#ifndef regex_impl_hh_INCLUDED #define regex_impl_hh_INCLUDED #include "exception.hh" #include "flags.hh" #include "ref_ptr.hh" #include "unicode.hh" #include "utf8.hh" #include "vector.hh" #include "utils.hh" namespace Kakoune { struct regex_error : runtime_error { using runtime_error::runtime_error; }; enum class MatchDirection { Forward, Backward }; enum class CharacterType : unsigned char { None = 0, Whitespace = 1 << 0, HorizontalWhitespace = 1 << 1, Word = 1 << 2, Digit = 1 << 3, NotWhitespace = 1 << 4, NotHorizontalWhitespace = 1 << 5, NotWord = 1 << 6, NotDigit = 1 << 7 }; constexpr bool with_bit_ops(Meta::Type) { return true; } struct CharacterClass { struct Range { Codepoint min, max; }; Vector ranges; CharacterType ctypes = CharacterType::None; bool negative = false; bool ignore_case = false; }; bool is_character_class(const CharacterClass& character_class, Codepoint cp); bool is_ctype(CharacterType ctype, Codepoint cp); struct CompiledRegex : RefCountable, UseMemoryDomain { enum Op : char { Match, FindNextStart, Literal, Literal_IgnoreCase, AnyChar, AnyCharExceptNewLine, Class, CharacterType, Jump, Split_PrioritizeParent, Split_PrioritizeChild, Save, LineStart, LineEnd, WordBoundary, NotWordBoundary, SubjectBegin, SubjectEnd, LookAhead, NegativeLookAhead, LookBehind, NegativeLookBehind, LookAhead_IgnoreCase, NegativeLookAhead_IgnoreCase, LookBehind_IgnoreCase, NegativeLookBehind_IgnoreCase, }; enum class Lookaround : Codepoint { OpBegin = 0xF0000, AnyChar = 0xF0000, AnyCharExceptNewLine = 0xF0001, CharacterClass = 0xF0002, CharacterType = 0xF8000, OpEnd = 0xFFFFF, EndOfLookaround = static_cast(-1) }; struct Instruction { Op op; // Those mutables are used during execution mutable bool scheduled; mutable uint16_t last_step; uint32_t param; }; static_assert(sizeof(Instruction) == 8, ""); static constexpr uint16_t search_prefix_size = 3; explicit operator bool() const { return not instructions.empty(); } Vector instructions; Vector character_classes; Vector lookarounds; uint32_t first_backward_inst; // -1 if no backward support, 0 if only backward, >0 if both forward and backward uint32_t save_count; struct StartDesc : UseMemoryDomain { static constexpr Codepoint count = 128; static constexpr Codepoint other = 0; bool map[count]; }; std::unique_ptr forward_start_desc; std::unique_ptr backward_start_desc; }; String dump_regex(const CompiledRegex& program); enum class RegexCompileFlags { None = 0, NoSubs = 1 << 0, Optimize = 1 << 1, Backward = 1 << 2, NoForward = 1 << 3, }; constexpr bool with_bit_ops(Meta::Type) { return true; } CompiledRegex compile_regex(StringView re, RegexCompileFlags flags); enum class RegexExecFlags { None = 0, Search = 1 << 0, NotBeginOfLine = 1 << 1, NotEndOfLine = 1 << 2, NotBeginOfWord = 1 << 3, NotEndOfWord = 1 << 4, NotInitialNull = 1 << 5, AnyMatch = 1 << 6, NoSaves = 1 << 7, }; constexpr bool with_bit_ops(Meta::Type) { return true; } template struct SentinelType { using Type = It; }; template struct SentinelType> { using Type = typename It::Sentinel; }; template class ThreadedRegexVM { public: ThreadedRegexVM(const CompiledRegex& program) : m_program{program} { kak_assert((forward and program.first_backward_inst != 0) or (direction == MatchDirection::Backward and program.first_backward_inst != -1)); } ThreadedRegexVM(const ThreadedRegexVM&) = delete; ThreadedRegexVM& operator=(const ThreadedRegexVM&) = delete; ~ThreadedRegexVM() { for (auto* saves : m_saves) { for (size_t i = m_program.save_count-1; i > 0; --i) saves->pos[i].~Iterator(); saves->~Saves(); operator delete(saves); } } bool exec(Iterator begin, Iterator end, Iterator subject_begin, Iterator subject_end, RegexExecFlags flags) { if (flags & RegexExecFlags::NotInitialNull and begin == end) return false; const bool search = (flags & RegexExecFlags::Search); ConstArrayView instructions{m_program.instructions}; if (forward) instructions = instructions.subrange(0, m_program.first_backward_inst); else instructions = instructions.subrange(m_program.first_backward_inst); if (not search) instructions = instructions.subrange(CompiledRegex::search_prefix_size); const ExecConfig config{ Sentinel{forward ? begin : end}, Sentinel{forward ? end : begin}, Sentinel{subject_begin}, Sentinel{subject_end}, flags, instructions }; Iterator start = forward ? begin : end; if (const auto& start_desc = forward ? m_program.forward_start_desc : m_program.backward_start_desc) { if (search) { to_next_start(start, config, *start_desc); if (start == config.end) // If start_desc is not null, it means we consume at least one char return false; } else if (start != config.end) { const Codepoint cp = codepoint(start, config); if (not start_desc->map[cp < StartDesc::count ? cp : StartDesc::other]) return false; } } return exec_program(std::move(start), config); } ArrayView captures() const { if (m_captures >= 0) return { m_saves[m_captures]->pos, m_program.save_count }; return {}; } private: struct Saves { int16_t refcount; int16_t next_free; Iterator pos[1]; }; template int16_t new_saves(Iterator* pos) { kak_assert(not copy or pos != nullptr); const auto count = m_program.save_count; if (m_first_free >= 0) { const int16_t res = m_first_free; Saves& saves = *m_saves[res]; m_first_free = saves.next_free; kak_assert(saves.refcount == 1); if (copy) std::copy_n(pos, count, saves.pos); else std::fill_n(saves.pos, count, Iterator{}); return res; } void* ptr = operator new (sizeof(Saves) + (count-1) * sizeof(Iterator)); Saves* saves = new (ptr) Saves{1, 0, {copy ? pos[0] : Iterator{}}}; for (size_t i = 1; i < count; ++i) new (&saves->pos[i]) Iterator{copy ? pos[i] : Iterator{}}; m_saves.push_back(saves); return static_cast(m_saves.size() - 1); } void release_saves(int16_t index) { if (index < 0) return; auto& saves = *m_saves[index]; if (saves.refcount == 1) { saves.next_free = m_first_free; m_first_free = index; } else --saves.refcount; }; struct alignas(int32_t) Thread { int16_t inst; int16_t saves; }; using StartDesc = CompiledRegex::StartDesc; using Sentinel = typename SentinelType::Type; struct ExecConfig { const Sentinel begin; const Sentinel end; const Sentinel subject_begin; const Sentinel subject_end; const RegexExecFlags flags; ConstArrayView instructions; }; // Steps a thread until it consumes the current character, matches or fail void step_thread(const Iterator& pos, uint16_t current_step, Thread thread, const ExecConfig& config) { auto failed = [this](const Thread& thread) { release_saves(thread.saves); }; auto consumed = [this](const Thread& thread) { if (m_program.instructions[thread.inst].scheduled) return release_saves(thread.saves); m_program.instructions[thread.inst].scheduled = true; m_threads.push_next(thread); }; auto* instructions = m_program.instructions.data(); while (true) { auto& inst = instructions[thread.inst++]; // if this instruction was already executed for this step in another thread, // then this thread is redundant and can be dropped if (inst.last_step == current_step) return failed(thread); inst.last_step = current_step; switch (inst.op) { case CompiledRegex::Literal: if (pos != config.end and inst.param == codepoint(pos, config)) return consumed(thread); return failed(thread); case CompiledRegex::Literal_IgnoreCase: if (pos != config.end and inst.param == to_lower(codepoint(pos, config))) return consumed(thread); return failed(thread); case CompiledRegex::AnyChar: return consumed(thread); case CompiledRegex::AnyCharExceptNewLine: if (pos != config.end and codepoint(pos, config) != '\n') return consumed(thread); return failed(thread); case CompiledRegex::Jump: thread.inst = static_cast(inst.param); break; case CompiledRegex::Split_PrioritizeParent: { if (thread.saves >= 0) ++m_saves[thread.saves]->refcount; m_threads.push_current({static_cast(inst.param), thread.saves}); break; } case CompiledRegex::Split_PrioritizeChild: { if (thread.saves >= 0) ++m_saves[thread.saves]->refcount; m_threads.push_current({thread.inst, thread.saves}); thread.inst = static_cast(inst.param); break; } case CompiledRegex::Save: { if (config.flags & RegexExecFlags::NoSaves) break; if (thread.saves < 0) thread.saves = new_saves(nullptr); else if (m_saves[thread.saves]->refcount > 1) { --m_saves[thread.saves]->refcount; thread.saves = new_saves(m_saves[thread.saves]->pos); } m_saves[thread.saves]->pos[inst.param] = pos; break; } case CompiledRegex::Class: if (pos == config.end) return failed(thread); return is_character_class(m_program.character_classes[inst.param], codepoint(pos, config)) ? consumed(thread) : failed(thread); case CompiledRegex::CharacterType: if (pos == config.end) return failed(thread); return