // If true, this node is never checked at the start of the input. + // Allows a new trace to start with at_start() set to false. + bool not_at_start_; + bool being_calculated_; +}; + +class NegativeLookaheadChoiceNode : public ChoiceNode +{ + public: + explicit NegativeLookaheadChoiceNode(LifoAlloc* alloc, + GuardedAlternative this_must_fail, + GuardedAlternative then_do_this) + : ChoiceNode(alloc, 2) + { + AddAlternative(this_must_fail); + AddAlternative(then_do_this); + } + virtual int EatsAtLeast(int still_to_find, int budget, bool not_at_start); + virtual void GetQuickCheckDetails(QuickCheckDetails* details, + RegExpCompiler* compiler, + int characters_filled_in, + bool not_at_start); + virtual bool FillInBMInfo(int offset, + int budget, + BoyerMooreLookahead* bm, + bool not_at_start); + + // For a negative lookahead we don't emit the quick check for the + // alternative that is expected to fail. This is because quick check code + // starts by loading enough characters for the alternative that takes fewest + // characters, but on a negative lookahead the negative branch did not take + // part in that calculation (EatsAtLeast) so the assumptions don't hold. + virtual bool try_to_emit_quick_check_for_alternative(int i) { return i != 0; } + virtual RegExpNode* FilterASCII(int depth, bool ignore_case, bool unicode); +}; + +class LoopChoiceNode : public ChoiceNode +{ + public: + explicit LoopChoiceNode(LifoAlloc* alloc, bool body_can_be_zero_length) + : ChoiceNode(alloc, 2), + loop_node_(nullptr), + continue_node_(nullptr), + body_can_be_zero_length_(body_can_be_zero_length) + {} + + void AddLoopAlternative(GuardedAlternative alt); + void AddContinueAlternative(GuardedAlternative alt); + virtual void Emit(RegExpCompiler* compiler, Trace* trace); + virtual int EatsAtLeast(int still_to_find, int budget, bool not_at_start); + virtual void GetQuickCheckDetails(QuickCheckDetails* details, + RegExpCompiler* compiler, + int characters_filled_in, + bool not_at_start); + virtual bool FillInBMInfo(int offset, + int budget, + BoyerMooreLookahead* bm, + bool not_at_start); + RegExpNode* loop_node() { return loop_node_; } + RegExpNode* continue_node() { return continue_node_; } + bool body_can_be_zero_length() { return body_can_be_zero_length_; } + virtual void Accept(NodeVisitor* visitor); + virtual RegExpNode* FilterASCII(int depth, bool ignore_case, bool unicode); + + private: + // AddAlternative is made private for loop nodes because alternatives + // should not be added freely, we need to keep track of which node + // goes back to the node itself. + void AddAlternative(GuardedAlternative node) { + ChoiceNode::AddAlternative(node); + } + + RegExpNode* loop_node_; + RegExpNode* continue_node_; + bool body_can_be_zero_length_; +}; + +// Improve the speed that we scan for an initial point where a non-anchored +// regexp can match by using a Boyer-Moore-like table. This is done by +// identifying non-greedy non-capturing loops in the nodes that eat any +// character one at a time. For example in the middle of the regexp +// /foo[\s\S]*?bar/ we find such a loop. There is also such a loop implicitly +// inserted at the start of any non-anchored regexp. +// +// When we have found such a loop we look ahead in the nodes to find the set of +// characters that can come at given distances. For example for the regexp +// /.?foo/ we know that there are at least 3 characters ahead of us, and the +// sets of characters that can occur are [any, [f, o], [o]]. We find a range in +// the lookahead info where the set of characters is reasonably constrained. In +// our example this is from index 1 to 2 (0 is not constrained). We can now +// look 3 characters ahead and if we don't find one of [f, o] (the union of +// [f, o] and [o]) then we can skip forwards by the range size (in this case 2). +// +// For Unicode input strings we do the same, but modulo 128. +// +// We also look at the first string fed to the regexp and use that to get a hint +// of the character frequencies in the inputs. This affects the assessment of +// whether the set of characters is 'reasonably constrained'. +// +// We also have another lookahead mechanism (called quick check in the code), +// which uses a wide load of multiple characters followed by a mask and compare +// to determine whether a match is possible at this point. +enum ContainedInLattice { + kNotYet = 0, + kLatticeIn = 1, + kLatticeOut = 2, + kLatticeUnknown = 3 // Can also mean both in and out. +}; + +inline ContainedInLattice +Combine(ContainedInLattice a, ContainedInLattice b) { + return static_cast(a | b); +} + +ContainedInLattice +AddRange(ContainedInLattice a, + const int* ranges, + int ranges_size, + Interval new_range); + +class BoyerMoorePositionInfo +{ + public: + explicit BoyerMoorePositionInfo(LifoAlloc* alloc) + : map_(*alloc), + map_count_(0), + w_(kNotYet), + s_(kNotYet), + d_(kNotYet), + surrogate_(kNotYet) + { + map_.reserve(kMapSize); + for (int i = 0; i < kMapSize; i++) + map_.append(false); + } + + bool& at(int i) { return map_[i]; } + + static const int kMapSize = 128; + static const int kMask = kMapSize - 1; + + int map_count() const { return map_count_; } + + void Set(int character); + void SetInterval(const Interval& interval); + void SetAll(); + bool is_non_word() { return w_ == kLatticeOut; } + bool is_word() { return w_ == kLatticeIn; } + + private: + InfallibleVector map_; + int map_count_; // Number of set bits in the map. + ContainedInLattice w_; // The \w character class. + ContainedInLattice s_; // The \s character class. + ContainedInLattice d_; // The \d character class. + ContainedInLattice surrogate_; // Surrogate UTF-16 code units. +}; + +typedef InfallibleVector BoyerMoorePositionInfoVector; + +class BoyerMooreLookahead +{ + public: + BoyerMooreLookahead(LifoAlloc* alloc, size_t length, RegExpCompiler* compiler); + + int length() { return length_; } + int max_char() { return max_char_; } + RegExpCompiler* compiler() { return compiler_; } + + int Count(int map_number) { + return bitmaps_[map_number]->map_count(); + } + + BoyerMoorePositionInfo* at(int i) { return bitmaps_[i]; } + + void Set(int map_number, int character) { + if (character > max_char_) return; + BoyerMoorePositionInfo* info = bitmaps_[map_number]; + info->Set(character); + } + + void SetInterval(int map_number, const Interval& interval) { + if (interval.from() > max_char_) return; + BoyerMoorePositionInfo* info = bitmaps_[map_number]; + if (interval.to() > max_char_) { + info->SetInterval(Interval(interval.from(), max_char_)); + } else { + info->SetInterval(interval); + } + } + + void SetAll(int map_number) { + bitmaps_[map_number]->SetAll(); + } + + void SetRest(int from_map) { + for (int i = from_map; i < length_; i++) SetAll(i); + } + bool EmitSkipInstructions(RegExpMacroAssembler* masm); + + bool CheckOverRecursed(); + + private: + // This is the value obtained by EatsAtLeast. If we do not have at least this + // many characters left in the sample string then the match is bound to fail. + // Therefore it is OK to read a character this far ahead of the current match + // point. + int length_; + RegExpCompiler* compiler_; + + // 0x7f for ASCII, 0xffff for UTF-16. + int max_char_; + BoyerMoorePositionInfoVector bitmaps_; + + int GetSkipTable(int min_lookahead, + int max_lookahead, + uint8_t* boolean_skip_table); + bool FindWorthwhileInterval(int* from, int* to); + int FindBestInterval(int max_number_of_chars, int old_biggest_points, int* from, int* to); +}; + +// There are many ways to generate code for a node. This class encapsulates +// the current way we should be generating. In other words it encapsulates +// the current state of the code generator. The effect of this is that we +// generate code for paths that the matcher can take through the regular +// expression. A given node in the regexp can be code-generated several times +// as it can be part of several traces. For example for the regexp: +// /foo(bar|ip)baz/ the code to match baz will be generated twice, once as part +// of the foo-bar-baz trace and once as part of the foo-ip-baz trace. The code +// to match foo is generated only once (the traces have a common prefix). The +// code to store the capture is deferred and generated (twice) after the places +// where baz has been matched. +class Trace +{ + public: + // A value for a property that is either known to be true, know to be false, + // or not known. + enum TriBool { + UNKNOWN = -1, FALSE_VALUE = 0, TRUE_VALUE = 1 + }; + + class DeferredAction { + public: + DeferredAction(ActionNode::ActionType action_type, int reg) + : action_type_(action_type), reg_(reg), next_(nullptr) + {} + + DeferredAction* next() { return next_; } + bool Mentions(int reg); + int reg() { return reg_; } + ActionNode::ActionType action_type() { return action_type_; } + private: + ActionNode::ActionType action_type_; + int reg_; + DeferredAction* next_; + friend class Trace; + }; + + class DeferredCapture : public DeferredAction { + public: + DeferredCapture(int reg, bool is_capture, Trace* trace) + : DeferredAction(ActionNode::STORE_POSITION, reg), + cp_offset_(trace->cp_offset()), + is_capture_(is_capture) + {} + + int cp_offset() { return cp_offset_; } + bool is_capture() { return is_capture_; } + private: + int cp_offset_; + bool is_capture_; + void set_cp_offset(int cp_offset) { cp_offset_ = cp_offset; } + }; + + class DeferredSetRegister : public DeferredAction { + public: + DeferredSetRegister(int reg, int value) + : DeferredAction(ActionNode::SET_REGISTER, reg), + value_(value) + {} + int value() { return value_; } + private: + int value_; + }; + + class DeferredClearCaptures : public DeferredAction { + public: + explicit DeferredClearCaptures(Interval range) + : DeferredAction(ActionNode::CLEAR_CAPTURES, -1), + range_(range) + {} + + Interval range() { return range_; } + private: + Interval range_; + }; + + class DeferredIncrementRegister : public DeferredAction { + public: + explicit DeferredIncrementRegister(int reg) + : DeferredAction(ActionNode::INCREMENT_REGISTER, reg) + {} + }; + + Trace() + : cp_offset_(0), + actions_(nullptr), + backtrack_(nullptr), + stop_node_(nullptr), + loop_label_(nullptr), + characters_preloaded_(0), + bound_checked_up_to_(0), + flush_budget_(100), + at_start_(UNKNOWN) + {} + + // End the trace. This involves flushing the deferred actions in the trace + // and pushing a backtrack location onto the backtrack stack. Once this is + // done we can start a new trace or go to one that has already been + // generated. + void Flush(RegExpCompiler* compiler, RegExpNode* successor); + + int cp_offset() { return cp_offset_; } + DeferredAction* actions() { return actions_; } + + // A trivial trace is one that has no deferred actions or other state that + // affects the assumptions used when generating code. There is no recorded + // backtrack location in a trivial trace, so with a trivial trace we will + // generate code that, on a failure to match, gets the backtrack location + // from the backtrack stack rather than using a direct jump instruction. We + // always start code generation with a trivial trace and non-trivial traces + // are created as we emit code for nodes or add to the list of deferred + // actions in the trace. The location of the code generated for a node using + // a trivial trace is recorded in a label in the node so that gotos can be + // generated to that code. + bool is_trivial() { + return backtrack_ == nullptr && + actions_ == nullptr && + cp_offset_ == 0 && + characters_preloaded_ == 0 && + bound_checked_up_to_ == 0 && + quick_check_performed_.characters() == 0 && + at_start_ == UNKNOWN; + } + + TriBool at_start() { return at_start_; } + void set_at_start(bool at_start) { + at_start_ = at_start ? TRUE_VALUE : FALSE_VALUE; + } + jit::Label* backtrack() { return backtrack_; } + jit::Label* loop_label() { return loop_label_; } + RegExpNode* stop_node() { return stop_node_; } + int characters_preloaded() { return characters_preloaded_; } + int bound_checked_up_to() { return bound_checked_up_to_; } + int flush_budget() { return flush_budget_; } + QuickCheckDetails* quick_check_performed() { return &quick_check_performed_; } + bool mentions_reg(int reg); + + // Returns true if a deferred position store exists to the specified + // register and stores the offset in the out-parameter. Otherwise + // returns false. + bool GetStoredPosition(int reg, int* cp_offset); + + // These set methods and AdvanceCurrentPositionInTrace should be used only on + // new traces - the intention is that traces are immutable after creation. + void add_action(DeferredAction* new_action) { + MOZ_ASSERT(new_action->next_ == nullptr); + new_action->next_ = actions_; + actions_ = new_action; + } + + void set_backtrack(jit::Label* backtrack) { backtrack_ = backtrack; } + void set_stop_node(RegExpNode* node) { stop_node_ = node; } + void set_loop_label(jit::Label* label) { loop_label_ = label; } + void set_characters_preloaded(int count) { characters_preloaded_ = count; } + void set_bound_checked_up_to(int to) { bound_checked_up_to_ = to; } + void set_flush_budget(int to) { flush_budget_ = to; } + void set_quick_check_performed(QuickCheckDetails* d) { + quick_check_performed_ = *d; + } + void InvalidateCurrentCharacter(); + void AdvanceCurrentPositionInTrace(int by, RegExpCompiler* compiler); + + private: + int FindAffectedRegisters(LifoAlloc* alloc, OutSet* affected_registers); + void PerformDeferredActions(LifoAlloc* alloc, + RegExpMacroAssembler* macro, + int max_register, + OutSet& affected_registers, + OutSet* registers_to_pop, + OutSet* registers_to_clear); + void RestoreAffectedRegisters(RegExpMacroAssembler* macro, + int max_register, + OutSet& registers_to_pop, + OutSet& registers_to_clear); + int cp_offset_; + DeferredAction* actions_; + jit::Label* backtrack_; + RegExpNode* stop_node_; + jit::Label* loop_label_; + int characters_preloaded_; + int bound_checked_up_to_; + QuickCheckDetails quick_check_performed_; + int flush_budget_; + TriBool at_start_; +}; + +class NodeVisitor +{ + public: + virtual ~NodeVisitor() { } +#define DECLARE_VISIT(Type) \ + virtual void Visit##Type(Type##Node* that) = 0; + FOR_EACH_NODE_TYPE(DECLARE_VISIT) +#undef