1
Fork 0
mirror of https://github.com/RGBCube/serenity synced 2025-07-25 04:17:35 +00:00

LibRegex: Make infinite repetitions short-circuit on empty matches

This makes (addmittedly weird) patterns like `(a*)*` work correctly
without going into an infinite fork loop.
This commit is contained in:
Ali Mohammad Pur 2021-09-06 02:38:47 +04:30 committed by Ali Mohammad Pur
parent f8570bd773
commit abbe9da255
4 changed files with 131 additions and 69 deletions

View file

@ -27,6 +27,7 @@ using ByteCodeValueType = u64;
#define ENUMERATE_OPCODES \
__ENUMERATE_OPCODE(Compare) \
__ENUMERATE_OPCODE(Jump) \
__ENUMERATE_OPCODE(JumpNonEmpty) \
__ENUMERATE_OPCODE(ForkJump) \
__ENUMERATE_OPCODE(ForkStay) \
__ENUMERATE_OPCODE(FailForks) \
@ -42,6 +43,7 @@ using ByteCodeValueType = u64;
__ENUMERATE_OPCODE(ClearCaptureGroup) \
__ENUMERATE_OPCODE(Repeat) \
__ENUMERATE_OPCODE(ResetRepeat) \
__ENUMERATE_OPCODE(Checkpoint) \
__ENUMERATE_OPCODE(Exit)
// clang-format off
@ -319,16 +321,14 @@ public:
empend(static_cast<ByteCodeValueType>(OpCodeId::ForkJump));
empend(right.size() + 2); // Jump to the _ALT label
for (auto& op : right)
append(move(op));
extend(right);
empend(static_cast<ByteCodeValueType>(OpCodeId::Jump));
empend(left.size()); // Jump to the _END label
// LABEL _ALT = bytecode.size() + 2
for (auto& op : left)
append(move(op));
extend(left);
// LABEL _END = alterantive_bytecode.size
}
@ -376,10 +376,21 @@ public:
new_bytecode[pre_loop_fork_jump_index - 1] = (ByteCodeValueType)(fork_jump_address - pre_loop_fork_jump_index);
}
} else {
// no maximum value set, repeat finding if possible
// no maximum value set, repeat finding if possible:
// (REPEAT REGEXP MIN)
// LABEL _START
// CHECKPOINT _C
// REGEXP
// JUMP_NONEMPTY _C _START FORK
// Note: This is only safe because REPEAT will leave one iteration outside (see repetition_n)
new_bytecode.insert(new_bytecode.size() - bytecode_to_repeat.size(), (ByteCodeValueType)OpCodeId::Checkpoint);
auto jump_kind = static_cast<ByteCodeValueType>(greedy ? OpCodeId::ForkJump : OpCodeId::ForkStay);
new_bytecode.empend((ByteCodeValueType)OpCodeId::JumpNonEmpty);
new_bytecode.empend(-bytecode_to_repeat.size() - 4 - 1); // Jump to the last iteration
new_bytecode.empend(-bytecode_to_repeat.size() - 4 - 1); // if _C is not empty.
new_bytecode.empend(jump_kind);
new_bytecode.empend(-bytecode_to_repeat.size() - 2); // Jump to the last iteration
}
bytecode_to_repeat = move(new_bytecode);
@ -412,23 +423,29 @@ public:
static void transform_bytecode_repetition_min_one(ByteCode& bytecode_to_repeat, bool greedy)
{
// LABEL _START = -bytecode_to_repeat.size()
// CHECKPOINT _C
// REGEXP
// FORKSTAY _START (FORKJUMP -> Greedy)
// JUMP_NONEMPTY _C _START FORKSTAY (FORKJUMP -> Greedy)
bytecode_to_repeat.prepend((ByteCodeValueType)OpCodeId::Checkpoint);
bytecode_to_repeat.empend((ByteCodeValueType)OpCodeId::JumpNonEmpty);
bytecode_to_repeat.empend(-bytecode_to_repeat.size() - 3); // Jump to the _START label...
bytecode_to_repeat.empend(-bytecode_to_repeat.size() - 2); // ...if _C is not empty
if (greedy)
bytecode_to_repeat.empend(static_cast<ByteCodeValueType>(OpCodeId::ForkJump));
else
bytecode_to_repeat.empend(static_cast<ByteCodeValueType>(OpCodeId::ForkStay));
bytecode_to_repeat.empend(-(bytecode_to_repeat.size() + 1)); // Jump to the _START label
}
static void transform_bytecode_repetition_any(ByteCode& bytecode_to_repeat, bool greedy)
{
// LABEL _START
// FORKJUMP _END (FORKSTAY -> Greedy)
// CHECKPOINT _C
// REGEXP
// JUMP _START
// JUMP_NONEMPTY _C _START JUMP
// LABEL _END
// LABEL _START = m_bytes.size();
@ -439,13 +456,17 @@ public:
else
bytecode.empend(static_cast<ByteCodeValueType>(OpCodeId::ForkJump));
bytecode.empend(bytecode_to_repeat.size() + 2); // Jump to the _END label
bytecode.empend(bytecode_to_repeat.size() + 1 + 4); // Jump to the _END label
for (auto& op : bytecode_to_repeat)
bytecode.append(move(op));
auto c_label = bytecode.size();
bytecode.empend(static_cast<ByteCodeValueType>(OpCodeId::Checkpoint));
bytecode.empend(static_cast<ByteCodeValueType>(OpCodeId::Jump));
bytecode.empend(-bytecode.size() - 1); // Jump to the _START label
bytecode.extend(bytecode_to_repeat);
bytecode.empend(static_cast<ByteCodeValueType>(OpCodeId::JumpNonEmpty));
bytecode.empend(-bytecode.size() - 3); // Jump(...) to the _START label...
bytecode.empend(c_label - bytecode.size() - 2); // ...only if _C passes.
bytecode.empend((ByteCodeValueType)OpCodeId::Jump);
// LABEL _END = bytecode.size()
bytecode_to_repeat = move(bytecode);
@ -744,6 +765,31 @@ public:
}
};
class OpCode_Checkpoint final : public OpCode {
public:
ExecutionResult execute(MatchInput const& input, MatchState& state) const override;
ALWAYS_INLINE OpCodeId opcode_id() const override { return OpCodeId::Checkpoint; }
ALWAYS_INLINE size_t size() const override { return 1; }
String const arguments_string() const override { return ""; }
};
class OpCode_JumpNonEmpty final : public OpCode {
public:
ExecutionResult execute(MatchInput const& input, MatchState& state) const override;
ALWAYS_INLINE OpCodeId opcode_id() const override { return OpCodeId::JumpNonEmpty; }
ALWAYS_INLINE size_t size() const override { return 4; }
ALWAYS_INLINE ssize_t offset() const { return argument(0); }
ALWAYS_INLINE ssize_t checkpoint() const { return argument(1); }
ALWAYS_INLINE OpCodeId form() const { return (OpCodeId)argument(2); }
String const arguments_string() const override
{
return String::formatted("{} offset={} [&{}], cp={} [&{}]",
opcode_id_name(form()),
offset(), state().instruction_position + size() + offset(),
checkpoint(), state().instruction_position + size() + checkpoint());
}
};
template<typename T>
bool is(OpCode const&);