DECLARE_VISIT + virtual void VisitLoopChoice(LoopChoiceNode* that) { VisitChoice(that); } +}; + +// 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 Analysis : public NodeVisitor +{ + public: + Analysis(JSContext* cx, bool ignore_case, bool is_ascii, bool unicode) + : cx(cx), + ignore_case_(ignore_case), + is_ascii_(is_ascii), + unicode_(unicode), + error_message_(nullptr) + {} + + void EnsureAnalyzed(RegExpNode* node); + +#define DECLARE_VISIT(Type) \ + virtual void Visit##Type(Type##Node* that); + FOR_EACH_NODE_TYPE(DECLARE_VISIT) +#undef DECLARE_VISIT + virtual void VisitLoopChoice(LoopChoiceNode* that); + + bool has_failed() { return error_message_ != nullptr; } + const char* errorMessage() { + MOZ_ASSERT(error_message_ != nullptr); + return error_message_; + } + void failASCII(const char* error_message) { + error_message_ = error_message; + } + + private: + JSContext* cx; + bool ignore_case_; + bool is_ascii_; + bool unicode_; + const char* error_message_; + + Analysis(Analysis&) = delete; + void operator=(Analysis&) = delete; +}; + +} } // namespace js::irregexp + +#endif // V8_JSREGEXP_H_ diff --git a/js/src/irregexp/RegExpInterpreter.cpp b/js/src/irregexp/RegExpInterpreter.cpp new file mode 100644 index 000000000..7fd2d983a --- /dev/null +++ b/js/src/irregexp/RegExpInterpreter.cpp @@ -0,0 +1,501 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- + * vim: set ts=8 sts=4 et sw=4 tw=99: */ + +// Copyright 2011 the V8 project authors. All rights reserved. +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived +// from this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +// A simple interpreter for the Irregexp byte code. + +#include "irregexp/RegExpBytecode.h" +#include "irregexp/RegExpMacroAssembler.h" +#include "vm/MatchPairs.h" + +using namespace js; +using namespace js::irregexp; + +static const size_t kBitsPerByte = 8; +static const size_t kBitsPerByteLog2 = 3; + +class MOZ_STACK_CLASS RegExpStackCursor +{ + public: + explicit RegExpStackCursor(JSContext* cx) + : cx(cx), cursor(nullptr) + {} + + bool init() { + if (!stack.init()) { + ReportOutOfMemory(cx); + return false; + } + cursor = base(); + return true; + } + + bool push(int32_t value) { + *cursor++ = value; + if (cursor >= stack.limit()) { + int32_t pos = position(); + if (!stack.grow()) { + ReportOverRecursed(cx); + return false; + } + setPosition(pos); + } + return true; + } + + int32_t pop() { + MOZ_ASSERT(cursor > base()); + return *--cursor; + } + + int32_t peek() { + MOZ_ASSERT(cursor > base()); + return *(cursor - 1); + } + + int32_t position() { + MOZ_ASSERT(cursor >= base()); + return cursor - base(); + } + + void setPosition(int32_t position) { + cursor = base() + position; + MOZ_ASSERT(cursor < stack.limit()); + } + + private: + JSContext* cx; + RegExpStack stack; + + int32_t* cursor; + + int32_t* base() { return (int32_t*) stack.base(); } +}; + +static int32_t +Load32Aligned(const uint8_t* pc) +{ + MOZ_ASSERT((reinterpret_cast(pc) & 3) == 0); + return *reinterpret_cast(pc); +} + +static int32_t +Load16Aligned(const uint8_t* pc) +{ + MOZ_ASSERT((reinterpret_cast(pc) & 1) == 0); + return *reinterpret_cast(pc); +} + +#define BYTECODE(name) case BC_##name: + +template +RegExpRunStatus +irregexp::InterpretCode(JSContext* cx, const uint8_t* byteCode, const CharT* chars, size_t current, + size_t length, MatchPairs* matches, size_t* endIndex) +{ + const uint8_t* pc = byteCode; + + uint32_t current_char = current ? chars[current - 1] : '\n'; + + RegExpStackCursor stack(cx); + + if (!stack.init()) + return RegExpRunStatus_Error; + + int32_t numRegisters = Load32Aligned(pc); + pc += 4; + + Vector registers; + if (!registers.growByUninitialized(numRegisters)) + return RegExpRunStatus_Error; + for (size_t i = 0; i < (size_t) numRegisters; i++) + registers[i] = -1; + + while (true) { + int32_t insn = Load32Aligned(pc); + switch (insn & BYTECODE_MASK) { + BYTECODE(BREAK) + MOZ_CRASH("Bad bytecode: BREAK"); + BYTECODE(PUSH_CP) + if (!stack.push(current)) + return RegExpRunStatus_Error; + pc += BC_PUSH_CP_LENGTH; + break; + BYTECODE(PUSH_BT) + if (!stack.push(Load32Aligned(pc + 4))) + return RegExpRunStatus_Error; + pc += BC_PUSH_BT_LENGTH; + break; + BYTECODE(PUSH_REGISTER) + if (!stack.push(registers[insn >> BYTECODE_SHIFT])) + return RegExpRunStatus_Error; + pc += BC_PUSH_REGISTER_LENGTH; + break; + BYTECODE(SET_REGISTER) + registers[insn >> BYTECODE_SHIFT] = Load32Aligned(pc + 4); + pc += BC_SET_REGISTER_LENGTH; + break; + BYTECODE(ADVANCE_REGISTER) + registers[insn >> BYTECODE_SHIFT] += Load32Aligned(pc + 4); + pc += BC_ADVANCE_REGISTER_LENGTH; + break; + BYTECODE(SET_REGISTER_TO_CP) + registers[insn >> BYTECODE_SHIFT] = current + Load32Aligned(pc + 4); + pc += BC_SET_REGISTER_TO_CP_LENGTH; + break; + BYTECODE(SET_CP_TO_REGISTER) + current = registers[insn >> BYTECODE_SHIFT]; + pc += BC_SET_CP_TO_REGISTER_LENGTH; + break; + BYTECODE(SET_REGISTER_TO_SP) + registers[insn >> BYTECODE_SHIFT] = stack.position(); + pc += BC_SET_REGISTER_TO_SP_LENGTH; + break; + BYTECODE(SET_SP_TO_REGISTER) + stack.setPosition(registers[insn >> BYTECODE_SHIFT]); + pc += BC_SET_SP_TO_REGISTER_LENGTH; + break; + BYTECODE(POP_CP) + current = stack.pop(); + pc += BC_POP_CP_LENGTH; + break; + BYTECODE(POP_BT) + if (!CheckForInterrupt(cx)) + return RegExpRunStatus_Error; + pc = byteCode + stack.pop(); + break; + BYTECODE(POP_REGISTER) + registers[insn >> BYTECODE_SHIFT] = stack.pop(); + pc += BC_POP_REGISTER_LENGTH; + break; + BYTECODE(FAIL) + return RegExpRunStatus_Success_NotFound; + BYTECODE(SUCCEED) + if (matches) + memcpy(matches->pairsRaw(), registers.begin(), matches->length() * 2 * sizeof(int32_t)); + else if (endIndex) + *endIndex = registers[1]; + return RegExpRunStatus_Success; + BYTECODE(ADVANCE_CP) + current += insn >> BYTECODE_SHIFT; + pc += BC_ADVANCE_CP_LENGTH; + break; + BYTECODE(GOTO) + pc = byteCode + Load32Aligned(pc + 4); + break; + BYTECODE(ADVANCE_CP_AND_GOTO) + current += insn >> BYTECODE_SHIFT; + pc = byteCode + Load32Aligned(pc + 4); + break; + BYTECODE(CHECK_GREEDY) + if ((int32_t)current == stack.peek()) { + stack.pop(); + pc = byteCode + Load32Aligned(pc + 4); + } else { + pc += BC_CHECK_GREEDY_LENGTH; + } + break; + BYTECODE(LOAD_CURRENT_CHAR) { + size_t pos = current + (insn >> BYTECODE_SHIFT); + if (pos >= length) { + pc = byteCode + Load32Aligned(pc + 4); + } else { + current_char = chars[pos]; + pc += BC_LOAD_CURRENT_CHAR_LENGTH; + } + break; + } + BYTECODE(LOAD_CURRENT_CHAR_UNCHECKED) { + int pos = current + (insn >> BYTECODE_SHIFT); + current_char = chars[pos]; + pc += BC_LOAD_CURRENT_CHAR_UNCHECKED_LENGTH; + break; + } + BYTECODE(LOAD_2_CURRENT_CHARS) { + size_t pos = current + (insn >> BYTECODE_SHIFT); + if (pos + 2 > length) { + pc = byteCode + Load32Aligned(pc + 4); + } else { + CharT next = chars[pos + 1]; + current_char = (chars[pos] | (next << (kBitsPerByte * sizeof(CharT)))); + pc += BC_LOAD_2_CURRENT_CHARS_LENGTH; + } + break; + } + BYTECODE(LOAD_2_CURRENT_CHARS_UNCHECKED) { + int pos = current + (insn >> BYTECODE_SHIFT); + char16_t next = chars[pos + 1]; + current_char = (chars[pos] | (next << (kBitsPerByte * sizeof(char16_t)))); + pc += BC_LOAD_2_CURRENT_CHARS_UNCHECKED_LENGTH; + break; + } + BYTECODE(LOAD_4_CURRENT_CHARS) + MOZ_CRASH("ASCII handling implemented"); + BYTECODE(LOAD_4_CURRENT_CHARS_UNCHECKED) + MOZ_CRASH("ASCII handling implemented"); + BYTECODE(CHECK_4_CHARS) { + uint32_t c = Load32Aligned(pc + 4); + if (c == current_char) + pc = byteCode + Load32Aligned(pc + 8); + else + pc += BC_CHECK_4_CHARS_LENGTH; + break; + } + BYTECODE(CHECK_CHAR) { + uint32_t c = (insn >> BYTECODE_SHIFT); + if (c == current_char) + pc = byteCode + Load32Aligned(pc + 4); + else + pc += BC_CHECK_CHAR_LENGTH; + break; + } + BYTECODE(CHECK_NOT_4_CHARS) { + uint32_t c = Load32Aligned(pc + 4); + if (c != current_char) + pc = byteCode + Load32Aligned(pc + 8); + else + pc += BC_CHECK_NOT_4_CHARS_LENGTH; + break; + } + BYTECODE(CHECK_NOT_CHAR) { + uint32_t c = (insn >> BYTECODE_SHIFT); + if (c != current_char) + pc = byteCode + Load32Aligned(pc + 4); + else + pc += BC_CHECK_NOT_CHAR_LENGTH; + break; + } + BYTECODE(AND_CHECK_4_CHARS) { + uint32_t c = Load32Aligned(pc + 4); + if (c == (current_char & Load32Aligned(pc + 8))) + pc = byteCode + Load32Aligned(pc + 12); + else + pc += BC_AND_CHECK_4_CHARS_LENGTH; + break; + } + BYTECODE(AND_CHECK_CHAR) { + uint32_t c = (insn >> BYTECODE_SHIFT); + if (c == (current_char & Load32Aligned(pc + 4))) + pc = byteCode + Load32Aligned(pc + 8); + else + pc += BC_AND_CHECK_CHAR_LENGTH; + break; + } + BYTECODE(AND_CHECK_NOT_4_CHARS) { + uint32_t c = Load32Aligned(pc + 4); + if (c != (current_char & Load32Aligned(pc + 8))) + pc = byteCode + Load32Aligned(pc + 12); + else + pc += BC_AND_CHECK_NOT_4_CHARS_LENGTH; + break; + } + BYTECODE(AND_CHECK_NOT_CHAR) { + uint32_t c = (insn >> BYTECODE_SHIFT); + if (c != (current_char & Load32Aligned(pc + 4))) + pc = byteCode + Load32Aligned(pc + 8); + else + pc += BC_AND_CHECK_NOT_CHAR_LENGTH; + break; + } + BYTECODE(MINUS_AND_CHECK_NOT_CHAR) { + uint32_t c = (insn >> BYTECODE_SHIFT); + uint32_t minus = Load16Aligned(pc + 4); + uint32_t mask = Load16Aligned(pc + 6); + if (c != ((current_char - minus) & mask)) + pc = byteCode + Load32Aligned(pc + 8); + else + pc += BC_MINUS_AND_CHECK_NOT_CHAR_LENGTH; + break; + } + BYTECODE(CHECK_CHAR_IN_RANGE) { + uint32_t from = Load16Aligned(pc + 4); + uint32_t to = Load16Aligned(pc + 6); + if (from <= current_char && current_char <= to) + pc = byteCode + Load32Aligned(pc + 8); + else + pc += BC_CHECK_CHAR_IN_RANGE_LENGTH; + break; + } + BYTECODE(CHECK_CHAR_NOT_IN_RANGE) { + uint32_t from = Load16Aligned(pc + 4); + uint32_t to = Load16Aligned(pc + 6); + if (from > current_char || current_char > to) + pc = byteCode + Load32Aligned(pc + 8); + else + pc += BC_CHECK_CHAR_NOT_IN_RANGE_LENGTH; + break; + } + BYTECODE(CHECK_BIT_IN_TABLE) { + int mask = RegExpMacroAssembler::kTableMask; + uint8_t b = pc[8 + ((current_char & mask) >> kBitsPerByteLog2)]; + int bit = (current_char & (kBitsPerByte - 1)); + if ((b & (1 << bit)) != 0) + pc = byteCode + Load32Aligned(pc + 4); + else + pc += BC_CHECK_BIT_IN_TABLE_LENGTH; + break; + } + BYTECODE(CHECK_LT) { + uint32_t limit = (insn >> BYTECODE_SHIFT); + if (current_char < limit) + pc = byteCode + Load32Aligned(pc + 4); + else + pc += BC_CHECK_LT_LENGTH; + break; + } + BYTECODE(CHECK_GT) { + uint32_t limit = (insn >> BYTECODE_SHIFT); + if (current_char > limit) + pc = byteCode + Load32Aligned(pc + 4); + else + pc += BC_CHECK_GT_LENGTH; + break; + } + BYTECODE(CHECK_REGISTER_LT) + if (registers[insn >> BYTECODE_SHIFT] < Load32Aligned(pc + 4)) + pc = byteCode + Load32Aligned(pc + 8); + else + pc += BC_CHECK_REGISTER_LT_LENGTH; + break; + BYTECODE(CHECK_REGISTER_GE) + if (registers[insn >> BYTECODE_SHIFT] >= Load32Aligned(pc + 4)) + pc = byteCode + Load32Aligned(pc + 8); + else + pc += BC_CHECK_REGISTER_GE_LENGTH; + break; + BYTECODE(CHECK_REGISTER_EQ_POS) + if (registers[insn >> BYTECODE_SHIFT] == (int32_t) current) + pc = byteCode + Load32Aligned(pc + 4); + else + pc += BC_CHECK_REGISTER_EQ_POS_LENGTH; + break; + BYTECODE(CHECK_NOT_REGS_EQUAL) + if (registers[insn >> BYTECODE_SHIFT] == registers[Load32Aligned(pc + 4)]) + pc += BC_CHECK_NOT_REGS_EQUAL_LENGTH; + else + pc = byteCode + Load32Aligned(pc + 8); + break; + BYTECODE(CHECK_NOT_BACK_REF) { + int from = registers[insn >> BYTECODE_SHIFT]; + int len = registers[(insn >> BYTECODE_SHIFT) + 1] - from; + if (from < 0 || len <= 0) { + pc += BC_CHECK_NOT_BACK_REF_LENGTH; + break; + } + if (current + len > length) { + pc = byteCode + Load32Aligned(pc + 4); + break; + } else { + int i; + for (i = 0; i < len; i++) { + if (chars[from + i] != chars[current + i]) { + pc = byteCode + Load32Aligned(pc + 4); + break; + } + } + if (i < len) break; + current += len; + } + pc += BC_CHECK_NOT_BACK_REF_LENGTH; + break; + } + BYTECODE(CHECK_NOT_BACK_REF_NO_CASE) { + int from = registers[insn >> BYTECODE_SHIFT]; + int len = registers[(insn >> BYTECODE_SHIFT) + 1] - from; + if (from < 0 || len <= 0) { + pc += BC_CHECK_NOT_BACK_REF_NO_CASE_LENGTH; + break; + } + if (current + len > length) { + pc = byteCode + Load32Aligned(pc + 4); + break; + } + if (CaseInsensitiveCompareStrings(chars + from, chars + current, len * sizeof(CharT))) { + current += len; + pc += BC_CHECK_NOT_BACK_REF_NO_CASE_LENGTH; + } else { + pc = byteCode + Load32Aligned(pc + 4); + } + break; + } + BYTECODE(CHECK_NOT_BACK_REF_NO_CASE_UNICODE) { + int from = registers[insn >> BYTECODE_SHIFT]; + int len = registers[(insn >> BYTECODE_SHIFT) + 1] - from; + if (from < 0 || len <= 0) { + pc += BC_CHECK_NOT_BACK_REF_NO_CASE_UNICODE_LENGTH; + break; + } + if (current + len > length) { + pc = byteCode + Load32Aligned(pc + 4); + break; + } + if (CaseInsensitiveCompareUCStrings(chars + from, chars + current, + len * sizeof(CharT))) + { + current += len; + pc += BC_CHECK_NOT_BACK_REF_NO_CASE_UNICODE_LENGTH; + } else { + pc = byteCode + Load32Aligned(pc + 4); + } + break; + } + BYTECODE(CHECK_AT_START) + if (current == 0) + pc = byteCode + Load32Aligned(pc + 4); + else + pc += BC_CHECK_AT_START_LENGTH; + break; + BYTECODE(CHECK_NOT_AT_START) + if (current == 0) + pc += BC_CHECK_NOT_AT_START_LENGTH; + else + pc = byteCode + Load32Aligned(pc + 4); + break; + BYTECODE(SET_CURRENT_POSITION_FROM_END) { + size_t by = static_cast(insn) >> BYTECODE_SHIFT; + if (length - current > by) { + current = length - by; + current_char = chars[current - 1]; + } + pc += BC_SET_CURRENT_POSITION_FROM_END_LENGTH; + break; + } + default: + MOZ_CRASH("Bad bytecode"); + } + } +} + +template RegExpRunStatus +irregexp::InterpretCode(JSContext* cx, const uint8_t* byteCode, const Latin1Char* chars, size_t current, + size_t length, MatchPairs* matches, size_t* endIndex); + +template RegExpRunStatus +irregexp::InterpretCode(JSContext* cx, const uint8_t* byteCode, const char16_t* chars, size_t current, + size_t length, MatchPairs* matches, size_t* endIndex); diff --git a/js/src/irregexp/RegExpMacroAssembler.cpp b/js/src/irregexp/RegExpMacroAssembler.cpp new file mode 100644 index 000000000..d66d0d204 --- /dev/null +++ b/js/src/irregexp/RegExpMacroAssembler.cpp @@ -0,0 +1,573 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- + * vim: set ts=8 sts=4 et sw=4 tw=99: */ + +// Copyright 2012 the V8 project authors. All rights reserved. +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived +// from this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include "irregexp/RegExpMacroAssembler.h" + +#include "irregexp/RegExpBytecode.h" + +using namespace js; +using namespace js::irregexp; + +template +int +irregexp::CaseInsensitiveCompareStrings(const CharT* substring1, const CharT* substring2, + size_t byteLength) +{ + MOZ_ASSERT(byteLength % sizeof(CharT) == 0); + size_t length = byteLength / sizeof(CharT); + + for (size_t i = 0; i < length; i++) { + char16_t c1 = substring1[i]; + char16_t c2 = substring2[i]; + if (c1 != c2) { + c1 = unicode::ToLowerCase(c1); + c2 = unicode::ToLowerCase(c2); + if (c1 != c2) + return 0; + } + } + + return 1; +} + +template int +irregexp::CaseInsensitiveCompareStrings(const Latin1Char* substring1, const Latin1Char* substring2, + size_t byteLength); + +template int +irregexp::CaseInsensitiveCompareStrings(const char16_t* substring1, const char16_t* substring2, + size_t byteLength); + +template +int +irregexp::CaseInsensitiveCompareUCStrings(const CharT* substring1, const CharT* substring2, + size_t byteLength) +{ + MOZ_ASSERT(byteLength % sizeof(CharT) == 0); + size_t length = byteLength / sizeof(CharT); + + for (size_t i = 0; i < length; i++) { + char16_t c1 = substring1[i]; + char16_t c2 = substring2[i]; + if (c1 != c2) { + c1 = unicode::FoldCase(c1); + c2 = unicode::FoldCase(c2); + if (c1 != c2) + return 0; + } + } + + return 1; +} + +template int +irregexp::CaseInsensitiveCompareUCStrings(const Latin1Char* substring1, + const Latin1Char* substring2, + size_t byteLength); + +template int +irregexp::CaseInsensitiveCompareUCStrings(const char16_t* substring1, + const char16_t* substring2, + size_t byteLength); + +InterpretedRegExpMacroAssembler::InterpretedRegExpMacroAssembler(LifoAlloc* alloc, RegExpShared* shared, + size_t numSavedRegisters) + : RegExpMacroAssembler(*alloc, shared, numSavedRegisters), + pc_(0), + advance_current_start_(0), + advance_current_offset_(0), + advance_current_end_(kInvalidPC), + buffer_(nullptr), + length_(0) +{ + // The first int32 word is the number of registers. + Emit32(0); +} + +InterpretedRegExpMacroAssembler::~InterpretedRegExpMacroAssembler() +{ + js_free(buffer_); +} + +RegExpCode +InterpretedRegExpMacroAssembler::GenerateCode(JSContext* cx, bool match_only) +{ + Bind(&backtrack_); + Emit(BC_POP_BT, 0); + + // Update the number of registers. + *(int32_t*)buffer_ = num_registers_; + + RegExpCode res; + res.byteCode = buffer_; + buffer_ = nullptr; + return res; +} + +void +InterpretedRegExpMacroAssembler::AdvanceCurrentPosition(int by) +{ + MOZ_ASSERT(by >= kMinCPOffset); + MOZ_ASSERT(by <= kMaxCPOffset); + advance_current_start_ = pc_; + advance_current_offset_ = by; + Emit(BC_ADVANCE_CP, by); + advance_current_end_ = pc_; +} + +void +InterpretedRegExpMacroAssembler::AdvanceRegister(int reg, int by) +{ + checkRegister(reg); + Emit(BC_ADVANCE_REGISTER, reg); + Emit32(by); +} + +void +InterpretedRegExpMacroAssembler::Backtrack() +{ + Emit(BC_POP_BT, 0); +} + +void +InterpretedRegExpMacroAssembler::Bind(jit::Label* label) +{ + advance_current_end_ = kInvalidPC; + MOZ_ASSERT(!label->bound()); + if (label->used()) { + int pos = label->offset(); + while (pos != jit::Label::INVALID_OFFSET) { + int fixup = pos; + pos = *reinterpret_cast(buffer_ + fixup); + *reinterpret_cast(buffer_ + fixup) = pc_; + } + } + label->bind(pc_); +} + +void +InterpretedRegExpMacroAssembler::CheckAtStart(jit::Label* on_at_start) +{ + Emit(BC_CHECK_AT_START, 0); + EmitOrLink(on_at_start); +} + +void +InterpretedRegExpMacroAssembler::CheckCharacter(unsigned c, jit::Label* on_equal) +{ + if (c > MAX_FIRST_ARG) { + Emit(BC_CHECK_4_CHARS, 0); + Emit32(c); + } else { + Emit(BC_CHECK_CHAR, c); + } + EmitOrLink(on_equal); +} + +void +InterpretedRegExpMacroAssembler::CheckCharacterAfterAnd(unsigned c, unsigned and_with, jit::Label* on_equal) +{ + if (c > MAX_FIRST_ARG) { + Emit(BC_AND_CHECK_4_CHARS, 0); + Emit32(c); + } else { + Emit(BC_AND_CHECK_CHAR, c); + } + Emit32(and_with); + EmitOrLink(on_equal); +} + +void +InterpretedRegExpMacroAssembler::CheckCharacterGT(char16_t limit, jit::Label* on_greater) +{ + Emit(BC_CHECK_GT, limit); + EmitOrLink(on_greater); +} + +void +InterpretedRegExpMacroAssembler::CheckCharacterLT(char16_t limit, jit::Label* on_less) +{ + Emit(BC_CHECK_LT, limit); + EmitOrLink(on_less); +} + +void +InterpretedRegExpMacroAssembler::CheckGreedyLoop(jit::Label* on_tos_equals_current_position) +{ + Emit(BC_CHECK_GREEDY, 0); + EmitOrLink(on_tos_equals_current_position); +} + +void +InterpretedRegExpMacroAssembler::CheckNotAtStart(jit::Label* on_not_at_start) +{ + Emit(BC_CHECK_NOT_AT_START, 0); + EmitOrLink(on_not_at_start); +} + +void +InterpretedRegExpMacroAssembler::CheckNotBackReference(int start_reg, jit::Label* on_no_match) +{ + MOZ_ASSERT(start_reg >= 0); + MOZ_ASSERT(start_reg <= kMaxRegister); + Emit(BC_CHECK_NOT_BACK_REF, start_reg); + EmitOrLink(on_no_match); +} + +void +InterpretedRegExpMacroAssembler::CheckNotBackReferenceIgnoreCase(int start_reg, + jit::Label* on_no_match, + bool unicode) +{ + MOZ_ASSERT(start_reg >= 0); + MOZ_ASSERT(start_reg <= kMaxRegister); + if (unicode) + Emit(BC_CHECK_NOT_BACK_REF_NO_CASE_UNICODE, start_reg); + else + Emit(BC_CHECK_NOT_BACK_REF_NO_CASE, start_reg); + EmitOrLink(on_no_match); +} + +void +InterpretedRegExpMacroAssembler::CheckNotCharacter(unsigned c, jit::Label* on_not_equal) +{ + if (c > MAX_FIRST_ARG) { + Emit(BC_CHECK_NOT_4_CHARS, 0); + Emit32(c); + } else { + Emit(BC_CHECK_NOT_CHAR, c); + } + EmitOrLink(on_not_equal); +} + +void +InterpretedRegExpMacroAssembler::CheckNotCharacterAfterAnd(unsigned c, unsigned and_with, + jit::Label* on_not_equal) +{ + if (c > MAX_FIRST_ARG) { + Emit(BC_AND_CHECK_NOT_4_CHARS, 0); + Emit32(c); + } else { + Emit(BC_AND_CHECK_NOT_CHAR, c); + } + Emit32(and_with); + EmitOrLink(on_not_equal); +} + +void +InterpretedRegExpMacroAssembler::CheckNotCharacterAfterMinusAnd(char16_t c, char16_t minus, char16_t and_with, + jit::Label* on_not_equal) +{ + Emit(BC_MINUS_AND_CHECK_NOT_CHAR, c); + Emit16(minus); + Emit16(and_with); + EmitOrLink(on_not_equal); +} + +void +InterpretedRegExpMacroAssembler::CheckCharacterInRange(char16_t from, char16_t to, + jit::Label* on_in_range) +{ + Emit(BC_CHECK_CHAR_IN_RANGE, 0); + Emit16(from); + Emit16(to); + EmitOrLink(on_in_range); +} + +void +InterpretedRegExpMacroAssembler::CheckCharacterNotInRange(char16_t from, char16_t to, + jit::Label* on_not_in_range) +{ + Emit(BC_CHECK_CHAR_NOT_IN_RANGE, 0); + Emit16(from); + Emit16(to); + EmitOrLink(on_not_in_range); +} + +void +InterpretedRegExpMacroAssembler::CheckBitInTable(uint8_t* table, jit::Label* on_bit_set) +{ + static const int kBitsPerByte = 8; + + Emit(BC_CHECK_BIT_IN_TABLE, 0); + EmitOrLink(on_bit_set); + for (int i = 0; i < kTableSize; i += kBitsPerByte) { + int byte = 0; + for (int j = 0; j < kBitsPerByte; j++) { + if (table[i + j] != 0) + byte |= 1 << j; + } + Emit8(byte); + } +} + +void +InterpretedRegExpMacroAssembler::JumpOrBacktrack(jit::Label* to) +{ + if (advance_current_end_ == pc_) { + // Combine advance current and goto. + pc_ = advance_current_start_; + Emit(BC_ADVANCE_CP_AND_GOTO, advance_current_offset_); + EmitOrLink(to); + advance_current_end_ = kInvalidPC; + } else { + // Regular goto. + Emit(BC_GOTO, 0); + EmitOrLink(to); + } +} + +void +InterpretedRegExpMacroAssembler::Fail() +{ + Emit(BC_FAIL, 0); +} + +void +InterpretedRegExpMacroAssembler::IfRegisterGE(int reg, int comparand, jit::Label* if_ge) +{ + checkRegister(reg); + Emit(BC_CHECK_REGISTER_GE, reg); + Emit32(comparand); + EmitOrLink(if_ge); +} + +void +InterpretedRegExpMacroAssembler::IfRegisterLT(int reg, int comparand, jit::Label* if_lt) +{ + checkRegister(reg); + Emit(BC_CHECK_REGISTER_LT, reg); + Emit32(comparand); + EmitOrLink(if_lt); +} + +void +InterpretedRegExpMacroAssembler::IfRegisterEqPos(int reg, jit::Label* if_eq) +{ + checkRegister(reg); + Emit(BC_CHECK_REGISTER_EQ_POS, reg); + EmitOrLink(if_eq); +} + +void +InterpretedRegExpMacroAssembler::LoadCurrentCharacter(int cp_offset, jit::Label* on_end_of_input, + bool check_bounds, int characters) +{ + MOZ_ASSERT(cp_offset >= kMinCPOffset); + MOZ_ASSERT(cp_offset <= kMaxCPOffset); + int bytecode; + if (check_bounds) { + if (characters == 4) { + bytecode = BC_LOAD_4_CURRENT_CHARS; + } else if (characters == 2) { + bytecode = BC_LOAD_2_CURRENT_CHARS; + } else { + MOZ_ASSERT(characters == 1); + bytecode = BC_LOAD_CURRENT_CHAR; + } + } else { + if (characters == 4) { + bytecode = BC_LOAD_4_CURRENT_CHARS_UNCHECKED; + } else if (characters == 2) { + bytecode = BC_LOAD_2_CURRENT_CHARS_UNCHECKED; + } else { + MOZ_ASSERT(characters == 1); + bytecode = BC_LOAD_CURRENT_CHAR_UNCHECKED; + } + } + Emit(bytecode, cp_offset); + if (check_bounds) + EmitOrLink(on_end_of_input); +} + +void +InterpretedRegExpMacroAssembler::PopCurrentPosition() +{ + Emit(BC_POP_CP, 0); +} + +void +InterpretedRegExpMacroAssembler::PopRegister(int reg) +{ + checkRegister(reg); + Emit(BC_POP_REGISTER, reg); +} + +void +InterpretedRegExpMacroAssembler::PushCurrentPosition() +{ + Emit(BC_PUSH_CP, 0); +} + +void +InterpretedRegExpMacroAssembler::PushRegister(int reg, StackCheckFlag check_stack_limit) +{ + checkRegister(reg); + Emit(BC_PUSH_REGISTER, reg); +} + +void +InterpretedRegExpMacroAssembler::ReadCurrentPositionFromRegister(int reg) +{ + checkRegister(reg); + Emit(BC_SET_CP_TO_REGISTER, reg); +} + +void +InterpretedRegExpMacroAssembler::ReadBacktrackStackPointerFromRegister(int reg) +{ + checkRegister(reg); + Emit(BC_SET_SP_TO_REGISTER, reg); +} + +void +InterpretedRegExpMacroAssembler::SetCurrentPositionFromEnd(int by) +{ + MOZ_ASSERT(by >= 0 && by < (1 << 24)); + Emit(BC_SET_CURRENT_POSITION_FROM_END, by); +} + +void +InterpretedRegExpMacroAssembler::SetRegister(int reg, int to) +{ + checkRegister(reg); + Emit(BC_SET_REGISTER, reg); + Emit32(to); +} + +bool +InterpretedRegExpMacroAssembler::Succeed() +{ + Emit(BC_SUCCEED, 0); + + // Restart matching for global regexp not supported. + return false; +} + +void +InterpretedRegExpMacroAssembler::WriteCurrentPositionToRegister(int reg, int cp_offset) +{ + checkRegister(reg); + Emit(BC_SET_REGISTER_TO_CP, reg); + Emit32(cp_offset); // Current position offset. +} + +void +InterpretedRegExpMacroAssembler::ClearRegisters(int reg_from, int reg_to) +{ + MOZ_ASSERT(reg_from <= reg_to); + for (int reg = reg_from; reg <= reg_to; reg++) + SetRegister(reg, -1); +} + +void +InterpretedRegExpMacroAssembler::WriteBacktrackStackPointerToRegister(int reg) +{ + checkRegister(reg); + Emit(BC_SET_REGISTER_TO_SP, reg); +} + +void +InterpretedRegExpMacroAssembler::PushBacktrack(jit::Label* label) +{ + Emit(BC_PUSH_BT, 0); + EmitOrLink(label); +} + +void +InterpretedRegExpMacroAssembler::BindBacktrack(jit::Label* label) +{ + Bind(label); +} + +void +InterpretedRegExpMacroAssembler::EmitOrLink(jit::Label* label) +{ + if (label == nullptr) + label = &backtrack_; + if (label->bound()) { + Emit32(label->offset()); + } else { + int pos = label->use(pc_); + Emit32(pos); + } +} + +void +InterpretedRegExpMacroAssembler::Emit(uint32_t byte, uint32_t twenty_four_bits) +{ + uint32_t word = ((twenty_four_bits << BYTECODE_SHIFT) | byte); + Emit32(word); +} + +void +InterpretedRegExpMacroAssembler::Emit32(uint32_t word) +{ + MOZ_ASSERT(pc_ <= length_); + if (pc_ + 3 >= length_) + Expand(); + *reinterpret_cast(buffer_ + pc_) = word; + pc_ += 4; +} + +void +InterpretedRegExpMacroAssembler::Emit16(uint32_t word) +{ + MOZ_ASSERT(pc_ <= length_); + if (pc_ + 1 >= length_) + Expand(); + *reinterpret_cast(buffer_ + pc_) = word; + pc_ += 2; +} + +void +InterpretedRegExpMacroAssembler::Emit8(uint32_t word) +{ + MOZ_ASSERT(pc_ <= length_); + if (pc_ == length_) + Expand(); + *reinterpret_cast(buffer_ + pc_) = word; + pc_ += 1; +} + +void +InterpretedRegExpMacroAssembler::Expand() +{ + AutoEnterOOMUnsafeRegion oomUnsafe; + + int newLength = Max(100, length_ * 2); + if (newLength < length_ + 4) + oomUnsafe.crash("InterpretedRegExpMacroAssembler::Expand"); + + buffer_ = (uint8_t*) js_realloc(buffer_, newLength); + if (!buffer_) + oomUnsafe.crash("InterpretedRegExpMacroAssembler::Expand"); + length_ = newLength; +} diff --git a/js/src/irregexp/RegExpMacroAssembler.h b/js/src/irregexp/RegExpMacroAssembler.h new file mode 100644 index 000000000..dca2edf90 --- /dev/null +++ b/js/src/irregexp/RegExpMacroAssembler.h @@ -0,0 +1,312 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- + * vim: set ts=8 sts=4 et sw=4 tw=99: */ + +// Copyright 2012 the V8 project authors. All rights reserved. +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived +// from this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#ifndef V8_REGEXP_MACRO_ASSEMBLER_H_ +#define V8_REGEXP_MACRO_ASSEMBLER_H_ + +#include "irregexp/RegExpAST.h" +#include "irregexp/RegExpEngine.h" +#include "jit/MacroAssembler.h" + +namespace js { +namespace irregexp { + +class MOZ_STACK_CLASS RegExpMacroAssembler +{ + public: + RegExpMacroAssembler(LifoAlloc& alloc, RegExpShared* shared, size_t numSavedRegisters) + : slow_safe_compiler_(false), + global_mode_(NOT_GLOBAL), + alloc_(alloc), + num_registers_(numSavedRegisters), + num_saved_registers_(numSavedRegisters), + shared(shared) + {} + + enum StackCheckFlag { + kNoStackLimitCheck = false, + kCheckStackLimit = true + }; + + // The implementation must be able to handle at least: + static const int kMaxRegister = (1 << 16) - 1; + static const int kMaxCPOffset = (1 << 15) - 1; + static const int kMinCPOffset = -(1 << 15); + + static const int kTableSizeBits = 7; + static const int kTableSize = 1 << kTableSizeBits; + static const int kTableMask = kTableSize - 1; + + // Controls the generation of large inlined constants in the code. + void set_slow_safe(bool ssc) { slow_safe_compiler_ = ssc; } + bool slow_safe() { return slow_safe_compiler_; } + + enum GlobalMode { NOT_GLOBAL, GLOBAL, GLOBAL_NO_ZERO_LENGTH_CHECK }; + + // Set whether the regular expression has the global flag. Exiting due to + // a failure in a global regexp may still mean success overall. + inline void set_global_mode(GlobalMode mode) { global_mode_ = mode; } + inline bool global() { return global_mode_ != NOT_GLOBAL; } + inline bool global_with_zero_length_check() { + return global_mode_ == GLOBAL; + } + + LifoAlloc& alloc() { return alloc_; } + + virtual RegExpCode GenerateCode(JSContext* cx, bool match_only) = 0; + + // The maximal number of pushes between stack checks. Users must supply + // kCheckStackLimit flag to push operations (instead of kNoStackLimitCheck) + // at least once for every stack_limit() pushes that are executed. + virtual int stack_limit_slack() = 0; + + virtual bool CanReadUnaligned() { return false; } + + virtual void AdvanceCurrentPosition(int by) = 0; // Signed cp change. + virtual void AdvanceRegister(int reg, int by) = 0; // r[reg] += by. + + // Continues execution from the position pushed on the top of the backtrack + // stack by an earlier PushBacktrack. + virtual void Backtrack() = 0; + + virtual void Bind(jit::Label* label) = 0; + virtual void CheckAtStart(jit::Label* on_at_start) = 0; + + // Dispatch after looking the current character up in a 2-bits-per-entry + // map. The destinations vector has up to 4 labels. + virtual void CheckCharacter(unsigned c, jit::Label* on_equal) = 0; + + // Bitwise and the current character with the given constant and then + // check for a match with c. + virtual void CheckCharacterAfterAnd(unsigned c, unsigned and_with, jit::Label* on_equal) = 0; + + virtual void CheckCharacterGT(char16_t limit, jit::Label* on_greater) = 0; + virtual void CheckCharacterLT(char16_t limit, jit::Label* on_less) = 0; + virtual void CheckGreedyLoop(jit::Label* on_tos_equals_current_position) = 0; + virtual void CheckNotAtStart(jit::Label* on_not_at_start) = 0; + virtual void CheckNotBackReference(int start_reg, jit::Label* on_no_match) = 0; + virtual void CheckNotBackReferenceIgnoreCase(int start_reg, jit::Label* on_no_match, + bool unicode) = 0; + + // Check the current character for a match with a literal character. If we + // fail to match then goto the on_failure label. End of input always + // matches. If the label is nullptr then we should pop a backtrack address off + // the stack and go to that. + virtual void CheckNotCharacter(unsigned c, jit::Label* on_not_equal) = 0; + virtual void CheckNotCharacterAfterAnd(unsigned c, unsigned and_with, jit::Label* on_not_equal) = 0; + + // Subtract a constant from the current character, then and with the given + // constant and then check for a match with c. + virtual void CheckNotCharacterAfterMinusAnd(char16_t c, + char16_t minus, + char16_t and_with, + jit::Label* on_not_equal) = 0; + + virtual void CheckCharacterInRange(char16_t from, char16_t to, // Both inclusive. + jit::Label* on_in_range) = 0; + + virtual void CheckCharacterNotInRange(char16_t from, char16_t to, // Both inclusive. + jit::Label* on_not_in_range) = 0; + + // The current character (modulus the kTableSize) is looked up in the byte + // array, and if the found byte is non-zero, we jump to the on_bit_set label. + virtual void CheckBitInTable(uint8_t* table, jit::Label* on_bit_set) = 0; + + // Checks whether the given offset from the current position is before + // the end of the string. May overwrite the current character. + virtual void CheckPosition(int cp_offset, jit::Label* on_outside_input) { + LoadCurrentCharacter(cp_offset, on_outside_input, true); + } + + // Jump to either the target label or the top of the backtrack stack. + virtual void JumpOrBacktrack(jit::Label* to) = 0; + + // Check whether a standard/default character class matches the current + // character. Returns false if the type of special character class does + // not have custom support. + // May clobber the current loaded character. + virtual bool CheckSpecialCharacterClass(char16_t type, jit::Label* on_no_match) { + return false; + } + + virtual void Fail() = 0; + + // Check whether a register is >= a given constant and go to a label if it + // is. Backtracks instead if the label is nullptr. + virtual void IfRegisterGE(int reg, int comparand, jit::Label* if_ge) = 0; + + // Check whether a register is < a given constant and go to a label if it is. + // Backtracks instead if the label is nullptr. + virtual void IfRegisterLT(int reg, int comparand, jit::Label* if_lt) = 0; + + // Check whether a register is == to the current position and go to a + // label if it is. + virtual void IfRegisterEqPos(int reg, jit::Label* if_eq) = 0; + + virtual void LoadCurrentCharacter(int cp_offset, + jit::Label* on_end_of_input, + bool check_bounds = true, + int characters = 1) = 0; + virtual void PopCurrentPosition() = 0; + virtual void PopRegister(int register_index) = 0; + + virtual void PushCurrentPosition() = 0; + virtual void PushRegister(int register_index, StackCheckFlag check_stack_limit) = 0; + virtual void ReadCurrentPositionFromRegister(int reg) = 0; + virtual void ReadBacktrackStackPointerFromRegister(int reg) = 0; + virtual void SetCurrentPositionFromEnd(int by) = 0; + virtual void SetRegister(int register_index, int to) = 0; + + // Return whether the matching (with a global regexp) will be restarted. + virtual bool Succeed() = 0; + + virtual void WriteCurrentPositionToRegister(int reg, int cp_offset) = 0; + virtual void ClearRegisters(int reg_from, int reg_to) = 0; + virtual void WriteBacktrackStackPointerToRegister(int reg) = 0; + + // Pushes the label on the backtrack stack, so that a following Backtrack + // will go to this label. Always checks the backtrack stack limit. + virtual void PushBacktrack(jit::Label* label) = 0; + + // Bind a label that was previously used by PushBacktrack. + virtual void BindBacktrack(jit::Label* label) = 0; + + private: + bool slow_safe_compiler_; + GlobalMode global_mode_; + LifoAlloc& alloc_; + + protected: + int num_registers_; + int num_saved_registers_; + + void checkRegister(int reg) { + MOZ_ASSERT(reg >= 0); + MOZ_ASSERT(reg <= kMaxRegister); + if (num_registers_ <= reg) + num_registers_ = reg + 1; + } + + public: + RegExpShared* shared; +}; + +template +int +CaseInsensitiveCompareStrings(const CharT* substring1, const CharT* substring2, size_t byteLength); + +template +int +CaseInsensitiveCompareUCStrings(const CharT* substring1, const CharT* substring2, + size_t byteLength); + +class MOZ_STACK_CLASS InterpretedRegExpMacroAssembler final : public RegExpMacroAssembler +{ + public: + InterpretedRegExpMacroAssembler(LifoAlloc* alloc, RegExpShared* shared, size_t numSavedRegisters); + ~InterpretedRegExpMacroAssembler(); + + // Inherited virtual methods. + RegExpCode GenerateCode(JSContext* cx, bool match_only); + void AdvanceCurrentPosition(int by); + void AdvanceRegister(int reg, int by); + void Backtrack(); + void Bind(jit::Label* label); + void CheckAtStart(jit::Label* on_at_start); + void CheckCharacter(unsigned c, jit::Label* on_equal); + void CheckCharacterAfterAnd(unsigned c, unsigned and_with, jit::Label* on_equal); + void CheckCharacterGT(char16_t limit, jit::Label* on_greater); + void CheckCharacterLT(char16_t limit, jit::Label* on_less); + void CheckGreedyLoop(jit::Label* on_tos_equals_current_position); + void CheckNotAtStart(jit::Label* on_not_at_start); + void CheckNotBackReference(int start_reg, jit::Label* on_no_match); + void CheckNotBackReferenceIgnoreCase(int start_reg, jit::Label* on_no_match, bool unicode); + void CheckNotCharacter(unsigned c, jit::Label* on_not_equal); + void CheckNotCharacterAfterAnd(unsigned c, unsigned and_with, jit::Label* on_not_equal); + void CheckNotCharacterAfterMinusAnd(char16_t c, char16_t minus, char16_t and_with, + jit::Label* on_not_equal); + void CheckCharacterInRange(char16_t from, char16_t to, + jit::Label* on_in_range); + void CheckCharacterNotInRange(char16_t from, char16_t to, + jit::Label* on_not_in_range); + void CheckBitInTable(uint8_t* table, jit::Label* on_bit_set); + void JumpOrBacktrack(jit::Label* to); + void Fail(); + void IfRegisterGE(int reg, int comparand, jit::Label* if_ge); + void IfRegisterLT(int reg, int comparand, jit::Label* if_lt); + void IfRegisterEqPos(int reg, jit::Label* if_eq); + void LoadCurrentCharacter(int cp_offset, jit::Label* on_end_of_input, + bool check_bounds = true, int characters = 1); + void PopCurrentPosition(); + void PopRegister(int register_index); + void PushCurrentPosition(); + void PushRegister(int register_index, StackCheckFlag check_stack_limit); + void ReadCurrentPositionFromRegister(int reg); + void ReadBacktrackStackPointerFromRegister(int reg); + void SetCurrentPositionFromEnd(int by); + void SetRegister(int register_index, int to); + bool Succeed(); + void WriteCurrentPositionToRegister(int reg, int cp_offset); + void ClearRegisters(int reg_from, int reg_to); + void WriteBacktrackStackPointerToRegister(int reg); + void PushBacktrack(jit::Label* label); + void BindBacktrack(jit::Label* label); + + // The byte-code interpreter checks on each push anyway. + int stack_limit_slack() { return 1; } + + private: + void Expand(); + + // Code and bitmap emission. + void EmitOrLink(jit::Label* label); + void Emit32(uint32_t x); + void Emit16(uint32_t x); + void Emit8(uint32_t x); + void Emit(uint32_t bc, uint32_t arg); + + jit::Label backtrack_; + + // The program counter. + int pc_; + + int advance_current_start_; + int advance_current_offset_; + int advance_current_end_; + + static const int kInvalidPC = -1; + + uint8_t* buffer_; + int length_; +}; + +} } // namespace js::irregexp + +#endif // V8_REGEXP_MACRO_ASSEMBLER_H_ diff --git a/js/src/irregexp/RegExpParser.cpp b/js/src/irregexp/RegExpParser.cpp new file mode 100644 index 000000000..ccc6ae3eb --- /dev/null +++ b/js/src/irregexp/RegExpParser.cpp @@ -0,0 +1,1905 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- + * vim: set ts=8 sts=4 et sw=4 tw=99: */ + +// Copyright 2012 the V8 project authors. All rights reserved. +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived +// from this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include "irregexp/RegExpParser.h" + +#include "frontend/TokenStream.h" + +using namespace js; +using namespace js::irregexp; + +// ---------------------------------------------------------------------------- +// RegExpBuilder + +RegExpBuilder::RegExpBuilder(LifoAlloc* alloc) + : alloc(alloc), + pending_empty_(false), + characters_(nullptr) +#ifdef DEBUG + , last_added_(ADD_NONE) +#endif +{} + +void +RegExpBuilder::FlushCharacters() +{ + pending_empty_ = false; + if (characters_ != nullptr) { + RegExpTree* atom = alloc->newInfallible(characters_); + characters_ = nullptr; + text_.Add(alloc, atom); +#ifdef DEBUG + last_added_ = ADD_ATOM; +#endif + } +} + +void +RegExpBuilder::FlushText() +{ + FlushCharacters(); + int num_text = text_.length(); + if (num_text == 0) + return; + if (num_text == 1) { + terms_.Add(alloc, text_.last()); + } else { + RegExpText* text = alloc->newInfallible(alloc); + for (int i = 0; i < num_text; i++) + text_.Get(i)->AppendToText(text); + terms_.Add(alloc, text); + } + text_.Clear(); +} + +void +RegExpBuilder::AddCharacter(char16_t c) +{ + pending_empty_ = false; + if (characters_ == nullptr) + characters_ = alloc->newInfallible(*alloc); + characters_->append(c); +#ifdef DEBUG + last_added_ = ADD_CHAR; +#endif +} + +void +RegExpBuilder::AddEmpty() +{ + pending_empty_ = true; +} + +void +RegExpBuilder::AddAtom(RegExpTree* term) +{ + if (term->IsEmpty()) { + AddEmpty(); + return; + } + if (term->IsTextElement()) { + FlushCharacters(); + text_.Add(alloc, term); + } else { + FlushText(); + terms_.Add(alloc, term); + } +#ifdef DEBUG + last_added_ = ADD_ATOM; +#endif +} + +void +RegExpBuilder::AddAssertion(RegExpTree* assert) +{ + FlushText(); + terms_.Add(alloc, assert); +#ifdef DEBUG + last_added_ = ADD_ASSERT; +#endif +} + +void +RegExpBuilder::NewAlternative() +{ + FlushTerms(); +} + +void +RegExpBuilder::FlushTerms() +{ + FlushText(); + int num_terms = terms_.length(); + RegExpTree* alternative; + if (num_terms == 0) + alternative = RegExpEmpty::GetInstance(); + else if (num_terms == 1) + alternative = terms_.last(); + else + alternative = alloc->newInfallible(terms_.GetList(alloc)); + alternatives_.Add(alloc, alternative); + terms_.Clear(); +#ifdef DEBUG + last_added_ = ADD_NONE; +#endif +} + +RegExpTree* +RegExpBuilder::ToRegExp() +{ + FlushTerms(); + int num_alternatives = alternatives_.length(); + if (num_alternatives == 0) { + return RegExpEmpty::GetInstance(); + } + if (num_alternatives == 1) { + return alternatives_.last(); + } + return alloc->newInfallible(alternatives_.GetList(alloc)); +} + +void +RegExpBuilder::AddQuantifierToAtom(int min, int max, + RegExpQuantifier::QuantifierType quantifier_type) +{ + if (pending_empty_) { + pending_empty_ = false; + return; + } + RegExpTree* atom; + if (characters_ != nullptr) { + MOZ_ASSERT(last_added_ == ADD_CHAR); + // Last atom was character. + CharacterVector* char_vector = characters_; + int num_chars = char_vector->length(); + if (num_chars > 1) { + CharacterVector* prefix = alloc->newInfallible(*alloc); + prefix->append(char_vector->begin(), num_chars - 1); + text_.Add(alloc, alloc->newInfallible(prefix)); + char_vector = alloc->newInfallible(*alloc); + char_vector->append((*characters_)[num_chars - 1]); + } + characters_ = nullptr; + atom = alloc->newInfallible(char_vector); + FlushText(); + } else if (text_.length() > 0) { + MOZ_ASSERT(last_added_ == ADD_ATOM); + atom = text_.RemoveLast(); + FlushText(); + } else if (terms_.length() > 0) { + MOZ_ASSERT(last_added_ == ADD_ATOM); + atom = terms_.RemoveLast(); + if (atom->max_match() == 0) { + // Guaranteed to only match an empty string. +#ifdef DEBUG + last_added_ = ADD_TERM; +#endif + if (min == 0) + return; + terms_.Add(alloc, atom); + return; + } + } else { + // Only call immediately after adding an atom or character! + MOZ_CRASH("Bad call"); + } + terms_.Add(alloc, alloc->newInfallible(min, max, quantifier_type, atom)); +#ifdef DEBUG + last_added_ = ADD_TERM; +#endif +} + +// ---------------------------------------------------------------------------- +// RegExpParser + +template +RegExpParser::RegExpParser(frontend::TokenStream& ts, LifoAlloc* alloc, + const CharT* chars, const CharT* end, bool multiline_mode, + bool unicode, bool ignore_case) + : ts(ts), + alloc(alloc), + captures_(nullptr), + next_pos_(chars), + end_(end), + current_(kEndMarker), + capture_count_(0), + has_more_(true), + multiline_(multiline_mode), + unicode_(unicode), + ignore_case_(ignore_case), + simple_(false), + contains_anchor_(false), + is_scanned_for_captures_(false) +{ + Advance(); +} + +template +RegExpTree* +RegExpParser::ReportError(unsigned errorNumber) +{ + gc::AutoSuppressGC suppressGC(ts.context()); + ts.reportError(errorNumber); + return nullptr; +} + +template +void +RegExpParser::Advance() +{ + if (next_pos_ < end_) { + current_ = *next_pos_; + next_pos_++; + } else { + current_ = kEndMarker; + has_more_ = false; + } +} + +// Returns the value (0 .. 15) of a hexadecimal character c. +// If c is not a legal hexadecimal character, returns a value < 0. +inline int +HexValue(uint32_t c) +{ + c -= '0'; + if (static_cast(c) <= 9) return c; + c = (c | 0x20) - ('a' - '0'); // detect 0x11..0x16 and 0x31..0x36. + if (static_cast(c) <= 5) return c + 10; + return -1; +} + +template +widechar +RegExpParser::ParseOctalLiteral() +{ + MOZ_ASSERT('0' <= current() && current() <= '7'); + // For compatibility with some other browsers (not all), we parse + // up to three octal digits with a value below 256. + widechar value = current() - '0'; + Advance(); + if ('0' <= current() && current() <= '7') { + value = value * 8 + current() - '0'; + Advance(); + if (value < 32 && '0' <= current() && current() <= '7') { + value = value * 8 + current() - '0'; + Advance(); + } + } + return value; +} + +template +bool +RegExpParser::ParseHexEscape(int length, widechar* value) +{ + const CharT* start = position(); + uint32_t val = 0; + bool done = false; + for (int i = 0; !done; i++) { + widechar c = current(); + int d = HexValue(c); + if (d < 0) { + Reset(start); + return false; + } + val = val * 16 + d; + Advance(); + if (i == length - 1) { + done = true; + } + } + *value = val; + return true; +} + +template +bool +RegExpParser::ParseBracedHexEscape(widechar* value) +{ + MOZ_ASSERT(current() == '{'); + Advance(); + + bool first = true; + uint32_t code = 0; + while (true) { + widechar c = current(); + if (c == kEndMarker) { + ReportError(JSMSG_INVALID_UNICODE_ESCAPE); + return false; + } + if (c == '}') { + if (first) { + ReportError(JSMSG_INVALID_UNICODE_ESCAPE); + return false; + } + Advance(); + break; + } + + int d = HexValue(c); + if (d < 0) { + ReportError(JSMSG_INVALID_UNICODE_ESCAPE); + return false; + } + code = (code << 4) | d; + if (code > unicode::NonBMPMax) { + ReportError(JSMSG_UNICODE_OVERFLOW); + return false; + } + Advance(); + first = false; + } + + *value = code; + return true; +} + +template +bool +RegExpParser::ParseTrailSurrogate(widechar* value) +{ + if (current() != '\\') + return false; + + const CharT* start = position(); + Advance(); + if (current() != 'u') { + Reset(start); + return false; + } + Advance(); + if (!ParseHexEscape(4, value)) { + Reset(start); + return false; + } + if (!unicode::IsTrailSurrogate(*value)) { + Reset(start); + return false; + } + return true; +} + +template +bool +RegExpParser::ParseRawSurrogatePair(char16_t* lead, char16_t* trail) +{ + widechar c1 = current(); + if (!unicode::IsLeadSurrogate(c1)) + return false; + + const CharT* start = position(); + Advance(); + widechar c2 = current(); + if (!unicode::IsTrailSurrogate(c2)) { + Reset(start); + return false; + } + Advance(); + *lead = c1; + *trail = c2; + return true; +} + +static inline RegExpTree* +RangeAtom(LifoAlloc* alloc, char16_t from, char16_t to) +{ + CharacterRangeVector* ranges = alloc->newInfallible(*alloc); + ranges->append(CharacterRange::Range(from, to)); + return alloc->newInfallible(ranges, false); +} + +static inline RegExpTree* +NegativeLookahead(LifoAlloc* alloc, char16_t from, char16_t to) +{ + return alloc->newInfallible(RangeAtom(alloc, from, to), false, 0, 0); +} + +static bool +IsSyntaxCharacter(widechar c) +{ + switch (c) { + case '^': + case '$': + case '\\': + case '.': + case '*': + case '+': + case '?': + case '(': + case ')': + case '[': + case ']': + case '{': + case '}': + case '|': + case '/': + return true; + default: + return false; + } +} + +#ifdef DEBUG +// Currently only used in an assert.kASSERT. +static bool +IsSpecialClassEscape(widechar c) +{ + switch (c) { + case 'd': case 'D': + case 's': case 'S': + case 'w': case 'W': + return true; + default: + return false; + } +} +#endif + +template +bool +RegExpParser::ParseClassCharacterEscape(widechar* code) +{ + MOZ_ASSERT(current() == '\\'); + MOZ_ASSERT(has_next() && !IsSpecialClassEscape(Next())); + Advance(); + switch (current()) { + case 'b': + Advance(); + *code = '\b'; + return true; + // ControlEscape :: one of + // f n r t v + case 'f': + Advance(); + *code = '\f'; + return true; + case 'n': + Advance(); + *code = '\n'; + return true; + case 'r': + Advance(); + *code = '\r'; + return true; + case 't': + Advance(); + *code = '\t'; + return true; + case 'v': + Advance(); + *code = '\v'; + return true; + case 'c': { + widechar controlLetter = Next(); + widechar letter = controlLetter & ~('A' ^ 'a'); + // For compatibility with JSC, inside a character class + // we also accept digits and underscore as control characters, + // but only in non-unicode mode + if ((!unicode_ && + ((controlLetter >= '0' && controlLetter <= '9') || + controlLetter == '_')) || + (letter >= 'A' && letter <= 'Z')) + { + Advance(2); + // Control letters mapped to ASCII control characters in the range + // 0x00-0x1f. + *code = controlLetter & 0x1f; + return true; + } + if (unicode_) { + ReportError(JSMSG_INVALID_IDENTITY_ESCAPE); + return false; + } + // We match JSC in reading the backslash as a literal + // character instead of as starting an escape. + *code = '\\'; + return true; + } + case '0': case '1': case '2': case '3': case '4': case '5': + case '6': case '7': + if (unicode_) { + if (current() == '0') { + Advance(); + *code = 0; + return true; + } + ReportError(JSMSG_INVALID_IDENTITY_ESCAPE); + return false; + } + // For compatibility, outside of unicode mode, we interpret a decimal + // escape that isn't a back reference (and therefore either \0 or not + // valid according to the specification) as a 1..3 digit octal + // character code. + *code = ParseOctalLiteral(); + return true; + case 'x': { + Advance(); + widechar value; + if (ParseHexEscape(2, &value)) { + *code = value; + return true; + } + if (unicode_) { + ReportError(JSMSG_INVALID_IDENTITY_ESCAPE); + return false; + } + // If \x is not followed by a two-digit hexadecimal, treat it + // as an identity escape in non-unicode mode. + *code = 'x'; + return true; + } + case 'u': { + Advance(); + widechar value; + if (unicode_) { + if (current() == '{') { + if (!ParseBracedHexEscape(&value)) + return false; + *code = value; + return true; + } + if (ParseHexEscape(4, &value)) { + if (unicode::IsLeadSurrogate(value)) { + widechar trail; + if (ParseTrailSurrogate(&trail)) { + *code = unicode::UTF16Decode(value, trail); + return true; + } + } + *code = value; + return true; + } + ReportError(JSMSG_INVALID_UNICODE_ESCAPE); + return false; + } + if (ParseHexEscape(4, &value)) { + *code = value; + return true; + } + // If \u is not followed by a four-digit or braced hexadecimal, treat it + // as an identity escape. + *code = 'u'; + return true; + } + default: { + // Extended identity escape (non-unicode only). We accept any character + // that hasn't been matched by a more specific case, not just the subset + // required by the ECMAScript specification. + widechar result = current(); + if (unicode_ && result != '-' && !IsSyntaxCharacter(result)) { + ReportError(JSMSG_INVALID_IDENTITY_ESCAPE); + return false; + } + Advance(); + *code = result; + return true; + } + } + return true; +} + +class WideCharRange +{ + public: + WideCharRange() + : from_(0), to_(0) + {} + + WideCharRange(widechar from, widechar to) + : from_(from), to_(to) + {} + + static inline WideCharRange Singleton(widechar value) { + return WideCharRange(value, value); + } + static inline WideCharRange Range(widechar from, widechar to) { + MOZ_ASSERT(from <= to); + return WideCharRange(from, to); + } + + bool Contains(widechar i) const { return from_ <= i && i <= to_; } + widechar from() const { return from_; } + widechar to() const { return to_; } + + private: + widechar from_; + widechar to_; +}; + +typedef InfallibleVector WideCharRangeVector; + +static inline CharacterRange +LeadSurrogateRange() +{ + return CharacterRange::Range(unicode::LeadSurrogateMin, unicode::LeadSurrogateMax); +} + +static inline CharacterRange +TrailSurrogateRange() +{ + return CharacterRange::Range(unicode::TrailSurrogateMin, unicode::TrailSurrogateMax); +} + +static inline WideCharRange +NonBMPRange() +{ + return WideCharRange::Range(unicode::NonBMPMin, unicode::NonBMPMax); +} + +static const char16_t kNoCharClass = 0; + +// Adds a character or pre-defined character class to character ranges. +// If char_class is not kInvalidClass, it's interpreted as a class +// escape (i.e., 's' means whitespace, from '\s'). +static inline void +AddCharOrEscape(LifoAlloc* alloc, + CharacterRangeVector* ranges, + char16_t char_class, + widechar c) +{ + if (char_class != kNoCharClass) + CharacterRange::AddClassEscape(alloc, char_class, ranges); + else + ranges->append(CharacterRange::Singleton(c)); +} + +static inline void +AddCharOrEscapeUnicode(LifoAlloc* alloc, + CharacterRangeVector* ranges, + CharacterRangeVector* lead_ranges, + CharacterRangeVector* trail_ranges, + WideCharRangeVector* wide_ranges, + char16_t char_class, + widechar c, + bool ignore_case) +{ + if (char_class != kNoCharClass) { + CharacterRange::AddClassEscapeUnicode(alloc, char_class, ranges, ignore_case); + switch (char_class) { + case 'S': + case 'W': + case 'D': + lead_ranges->append(LeadSurrogateRange()); + trail_ranges->append(TrailSurrogateRange()); + wide_ranges->append(NonBMPRange()); + break; + case '.': + MOZ_CRASH("Bad char_class!"); + } + return; + } + + if (unicode::IsLeadSurrogate(c)) + lead_ranges->append(CharacterRange::Singleton(c)); + else if (unicode::IsTrailSurrogate(c)) + trail_ranges->append(CharacterRange::Singleton(c)); + else if (c >= unicode::NonBMPMin) + wide_ranges->append(WideCharRange::Singleton(c)); + else + ranges->append(CharacterRange::Singleton(c)); +} + +static inline void +AddUnicodeRange(LifoAlloc* alloc, + CharacterRangeVector* ranges, + CharacterRangeVector* lead_ranges, + CharacterRangeVector* trail_ranges, + WideCharRangeVector* wide_ranges, + widechar first, + widechar next) +{ + MOZ_ASSERT(first <= next); + if (first < unicode::LeadSurrogateMin) { + if (next < unicode::LeadSurrogateMin) { + ranges->append(CharacterRange::Range(first, next)); + return; + } + ranges->append(CharacterRange::Range(first, unicode::LeadSurrogateMin - 1)); + first = unicode::LeadSurrogateMin; + } + if (first <= unicode::LeadSurrogateMax) { + if (next <= unicode::LeadSurrogateMax) { + lead_ranges->append(CharacterRange::Range(first, next)); + return; + } + lead_ranges->append(CharacterRange::Range(first, unicode::LeadSurrogateMax)); + first = unicode::LeadSurrogateMax + 1; + } + MOZ_ASSERT(unicode::LeadSurrogateMax + 1 == unicode::TrailSurrogateMin); + if (first <= unicode::TrailSurrogateMax) { + if (next <= unicode::TrailSurrogateMax) { + trail_ranges->append(CharacterRange::Range(first, next)); + return; + } + trail_ranges->append(CharacterRange::Range(first, unicode::TrailSurrogateMax)); + first = unicode::TrailSurrogateMax + 1; + } + if (first <= unicode::UTF16Max) { + if (next <= unicode::UTF16Max) { + ranges->append(CharacterRange::Range(first, next)); + return; + } + ranges->append(CharacterRange::Range(first, unicode::UTF16Max)); + first = unicode::NonBMPMin; + } + MOZ_ASSERT(unicode::UTF16Max + 1 == unicode::NonBMPMin); + wide_ranges->append(WideCharRange::Range(first, next)); +} + +// Negate a vector of ranges by subtracting its ranges from a range +// encompassing the full range of possible values. +template +static inline void +NegateUnicodeRanges(LifoAlloc* alloc, InfallibleVector** ranges, + RangeType full_range) +{ + typedef InfallibleVector RangeVector; + RangeVector* tmp_ranges = alloc->newInfallible(*alloc); + tmp_ranges->append(full_range); + RangeVector* result_ranges = alloc->newInfallible(*alloc); + + // Perform the following calculation: + // result_ranges = tmp_ranges - ranges + // with the following steps: + // result_ranges = tmp_ranges - ranges[0] + // SWAP(result_ranges, tmp_ranges) + // result_ranges = tmp_ranges - ranges[1] + // SWAP(result_ranges, tmp_ranges) + // ... + // result_ranges = tmp_ranges - ranges[N-1] + // SWAP(result_ranges, tmp_ranges) + // The last SWAP is just for simplicity of the loop. + for (size_t i = 0; i < (*ranges)->length(); i++) { + result_ranges->clear(); + + const RangeType& range = (**ranges)[i]; + for (size_t j = 0; j < tmp_ranges->length(); j++) { + const RangeType& tmpRange = (*tmp_ranges)[j]; + auto from1 = tmpRange.from(); + auto to1 = tmpRange.to(); + auto from2 = range.from(); + auto to2 = range.to(); + + if (from1 < from2) { + if (to1 < from2) { + result_ranges->append(tmpRange); + } else if (to1 <= to2) { + result_ranges->append(RangeType::Range(from1, from2 - 1)); + } else { + result_ranges->append(RangeType::Range(from1, from2 - 1)); + result_ranges->append(RangeType::Range(to2 + 1, to1)); + } + } else if (from1 <= to2) { + if (to1 > to2) + result_ranges->append(RangeType::Range(to2 + 1, to1)); + } else { + result_ranges->append(tmpRange); + } + } + + auto tmp = tmp_ranges; + tmp_ranges = result_ranges; + result_ranges = tmp; + } + + // After the loop, result is pointed at by tmp_ranges, instead of + // result_ranges. + *ranges = tmp_ranges; +} + +static bool +WideCharRangesContain(WideCharRangeVector* wide_ranges, widechar c) +{ + for (size_t i = 0; i < wide_ranges->length(); i++) { + const WideCharRange& range = (*wide_ranges)[i]; + if (range.Contains(c)) + return true; + } + return false; +} + +static void +CalculateCaseInsensitiveRanges(LifoAlloc* alloc, widechar from, widechar to, int32_t diff, + WideCharRangeVector* wide_ranges, + WideCharRangeVector** tmp_wide_ranges) +{ + widechar contains_from = 0; + widechar contains_to = 0; + for (widechar c = from; c <= to; c++) { + if (WideCharRangesContain(wide_ranges, c) && + !WideCharRangesContain(wide_ranges, c + diff)) + { + if (contains_from == 0) + contains_from = c; + contains_to = c; + } else if (contains_from != 0) { + if (!*tmp_wide_ranges) + *tmp_wide_ranges = alloc->newInfallible(*alloc); + + (*tmp_wide_ranges)->append(WideCharRange::Range(contains_from + diff, + contains_to + diff)); + contains_from = 0; + } + } + + if (contains_from != 0) { + if (!*tmp_wide_ranges) + *tmp_wide_ranges = alloc->newInfallible(*alloc); + + (*tmp_wide_ranges)->append(WideCharRange::Range(contains_from + diff, + contains_to + diff)); + } +} + +static RegExpTree* +UnicodeRangesAtom(LifoAlloc* alloc, + CharacterRangeVector* ranges, + CharacterRangeVector* lead_ranges, + CharacterRangeVector* trail_ranges, + WideCharRangeVector* wide_ranges, + bool is_negated, + bool ignore_case) +{ + // Calculate case folding for non-BMP first and negate the range if needed. + if (ignore_case) { + WideCharRangeVector* tmp_wide_ranges = nullptr; +#define CALL_CALC(FROM, TO, LEAD, TRAIL_FROM, TRAIL_TO, DIFF) \ + CalculateCaseInsensitiveRanges(alloc, FROM, TO, DIFF, wide_ranges, &tmp_wide_ranges); + FOR_EACH_NON_BMP_CASE_FOLDING(CALL_CALC) + FOR_EACH_NON_BMP_REV_CASE_FOLDING(CALL_CALC) +#undef CALL_CALC + + if (tmp_wide_ranges) { + for (size_t i = 0; i < tmp_wide_ranges->length(); i++) + wide_ranges->append((*tmp_wide_ranges)[i]); + } + } + + if (is_negated) { + NegateUnicodeRanges(alloc, &lead_ranges, LeadSurrogateRange()); + NegateUnicodeRanges(alloc, &trail_ranges, TrailSurrogateRange()); + NegateUnicodeRanges(alloc, &wide_ranges, NonBMPRange()); + } + + RegExpBuilder* builder = alloc->newInfallible(alloc); + + bool added = false; + + if (is_negated) { + ranges->append(LeadSurrogateRange()); + ranges->append(TrailSurrogateRange()); + } + if (ranges->length() > 0) { + builder->AddAtom(alloc->newInfallible(ranges, is_negated)); + added = true; + } + + if (lead_ranges->length() > 0) { + if (added) + builder->NewAlternative(); + builder->AddAtom(alloc->newInfallible(lead_ranges, false)); + builder->AddAtom(NegativeLookahead(alloc, unicode::TrailSurrogateMin, + unicode::TrailSurrogateMax)); + added = true; + } + + if (trail_ranges->length() > 0) { + if (added) + builder->NewAlternative(); + builder->AddAssertion(alloc->newInfallible( + RegExpAssertion::NOT_AFTER_LEAD_SURROGATE)); + builder->AddAtom(alloc->newInfallible(trail_ranges, false)); + added = true; + } + + for (size_t i = 0; i < wide_ranges->length(); i++) { + if (added) + builder->NewAlternative(); + + const WideCharRange& range = (*wide_ranges)[i]; + widechar from = range.from(); + widechar to = range.to(); + char16_t from_lead, from_trail; + char16_t to_lead, to_trail; + + unicode::UTF16Encode(from, &from_lead, &from_trail); + if (from == to) { + builder->AddCharacter(from_lead); + builder->AddCharacter(from_trail); + } else { + unicode::UTF16Encode(to, &to_lead, &to_trail); + if (from_lead == to_lead) { + MOZ_ASSERT(from_trail != to_trail); + builder->AddCharacter(from_lead); + builder->AddAtom(RangeAtom(alloc, from_trail, to_trail)); + } else if (from_trail == unicode::TrailSurrogateMin && + to_trail == unicode::TrailSurrogateMax) + { + builder->AddAtom(RangeAtom(alloc, from_lead, to_lead)); + builder->AddAtom(RangeAtom(alloc, unicode::TrailSurrogateMin, + unicode::TrailSurrogateMax)); + } else if (from_lead + 1 == to_lead) { + builder->AddCharacter(from_lead); + builder->AddAtom(RangeAtom(alloc, from_trail, unicode::TrailSurrogateMax)); + + builder->NewAlternative(); + + builder->AddCharacter(to_lead); + builder->AddAtom(RangeAtom(alloc, unicode::TrailSurrogateMin, to_trail)); + } else if (from_lead + 2 == to_lead) { + builder->AddCharacter(from_lead); + builder->AddAtom(RangeAtom(alloc, from_trail, unicode::TrailSurrogateMax)); + + builder->NewAlternative(); + + builder->AddCharacter(from_lead + 1); + builder->AddAtom(RangeAtom(alloc, unicode::TrailSurrogateMin, + unicode::TrailSurrogateMax)); + + builder->NewAlternative(); + + builder->AddCharacter(to_lead); + builder->AddAtom(RangeAtom(alloc, unicode::TrailSurrogateMin, to_trail)); + } else { + builder->AddCharacter(from_lead); + builder->AddAtom(RangeAtom(alloc, from_trail, unicode::TrailSurrogateMax)); + + builder->NewAlternative(); + + builder->AddAtom(RangeAtom(alloc, from_lead + 1, to_lead - 1)); + builder->AddAtom(RangeAtom(alloc, unicode::TrailSurrogateMin, + unicode::TrailSurrogateMax)); + + builder->NewAlternative(); + + builder->AddCharacter(to_lead); + builder->AddAtom(RangeAtom(alloc, unicode::TrailSurrogateMin, to_trail)); + } + } + added = true; + } + + return builder->ToRegExp(); +} + +template +RegExpTree* +RegExpParser::ParseCharacterClass() +{ + MOZ_ASSERT(current() == '['); + Advance(); + bool is_negated = false; + if (current() == '^') { + is_negated = true; + Advance(); + } + CharacterRangeVector* ranges = alloc->newInfallible(*alloc); + CharacterRangeVector* lead_ranges = nullptr; + CharacterRangeVector* trail_ranges = nullptr; + WideCharRangeVector* wide_ranges = nullptr; + + if (unicode_) { + lead_ranges = alloc->newInfallible(*alloc); + trail_ranges = alloc->newInfallible(*alloc); + wide_ranges = alloc->newInfallible(*alloc); + } + + while (has_more() && current() != ']') { + char16_t char_class = kNoCharClass; + widechar first = 0; + if (!ParseClassAtom(&char_class, &first)) + return nullptr; + if (current() == '-') { + Advance(); + if (current() == kEndMarker) { + // If we reach the end we break out of the loop and let the + // following code report an error. + break; + } else if (current() == ']') { + if (unicode_) { + AddCharOrEscapeUnicode(alloc, ranges, lead_ranges, trail_ranges, wide_ranges, + char_class, first, ignore_case_); + } else { + AddCharOrEscape(alloc, ranges, char_class, first); + } + ranges->append(CharacterRange::Singleton('-')); + break; + } + char16_t char_class_2 = kNoCharClass; + widechar next = 0; + if (!ParseClassAtom(&char_class_2, &next)) + return nullptr; + if (char_class != kNoCharClass || char_class_2 != kNoCharClass) { + if (unicode_) + return ReportError(JSMSG_RANGE_WITH_CLASS_ESCAPE); + + // Either end is an escaped character class. Treat the '-' verbatim. + AddCharOrEscape(alloc, ranges, char_class, first); + ranges->append(CharacterRange::Singleton('-')); + AddCharOrEscape(alloc, ranges, char_class_2, next); + continue; + } + if (first > next) + return ReportError(JSMSG_BAD_CLASS_RANGE); + if (unicode_) + AddUnicodeRange(alloc, ranges, lead_ranges, trail_ranges,wide_ranges, first, next); + else + ranges->append(CharacterRange::Range(first, next)); + } else { + if (unicode_) { + AddCharOrEscapeUnicode(alloc, ranges, lead_ranges, trail_ranges, wide_ranges, + char_class, first, ignore_case_); + } else { + AddCharOrEscape(alloc, ranges, char_class, first); + } + } + } + if (!has_more()) + return ReportError(JSMSG_UNTERM_CLASS); + Advance(); + if (!unicode_) { + if (ranges->length() == 0) { + ranges->append(CharacterRange::Everything()); + is_negated = !is_negated; + } + return alloc->newInfallible(ranges, is_negated); + } + + if (!is_negated && ranges->length() == 0 && lead_ranges->length() == 0 && + trail_ranges->length() == 0 && wide_ranges->length() == 0) + { + ranges->append(CharacterRange::Everything()); + return alloc->newInfallible(ranges, true); + } + + return UnicodeRangesAtom(alloc, ranges, lead_ranges, trail_ranges, wide_ranges, is_negated, + ignore_case_); +} + +template +bool +RegExpParser::ParseClassAtom(char16_t* char_class, widechar* value) +{ + MOZ_ASSERT(*char_class == kNoCharClass); + widechar first = current(); + if (first == '\\') { + switch (Next()) { + case 'w': case 'W': case 'd': case 'D': case 's': case 'S': { + *char_class = Next(); + Advance(2); + return true; + } + case kEndMarker: + return ReportError(JSMSG_ESCAPE_AT_END_OF_REGEXP); + default: + if (!ParseClassCharacterEscape(value)) + return false; + return true; + } + } else { + if (unicode_) { + char16_t lead, trail; + if (ParseRawSurrogatePair(&lead, &trail)) { + *value = unicode::UTF16Decode(lead, trail); + return true; + } + } + Advance(); + *value = first; + return true; + } +} + +// In order to know whether an escape is a backreference or not we have to scan +// the entire regexp and find the number of capturing parentheses. However we +// don't want to scan the regexp twice unless it is necessary. This mini-parser +// is called when needed. It can see the difference between capturing and +// noncapturing parentheses and can skip character classes and backslash-escaped +// characters. +template +void +RegExpParser::ScanForCaptures() +{ + // Start with captures started previous to current position + int capture_count = captures_started(); + // Add count of captures after this position. + widechar n; + while ((n = current()) != kEndMarker) { + Advance(); + switch (n) { + case '\\': + Advance(); + break; + case '[': { + widechar c; + while ((c = current()) != kEndMarker) { + Advance(); + if (c == '\\') { + Advance(); + } else { + if (c == ']') break; + } + } + break; + } + case '(': + if (current() != '?') capture_count++; + break; + } + } + capture_count_ = capture_count; + is_scanned_for_captures_ = true; +} + +inline bool +IsInRange(int value, int lower_limit, int higher_limit) +{ + MOZ_ASSERT(lower_limit <= higher_limit); + return static_cast(value - lower_limit) <= + static_cast(higher_limit - lower_limit); +} + +inline bool +IsDecimalDigit(widechar c) +{ + // ECMA-262, 3rd, 7.8.3 (p 16) + return IsInRange(c, '0', '9'); +} + +template +bool +RegExpParser::ParseBackReferenceIndex(int* index_out) +{ + MOZ_ASSERT('\\' == current()); + MOZ_ASSERT('1' <= Next() && Next() <= '9'); + + // Try to parse a decimal literal that is no greater than the total number + // of left capturing parentheses in the input. + const CharT* start = position(); + int value = Next() - '0'; + Advance(2); + while (true) { + widechar c = current(); + if (IsDecimalDigit(c)) { + value = 10 * value + (c - '0'); + if (value > kMaxCaptures) { + Reset(start); + return false; + } + Advance(); + } else { + break; + } + } + if (value > captures_started()) { + if (!is_scanned_for_captures_) { + const CharT* saved_position = position(); + ScanForCaptures(); + Reset(saved_position); + } + if (value > capture_count_) { + Reset(start); + return false; + } + } + *index_out = value; + return true; +} + +// QuantifierPrefix :: +// { DecimalDigits } +// { DecimalDigits , } +// { DecimalDigits , DecimalDigits } +// +// Returns true if parsing succeeds, and set the min_out and max_out +// values. Values are truncated to RegExpTree::kInfinity if they overflow. +template +bool +RegExpParser::ParseIntervalQuantifier(int* min_out, int* max_out) +{ + MOZ_ASSERT(current() == '{'); + const CharT* start = position(); + Advance(); + int min = 0; + if (!IsDecimalDigit(current())) { + Reset(start); + return false; + } + while (IsDecimalDigit(current())) { + int next = current() - '0'; + if (min > (RegExpTree::kInfinity - next) / 10) { + // Overflow. Skip past remaining decimal digits and return -1. + do { + Advance(); + } while (IsDecimalDigit(current())); + min = RegExpTree::kInfinity; + break; + } + min = 10 * min + next; + Advance(); + } + int max = 0; + if (current() == '}') { + max = min; + Advance(); + } else if (current() == ',') { + Advance(); + if (current() == '}') { + max = RegExpTree::kInfinity; + Advance(); + } else { + while (IsDecimalDigit(current())) { + int next = current() - '0'; + if (max > (RegExpTree::kInfinity - next) / 10) { + do { + Advance(); + } while (IsDecimalDigit(current())); + max = RegExpTree::kInfinity; + break; + } + max = 10 * max + next; + Advance(); + } + if (current() != '}') { + Reset(start); + return false; + } + Advance(); + } + } else { + Reset(start); + return false; + } + *min_out = min; + *max_out = max; + return true; +} + +// Pattern :: +// Disjunction +template +RegExpTree* +RegExpParser::ParsePattern() +{ + RegExpTree* result = ParseDisjunction(); + MOZ_ASSERT_IF(result, !has_more()); + return result; +} + +static inline RegExpTree* +CaseFoldingSurrogatePairAtom(LifoAlloc* alloc, char16_t lead, char16_t trail, int32_t diff) +{ + RegExpBuilder* builder = alloc->newInfallible(alloc); + + builder->AddCharacter(lead); + CharacterRangeVector* ranges = alloc->newInfallible(*alloc); + ranges->append(CharacterRange::Range(trail, trail)); + ranges->append(CharacterRange::Range(trail + diff, trail + diff)); + builder->AddAtom(alloc->newInfallible(ranges, false)); + + return builder->ToRegExp(); +} + +static inline RegExpTree* +SurrogatePairAtom(LifoAlloc* alloc, char16_t lead, char16_t trail, bool ignore_case) +{ + if (ignore_case) { +#define CALL_ATOM(FROM, TO, LEAD, TRAIL_FROM, TRAIL_TO, DIFF) \ + if (lead == LEAD &&trail >= TRAIL_FROM && trail <= TRAIL_TO) \ + return CaseFoldingSurrogatePairAtom(alloc, lead, trail, DIFF); + FOR_EACH_NON_BMP_CASE_FOLDING(CALL_ATOM) + FOR_EACH_NON_BMP_REV_CASE_FOLDING(CALL_ATOM) +#undef CALL_ATOM + } + + RegExpBuilder* builder = alloc->newInfallible(alloc); + builder->AddCharacter(lead); + builder->AddCharacter(trail); + return builder->ToRegExp(); +} + +static inline RegExpTree* +LeadSurrogateAtom(LifoAlloc* alloc, char16_t value) +{ + RegExpBuilder* builder = alloc->newInfallible(alloc); + builder->AddCharacter(value); + builder->AddAtom(NegativeLookahead(alloc, unicode::TrailSurrogateMin, + unicode::TrailSurrogateMax)); + return builder->ToRegExp(); +} + +static inline RegExpTree* +TrailSurrogateAtom(LifoAlloc* alloc, char16_t value) +{ + RegExpBuilder* builder = alloc->newInfallible(alloc); + builder->AddAssertion(alloc->newInfallible( + RegExpAssertion::NOT_AFTER_LEAD_SURROGATE)); + builder->AddCharacter(value); + return builder->ToRegExp(); +} + +static inline RegExpTree* +UnicodeEverythingAtom(LifoAlloc* alloc) +{ + RegExpBuilder* builder = alloc->newInfallible(alloc); + + // everything except \x0a, \x0d, \u2028 and \u2029 + + CharacterRangeVector* ranges = alloc->newInfallible(*alloc); + ranges->append(CharacterRange::Range(0x0, 0x09)); + ranges->append(CharacterRange::Range(0x0b, 0x0c)); + ranges->append(CharacterRange::Range(0x0e, 0x2027)); + ranges->append(CharacterRange::Range(0x202A, unicode::LeadSurrogateMin - 1)); + ranges->append(CharacterRange::Range(unicode::TrailSurrogateMax + 1, unicode::UTF16Max)); + builder->AddAtom(alloc->newInfallible(ranges, false)); + + builder->NewAlternative(); + + builder->AddAtom(RangeAtom(alloc, unicode::LeadSurrogateMin, unicode::LeadSurrogateMax)); + builder->AddAtom(NegativeLookahead(alloc, unicode::TrailSurrogateMin, + unicode::TrailSurrogateMax)); + + builder->NewAlternative(); + + builder->AddAssertion(alloc->newInfallible( + RegExpAssertion::NOT_AFTER_LEAD_SURROGATE)); + builder->AddAtom(RangeAtom(alloc, unicode::TrailSurrogateMin, unicode::TrailSurrogateMax)); + + builder->NewAlternative(); + + builder->AddAtom(RangeAtom(alloc, unicode::LeadSurrogateMin, unicode::LeadSurrogateMax)); + builder->AddAtom(RangeAtom(alloc, unicode::TrailSurrogateMin, unicode::TrailSurrogateMax)); + + return builder->ToRegExp(); +} + +RegExpTree* +UnicodeCharacterClassEscapeAtom(LifoAlloc* alloc, char16_t char_class, bool ignore_case) +{ + CharacterRangeVector* ranges = alloc->newInfallible(*alloc); + CharacterRangeVector* lead_ranges = alloc->newInfallible(*alloc); + CharacterRangeVector* trail_ranges = alloc->newInfallible(*alloc); + WideCharRangeVector* wide_ranges = alloc->newInfallible(*alloc); + AddCharOrEscapeUnicode(alloc, ranges, lead_ranges, trail_ranges, wide_ranges, char_class, 0, + ignore_case); + + return UnicodeRangesAtom(alloc, ranges, lead_ranges, trail_ranges, wide_ranges, false, false); +} + +static inline RegExpTree* +UnicodeBackReferenceAtom(LifoAlloc* alloc, RegExpTree* atom) +{ + // If a back reference has a standalone lead surrogate as its last + // character, then that lead surrogate shouldn't match lead surrogates that + // are paired with a corresponding trail surrogate. + RegExpBuilder* builder = alloc->newInfallible(alloc); + + builder->AddAtom(atom); + builder->AddAssertion(alloc->newInfallible( + RegExpAssertion::NOT_IN_SURROGATE_PAIR)); + + return builder->ToRegExp(); +} + +// Disjunction :: +// Alternative +// Alternative | Disjunction +// Alternative :: +// [empty] +// Term Alternative +// Term :: +// Assertion +// Atom +// Atom Quantifier +template +RegExpTree* +RegExpParser::ParseDisjunction() +{ + // Used to store current state while parsing subexpressions. + RegExpParserState initial_state(alloc, nullptr, INITIAL, 0); + RegExpParserState* stored_state = &initial_state; + // Cache the builder in a local variable for quick access. + RegExpBuilder* builder = initial_state.builder(); + while (true) { + switch (current()) { + case kEndMarker: + if (stored_state->IsSubexpression()) { + // Inside a parenthesized group when hitting end of input. + return ReportError(JSMSG_MISSING_PAREN); + } + MOZ_ASSERT(INITIAL == stored_state->group_type()); + // Parsing completed successfully. + return builder->ToRegExp(); + case ')': { + if (!stored_state->IsSubexpression()) + return ReportError(JSMSG_UNMATCHED_RIGHT_PAREN); + MOZ_ASSERT(INITIAL != stored_state->group_type()); + + Advance(); + // End disjunction parsing and convert builder content to new single + // regexp atom. + RegExpTree* body = builder->ToRegExp(); + + int end_capture_index = captures_started(); + + int capture_index = stored_state->capture_index(); + SubexpressionType group_type = stored_state->group_type(); + + // Restore previous state. + stored_state = stored_state->previous_state(); + builder = stored_state->builder(); + + // Build result of subexpression. + if (group_type == CAPTURE) { + RegExpCapture* capture = alloc->newInfallible(body, capture_index); + (*captures_)[capture_index - 1] = capture; + body = capture; + } else if (group_type != GROUPING) { + MOZ_ASSERT(group_type == POSITIVE_LOOKAHEAD || + group_type == NEGATIVE_LOOKAHEAD); + bool is_positive = (group_type == POSITIVE_LOOKAHEAD); + body = alloc->newInfallible(body, + is_positive, + end_capture_index - capture_index, + capture_index); + } + builder->AddAtom(body); + if (unicode_ && (group_type == POSITIVE_LOOKAHEAD || group_type == NEGATIVE_LOOKAHEAD)) + continue; + // For compatability with JSC and ES3, we allow quantifiers after + // lookaheads, and break in all cases. + break; + } + case '|': { + Advance(); + builder->NewAlternative(); + continue; + } + case '*': + case '+': + case '?': + return ReportError(JSMSG_NOTHING_TO_REPEAT); + case '^': { + Advance(); + if (multiline_) { + builder->AddAssertion(alloc->newInfallible(RegExpAssertion::START_OF_LINE)); + } else { + builder->AddAssertion(alloc->newInfallible(RegExpAssertion::START_OF_INPUT)); + set_contains_anchor(); + } + continue; + } + case '$': { + Advance(); + RegExpAssertion::AssertionType assertion_type = + multiline_ ? RegExpAssertion::END_OF_LINE : + RegExpAssertion::END_OF_INPUT; + builder->AddAssertion(alloc->newInfallible(assertion_type)); + continue; + } + case '.': { + Advance(); + // everything except \x0a, \x0d, \u2028 and \u2029 + if (unicode_) { + builder->AddAtom(UnicodeEverythingAtom(alloc)); + break; + } + CharacterRangeVector* ranges = alloc->newInfallible(*alloc); + CharacterRange::AddClassEscape(alloc, '.', ranges); + RegExpTree* atom = alloc->newInfallible(ranges, false); + builder->AddAtom(atom); + break; + } + case '(': { + SubexpressionType subexpr_type = CAPTURE; + Advance(); + if (current() == '?') { + switch (Next()) { + case ':': + subexpr_type = GROUPING; + break; + case '=': + subexpr_type = POSITIVE_LOOKAHEAD; + break; + case '!': + subexpr_type = NEGATIVE_LOOKAHEAD; + break; + default: + return ReportError(JSMSG_INVALID_GROUP); + } + Advance(2); + } else { + if (captures_ == nullptr) + captures_ = alloc->newInfallible(*alloc); + if (captures_started() >= kMaxCaptures) + return ReportError(JSMSG_TOO_MANY_PARENS); + captures_->append((RegExpCapture*) nullptr); + } + // Store current state and begin new disjunction parsing. + stored_state = alloc->newInfallible(alloc, stored_state, subexpr_type, + captures_started()); + builder = stored_state->builder(); + continue; + } + case '[': { + RegExpTree* atom = ParseCharacterClass(); + if (!atom) + return nullptr; + builder->AddAtom(atom); + break; + } + // Atom :: + // \ AtomEscape + case '\\': + switch (Next()) { + case kEndMarker: + return ReportError(JSMSG_ESCAPE_AT_END_OF_REGEXP); + case 'b': + Advance(2); + builder->AddAssertion(alloc->newInfallible(RegExpAssertion::BOUNDARY)); + continue; + case 'B': + Advance(2); + builder->AddAssertion(alloc->newInfallible(RegExpAssertion::NON_BOUNDARY)); + continue; + // AtomEscape :: + // CharacterClassEscape + // + // CharacterClassEscape :: one of + // d D s S w W + case 'D': case 'S': case 'W': + if (unicode_) { + Advance(); + builder->AddAtom(UnicodeCharacterClassEscapeAtom(alloc, current(), + ignore_case_)); + Advance(); + break; + } + MOZ_FALLTHROUGH; + case 'd': case 's': case 'w': { + widechar c = Next(); + Advance(2); + CharacterRangeVector* ranges = + alloc->newInfallible(*alloc); + if (unicode_) + CharacterRange::AddClassEscapeUnicode(alloc, c, ranges, ignore_case_); + else + CharacterRange::AddClassEscape(alloc, c, ranges); + RegExpTree* atom = alloc->newInfallible(ranges, false); + builder->AddAtom(atom); + break; + } + case '1': case '2': case '3': case '4': case '5': case '6': + case '7': case '8': case '9': { + int index = 0; + if (ParseBackReferenceIndex(&index)) { + RegExpCapture* capture = nullptr; + if (captures_ != nullptr && index <= (int) captures_->length()) { + capture = (*captures_)[index - 1]; + } + if (capture == nullptr) { + builder->AddEmpty(); + break; + } + RegExpTree* atom = alloc->newInfallible(capture); + if (unicode_) + builder->AddAtom(UnicodeBackReferenceAtom(alloc, atom)); + else + builder->AddAtom(atom); + break; + } + if (unicode_) + return ReportError(JSMSG_BACK_REF_OUT_OF_RANGE); + widechar first_digit = Next(); + if (first_digit == '8' || first_digit == '9') { + // Treat as identity escape + builder->AddCharacter(first_digit); + Advance(2); + break; + } + MOZ_FALLTHROUGH; + } + case '0': { + if (unicode_) { + Advance(2); + if (IsDecimalDigit(current())) + return ReportError(JSMSG_INVALID_DECIMAL_ESCAPE); + builder->AddCharacter(0); + break; + } + + Advance(); + widechar octal = ParseOctalLiteral(); + builder->AddCharacter(octal); + break; + } + // ControlEscape :: one of + // f n r t v + case 'f': + Advance(2); + builder->AddCharacter('\f'); + break; + case 'n': + Advance(2); + builder->AddCharacter('\n'); + break; + case 'r': + Advance(2); + builder->AddCharacter('\r'); + break; + case 't': + Advance(2); + builder->AddCharacter('\t'); + break; + case 'v': + Advance(2); + builder->AddCharacter('\v'); + break; + case 'c': { + Advance(); + widechar controlLetter = Next(); + // Special case if it is an ASCII letter. + // Convert lower case letters to uppercase. + widechar letter = controlLetter & ~('a' ^ 'A'); + if (letter < 'A' || 'Z' < letter) { + if (unicode_) + return ReportError(JSMSG_INVALID_IDENTITY_ESCAPE); + // controlLetter is not in range 'A'-'Z' or 'a'-'z'. + // This is outside the specification. We match JSC in + // reading the backslash as a literal character instead + // of as starting an escape. + builder->AddCharacter('\\'); + } else { + Advance(2); + builder->AddCharacter(controlLetter & 0x1f); + } + break; + } + case 'x': { + Advance(2); + widechar value; + if (ParseHexEscape(2, &value)) { + builder->AddCharacter(value); + } else { + if (unicode_) + return ReportError(JSMSG_INVALID_IDENTITY_ESCAPE); + builder->AddCharacter('x'); + } + break; + } + case 'u': { + Advance(2); + widechar value; + if (unicode_) { + if (current() == '{') { + if (!ParseBracedHexEscape(&value)) + return nullptr; + if (unicode::IsLeadSurrogate(value)) { + builder->AddAtom(LeadSurrogateAtom(alloc, value)); + } else if (unicode::IsTrailSurrogate(value)) { + builder->AddAtom(TrailSurrogateAtom(alloc, value)); + } else if (value >= unicode::NonBMPMin) { + char16_t lead, trail; + unicode::UTF16Encode(value, &lead, &trail); + builder->AddAtom(SurrogatePairAtom(alloc, lead, trail, + ignore_case_)); + } else { + builder->AddCharacter(value); + } + } else if (ParseHexEscape(4, &value)) { + if (unicode::IsLeadSurrogate(value)) { + widechar trail; + if (ParseTrailSurrogate(&trail)) { + builder->AddAtom(SurrogatePairAtom(alloc, value, trail, + ignore_case_)); + } else { + builder->AddAtom(LeadSurrogateAtom(alloc, value)); + } + } else if (unicode::IsTrailSurrogate(value)) { + builder->AddAtom(TrailSurrogateAtom(alloc, value)); + } else { + builder->AddCharacter(value); + } + } else { + return ReportError(JSMSG_INVALID_UNICODE_ESCAPE); + } + break; + } + if (ParseHexEscape(4, &value)) { + builder->AddCharacter(value); + } else { + builder->AddCharacter('u'); + } + break; + } + default: + // Identity escape. + if (unicode_ && !IsSyntaxCharacter(Next())) + return ReportError(JSMSG_INVALID_IDENTITY_ESCAPE); + builder->AddCharacter(Next()); + Advance(2); + break; + } + break; + case '{': { + if (unicode_) + return ReportError(JSMSG_RAW_BRACE_IN_REGEP); + int dummy; + if (ParseIntervalQuantifier(&dummy, &dummy)) + return ReportError(JSMSG_NOTHING_TO_REPEAT); + MOZ_FALLTHROUGH; + } + default: + if (unicode_) { + char16_t lead, trail; + if (ParseRawSurrogatePair(&lead, &trail)) { + builder->AddAtom(SurrogatePairAtom(alloc, lead, trail, ignore_case_)); + } else { + widechar c = current(); + if (unicode::IsLeadSurrogate(c)) + builder->AddAtom(LeadSurrogateAtom(alloc, c)); + else if (unicode::IsTrailSurrogate(c)) + builder->AddAtom(TrailSurrogateAtom(alloc, c)); + else if (c == ']') + return ReportError(JSMSG_RAW_BRACKET_IN_REGEP); + else if (c == '}') + return ReportError(JSMSG_RAW_BRACE_IN_REGEP); + else + builder->AddCharacter(c); + Advance(); + } + break; + } + builder->AddCharacter(current()); + Advance(); + break; + } // end switch(current()) + + int min; + int max; + switch (current()) { + // QuantifierPrefix :: + // * + // + + // ? + // { + case '*': + min = 0; + max = RegExpTree::kInfinity; + Advance(); + break; + case '+': + min = 1; + max = RegExpTree::kInfinity; + Advance(); + break; + case '?': + min = 0; + max = 1; + Advance(); + break; + case '{': + if (ParseIntervalQuantifier(&min, &max)) { + if (max < min) + return ReportError(JSMSG_NUMBERS_OUT_OF_ORDER); + break; + } else { + continue; + } + default: + continue; + } + RegExpQuantifier::QuantifierType quantifier_type = RegExpQuantifier::GREEDY; + if (current() == '?') { + quantifier_type = RegExpQuantifier::NON_GREEDY; + Advance(); + } + builder->AddQuantifierToAtom(min, max, quantifier_type); + } +} + +template class irregexp::RegExpParser; +template class irregexp::RegExpParser; + +template +static bool +ParsePattern(frontend::TokenStream& ts, LifoAlloc& alloc, const CharT* chars, size_t length, + bool multiline, bool match_only, bool unicode, bool ignore_case, + bool global, bool sticky, RegExpCompileData* data) +{ + if (match_only) { + // Try to strip a leading '.*' from the RegExp, but only if it is not + // followed by a '?' (which will affect how the .* is parsed). This + // pattern will affect the captures produced by the RegExp, but not + // whether there is a match or not. + if (length >= 3 && chars[0] == '.' && chars[1] == '*' && chars[2] != '?') { + chars += 2; + length -= 2; + } + + // Try to strip a trailing '.*' from the RegExp, which as above will + // affect the captures but not whether there is a match. Only do this + // when the following conditions are met: + // 1. there are no other meta characters in the RegExp, so that we + // are sure this will not affect how the RegExp is parsed + // 2. global and sticky flags are not set, as lastIndex needs to be + // set properly on global or sticky match + if (length >= 3 && !HasRegExpMetaChars(chars, length - 2) && + !global && !sticky && + chars[length - 2] == '.' && chars[length - 1] == '*') + { + length -= 2; + } + } + + RegExpParser parser(ts, &alloc, chars, chars + length, multiline, unicode, ignore_case); + data->tree = parser.ParsePattern(); + if (!data->tree) + return false; + + data->simple = parser.simple(); + data->contains_anchor = parser.contains_anchor(); + data->capture_count = parser.captures_started(); + return true; +} + +bool +irregexp::ParsePattern(frontend::TokenStream& ts, LifoAlloc& alloc, JSAtom* str, + bool multiline, bool match_only, bool unicode, bool ignore_case, + bool global, bool sticky, RegExpCompileData* data) +{ + JS::AutoCheckCannotGC nogc; + return str->hasLatin1Chars() + ? ::ParsePattern(ts, alloc, str->latin1Chars(nogc), str->length(), + multiline, match_only, unicode, ignore_case, global, sticky, data) + : ::ParsePattern(ts, alloc, str->twoByteChars(nogc), str->length(), + multiline, match_only, unicode, ignore_case, global, sticky, data); +} + +template +static bool +ParsePatternSyntax(frontend::TokenStream& ts, LifoAlloc& alloc, const CharT* chars, size_t length, + bool unicode) +{ + LifoAllocScope scope(&alloc); + + RegExpParser parser(ts, &alloc, chars, chars + length, false, unicode, false); + return parser.ParsePattern() != nullptr; +} + +bool +irregexp::ParsePatternSyntax(frontend::TokenStream& ts, LifoAlloc& alloc, JSAtom* str, + bool unicode) +{ + JS::AutoCheckCannotGC nogc; + return str->hasLatin1Chars() + ? ::ParsePatternSyntax(ts, alloc, str->latin1Chars(nogc), str->length(), unicode) + : ::ParsePatternSyntax(ts, alloc, str->twoByteChars(nogc), str->length(), unicode); +} diff --git a/js/src/irregexp/RegExpParser.h b/js/src/irregexp/RegExpParser.h new file mode 100644 index 000000000..b5228a86f --- /dev/null +++ b/js/src/irregexp/RegExpParser.h @@ -0,0 +1,310 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- + * vim: set ts=8 sts=4 et sw=4 tw=99: */ + +// Copyright 2012 the V8 project authors. All rights reserved. +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived +// from this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#ifndef V8_PARSER_H_ +#define V8_PARSER_H_ + +#include "irregexp/RegExpAST.h" + +namespace js { + +namespace frontend { + class TokenStream; +} + +namespace irregexp { + +bool +ParsePattern(frontend::TokenStream& ts, LifoAlloc& alloc, JSAtom* str, + bool multiline, bool match_only, bool unicode, bool ignore_case, + bool global, bool sticky, RegExpCompileData* data); + +bool +ParsePatternSyntax(frontend::TokenStream& ts, LifoAlloc& alloc, JSAtom* str, + bool unicode); + +// A BufferedVector is an automatically growing list, just like (and backed +// by) a Vector, that is optimized for the case of adding and removing +// a single element. The last element added is stored outside the backing list, +// and if no more than one element is ever added, the ZoneList isn't even +// allocated. +// Elements must not be nullptr pointers. +template +class BufferedVector +{ + public: + typedef InfallibleVector VectorType; + + BufferedVector() : list_(nullptr), last_(nullptr) {} + + // Adds element at end of list. This element is buffered and can + // be read using last() or removed using RemoveLast until a new Add or until + // RemoveLast or GetList has been called. + void Add(LifoAlloc* alloc, T* value) { + if (last_ != nullptr) { + if (list_ == nullptr) { + list_ = alloc->newInfallible(*alloc); + list_->reserve(initial_size); + } + list_->append(last_); + } + last_ = value; + } + + T* last() { + MOZ_ASSERT(last_ != nullptr); + return last_; + } + + T* RemoveLast() { + MOZ_ASSERT(last_ != nullptr); + T* result = last_; + if ((list_ != nullptr) && (list_->length() > 0)) + last_ = list_->popCopy(); + else + last_ = nullptr; + return result; + } + + T* Get(int i) { + MOZ_ASSERT((0 <= i) && (i < length())); + if (list_ == nullptr) { + MOZ_ASSERT(0 == i); + return last_; + } else { + if (size_t(i) == list_->length()) { + MOZ_ASSERT(last_ != nullptr); + return last_; + } else { + return (*list_)[i]; + } + } + } + + void Clear() { + list_ = nullptr; + last_ = nullptr; + } + + int length() { + int length = (list_ == nullptr) ? 0 : list_->length(); + return length + ((last_ == nullptr) ? 0 : 1); + } + + VectorType* GetList(LifoAlloc* alloc) { + if (list_ == nullptr) + list_ = alloc->newInfallible(*alloc); + if (last_ != nullptr) { + list_->append(last_); + last_ = nullptr; + } + return list_; + } + + private: + VectorType* list_; + T* last_; +}; + + +// Accumulates RegExp atoms and assertions into lists of terms and alternatives. +class RegExpBuilder +{ + public: + explicit RegExpBuilder(LifoAlloc* alloc); + void AddCharacter(char16_t character); + // "Adds" an empty expression. Does nothing except consume a + // following quantifier + void AddEmpty(); + void AddAtom(RegExpTree* tree); + void AddAssertion(RegExpTree* tree); + void NewAlternative(); // '|' + void AddQuantifierToAtom(int min, int max, RegExpQuantifier::QuantifierType type); + RegExpTree* ToRegExp(); + + private: + void FlushCharacters(); + void FlushText(); + void FlushTerms(); + + LifoAlloc* alloc; + bool pending_empty_; + CharacterVector* characters_; + BufferedVector terms_; + BufferedVector text_; + BufferedVector alternatives_; + + enum LastAdded { + ADD_NONE, ADD_CHAR, ADD_TERM, ADD_ASSERT, ADD_ATOM + }; +#ifdef DEBUG + LastAdded last_added_; +#endif +}; + +// Characters parsed by RegExpParser can be either char16_t or kEndMarker. +typedef uint32_t widechar; + +template +class RegExpParser +{ + public: + RegExpParser(frontend::TokenStream& ts, LifoAlloc* alloc, + const CharT* chars, const CharT* end, bool multiline_mode, bool unicode, + bool ignore_case); + + RegExpTree* ParsePattern(); + RegExpTree* ParseDisjunction(); + RegExpTree* ParseCharacterClass(); + + // Parses a {...,...} quantifier and stores the range in the given + // out parameters. + bool ParseIntervalQuantifier(int* min_out, int* max_out); + + // Tries to parse the input as a single escaped character. If successful + // it stores the result in the output parameter and returns true. + // Otherwise it throws an error and returns false. The character must not + // be 'b' or 'B' since they are usually handled specially. + bool ParseClassCharacterEscape(widechar* code); + + // Checks whether the following is a length-digit hexadecimal number, + // and sets the value if it is. + bool ParseHexEscape(int length, widechar* value); + + bool ParseBracedHexEscape(widechar* value); + bool ParseTrailSurrogate(widechar* value); + bool ParseRawSurrogatePair(char16_t* lead, char16_t* trail); + + widechar ParseOctalLiteral(); + + // Tries to parse the input as a back reference. If successful it + // stores the result in the output parameter and returns true. If + // it fails it will push back the characters read so the same characters + // can be reparsed. + bool ParseBackReferenceIndex(int* index_out); + + bool ParseClassAtom(char16_t* char_class, widechar *value); + RegExpTree* ReportError(unsigned errorNumber); + void Advance(); + void Advance(int dist) { + next_pos_ += dist - 1; + Advance(); + } + + void Reset(const CharT* pos) { + next_pos_ = pos; + has_more_ = (pos < end_); + Advance(); + } + + // Reports whether the pattern might be used as a literal search string. + // Only use if the result of the parse is a single atom node. + bool simple() { return simple_; } + bool contains_anchor() { return contains_anchor_; } + void set_contains_anchor() { contains_anchor_ = true; } + int captures_started() { return captures_ == nullptr ? 0 : captures_->length(); } + const CharT* position() { return next_pos_ - 1; } + + static const int kMaxCaptures = 1 << 16; + static const widechar kEndMarker = (1 << 21); + + private: + enum SubexpressionType { + INITIAL, + CAPTURE, // All positive values represent captures. + POSITIVE_LOOKAHEAD, + NEGATIVE_LOOKAHEAD, + GROUPING + }; + + class RegExpParserState { + public: + RegExpParserState(LifoAlloc* alloc, + RegExpParserState* previous_state, + SubexpressionType group_type, + int disjunction_capture_index) + : previous_state_(previous_state), + builder_(alloc->newInfallible(alloc)), + group_type_(group_type), + disjunction_capture_index_(disjunction_capture_index) + {} + // Parser state of containing expression, if any. + RegExpParserState* previous_state() { return previous_state_; } + bool IsSubexpression() { return previous_state_ != nullptr; } + // RegExpBuilder building this regexp's AST. + RegExpBuilder* builder() { return builder_; } + // Type of regexp being parsed (parenthesized group or entire regexp). + SubexpressionType group_type() { return group_type_; } + // Index in captures array of first capture in this sub-expression, if any. + // Also the capture index of this sub-expression itself, if group_type + // is CAPTURE. + int capture_index() { return disjunction_capture_index_; } + + private: + // Linked list implementation of stack of states. + RegExpParserState* previous_state_; + // Builder for the stored disjunction. + RegExpBuilder* builder_; + // Stored disjunction type (capture, look-ahead or grouping), if any. + SubexpressionType group_type_; + // Stored disjunction's capture index (if any). + int disjunction_capture_index_; + }; + + widechar current() { return current_; } + bool has_more() { return has_more_; } + bool has_next() { return next_pos_ < end_; } + widechar Next() { + if (has_next()) + return *next_pos_; + return kEndMarker; + } + void ScanForCaptures(); + + frontend::TokenStream& ts; + LifoAlloc* alloc; + RegExpCaptureVector* captures_; + const CharT* next_pos_; + const CharT* end_; + widechar current_; + // The capture count is only valid after we have scanned for captures. + int capture_count_; + bool has_more_; + bool multiline_; + bool unicode_; + bool ignore_case_; + bool simple_; + bool contains_anchor_; + bool is_scanned_for_captures_; +}; + +} } // namespace js::irregexp + +#endif // V8_PARSER_H_ diff --git a/js/src/irregexp/RegExpStack.cpp b/js/src/irregexp/RegExpStack.cpp new file mode 100644 index 000000000..d00b5feef --- /dev/null +++ b/js/src/irregexp/RegExpStack.cpp @@ -0,0 +1,106 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- + * vim: set ts=8 sts=4 et sw=4 tw=99: */ + +// Copyright 2012 the V8 project authors. All rights reserved. +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived +// from this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include "irregexp/RegExpStack.h" + +#include "vm/Runtime.h" + +using namespace js; +using namespace js::irregexp; + +RegExpStackScope::RegExpStackScope(JSRuntime* rt) + : regexp_stack(&rt->regexpStack) +{} + +RegExpStackScope::~RegExpStackScope() +{ + regexp_stack->reset(); +} + +bool +irregexp::GrowBacktrackStack(JSRuntime* rt) +{ + return rt->regexpStack.grow(); +} + +RegExpStack::RegExpStack() + : base_(nullptr), size(0), limit_(nullptr) +{} + +RegExpStack::~RegExpStack() +{ + js_free(base_); +} + +bool +RegExpStack::init() +{ + base_ = js_malloc(kMinimumStackSize); + if (!base_) + return false; + + size = kMinimumStackSize; + updateLimit(); + return true; +} + +void +RegExpStack::reset() +{ + MOZ_ASSERT(size >= kMinimumStackSize); + + if (size != kMinimumStackSize) { + void* newBase = js_realloc(base_, kMinimumStackSize); + if (!newBase) + return; + + base_ = newBase; + size = kMinimumStackSize; + updateLimit(); + } +} + +bool +RegExpStack::grow() +{ + size_t newSize = size * 2; + if (newSize > kMaximumStackSize) + return false; + + void* newBase = js_realloc(base_, newSize); + if (!newBase) + return false; + + base_ = newBase; + size = newSize; + updateLimit(); + + return true; +} diff --git a/js/src/irregexp/RegExpStack.h b/js/src/irregexp/RegExpStack.h new file mode 100644 index 000000000..411ce757c --- /dev/null +++ b/js/src/irregexp/RegExpStack.h @@ -0,0 +1,122 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- + * vim: set ts=8 sts=4 et sw=4 tw=99: */ + +// Copyright 2009 the V8 project authors. All rights reserved. +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived +// from this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#ifndef V8_REGEXP_STACK_H_ +#define V8_REGEXP_STACK_H_ + +#include "jspubtd.h" +#include "js/Utility.h" + +namespace js { +namespace irregexp { + +class RegExpStack; + +// Maintains a per-thread stack area that can be used by irregexp +// implementation for its backtracking stack. +// +// Since there is only one stack area, the Irregexp implementation is not +// re-entrant. I.e., no regular expressions may be executed in the same thread +// during a preempted Irregexp execution. +class RegExpStackScope +{ + public: + // Create and delete an instance to control the life-time of a growing stack. + + // Initializes the stack memory area if necessary. + explicit RegExpStackScope(JSRuntime* rt); + + // Releases the stack if it has grown. + ~RegExpStackScope(); + + private: + RegExpStack* regexp_stack; +}; + +class RegExpStack +{ + public: + // Number of allocated locations on the stack above the limit. + // No sequence of pushes must be longer that this without doing a stack-limit + // check. + static const int kStackLimitSlack = 32; + + RegExpStack(); + ~RegExpStack(); + bool init(); + + // Resets the buffer if it has grown beyond the default/minimum size. + void reset(); + + // Attempts to grow the stack by at least kStackLimitSlack entries. + bool grow(); + + // Address of allocated memory. + const void* addressOfBase() { return &base_; } + const void* addressOfLimit() { return &limit_; } + + void* base() { return base_; } + void* limit() { return limit_; } + + private: + // Artificial limit used when no memory has been allocated. + static const uintptr_t kMemoryTop = static_cast(-1); + + // Minimal size of allocated stack area, in bytes. + static const size_t kMinimumStackSize = 1 * 1024; + + // Maximal size of allocated stack area, in bytes. + static const size_t kMaximumStackSize = 64 * 1024 * 1024; + + // If size > 0 then base must be non-nullptr. + void* base_; + + // Length in bytes of base. + size_t size; + + // If the stack pointer gets above the limit, we should react and + // either grow the stack or report an out-of-stack exception. + // There is only a limited number of locations above the stack limit, + // so users of the stack should check the stack limit during any + // sequence of pushes longer than this. + void* limit_; + + void updateLimit() { + MOZ_ASSERT(size >= kStackLimitSlack * sizeof(void*)); + limit_ = static_cast(base()) + size - (kStackLimitSlack * sizeof(void*)); + } +}; + +bool +GrowBacktrackStack(JSRuntime* rt); + +}} // namespace js::irregexp + +#endif // V8_REGEXP_STACK_H_ -- cgit v1.2.3