is_ctype((CharacterType)inst.param, codepoint(pos, config)) ? consumed(thread) : failed(thread); case CompiledRegex::LineStart: if (not is_line_start(pos, config)) return failed(thread); break; case CompiledRegex::LineEnd: if (not is_line_end(pos, config)) return failed(thread); break; case CompiledRegex::WordBoundary: if (not is_word_boundary(pos, config)) return failed(thread); break; case CompiledRegex::NotWordBoundary: if (is_word_boundary(pos, config)) return failed(thread); break; case CompiledRegex::SubjectBegin: if (pos != config.subject_begin) return failed(thread); break; case CompiledRegex::SubjectEnd: if (pos != config.subject_end) return failed(thread); break; case CompiledRegex::LookAhead: case CompiledRegex::NegativeLookAhead: if (lookaround(inst.param, pos, config) != (inst.op == CompiledRegex::LookAhead)) return failed(thread); break; case CompiledRegex::LookAhead_IgnoreCase: case CompiledRegex::NegativeLookAhead_IgnoreCase: if (lookaround(inst.param, pos, config) != (inst.op == CompiledRegex::LookAhead_IgnoreCase)) return failed(thread); break; case CompiledRegex::LookBehind: case CompiledRegex::NegativeLookBehind: if (lookaround(inst.param, pos, config) != (inst.op == CompiledRegex::LookBehind)) return failed(thread); break; case CompiledRegex::LookBehind_IgnoreCase: case CompiledRegex::NegativeLookBehind_IgnoreCase: if (lookaround(inst.param, pos, config) != (inst.op == CompiledRegex::LookBehind_IgnoreCase)) return failed(thread); break; case CompiledRegex::FindNextStart: // search thread should by construction be the lowest priority thread kak_assert(m_threads.current_is_empty()); if (not m_threads.next_is_empty()) return consumed(thread); m_threads.push_next(thread); m_find_next_start = true; return; case CompiledRegex::Match: if ((pos != config.end and not (config.flags & RegexExecFlags::Search)) or (config.flags & RegexExecFlags::NotInitialNull and pos == config.begin)) return failed(thread); release_saves(m_captures); m_captures = thread.saves; m_found_match = true; // remove lower priority threads while (not m_threads.current_is_empty()) release_saves(m_threads.pop_current().saves); return; } } return failed(thread); } bool exec_program(Iterator pos, const ExecConfig& config) { kak_assert(m_threads.current_is_empty() and m_threads.next_is_empty()); release_saves(m_captures); m_captures = -1; m_threads.grow_ifn(); m_threads.push_current({static_cast(&config.instructions[0] - &m_program.instructions[0]), -1}); const auto& start_desc = forward ? m_program.forward_start_desc : m_program.backward_start_desc; const bool any_match = config.flags & RegexExecFlags::AnyMatch; uint16_t current_step = -1; m_found_match = false; while (true) // Iterate on all codepoints and once at the end { if (++current_step == 0) { // We wrapped, avoid potential collision on inst.last_step by resetting them for (auto& inst : config.instructions) inst.last_step = 0; current_step = 1; // step 0 is never valid } m_find_next_start = false; while (not m_threads.current_is_empty()) step_thread(pos, current_step, m_threads.pop_current(), config); for (auto& thread : m_threads.next_threads()) m_program.instructions[thread.inst].scheduled = false; if (pos == config.end or m_threads.next_is_empty() or (m_found_match and any_match)) { for (auto& t : m_threads.next_threads()) release_saves(t.saves); m_threads.clear_next(); return m_found_match; } m_threads.swap_next(); forward ? utf8::to_next(pos, config.subject_end) : utf8::to_previous(pos, config.subject_begin); if (m_find_next_start and start_desc) to_next_start(pos, config, *start_desc); } } void to_next_start(Iterator& start, const ExecConfig& config, const StartDesc& start_desc) { while (start != config.end) { const Codepoint cp = read_codepoint(start, config); if (start_desc.map[(cp >= 0 and cp < StartDesc::count) ? cp : StartDesc::other]) { forward ? utf8::to_previous(start, config.subject_begin) : utf8::to_next(start, config.subject_end); return; } } } template bool lookaround(uint32_t index, Iterator pos, const ExecConfig& config) const { using Lookaround = CompiledRegex::Lookaround; if (look_direction == MatchDirection::Backward) { if (pos == config.subject_begin) return m_program.lookarounds[index] == Lookaround::EndOfLookaround; utf8::to_previous(pos, config.subject_begin); } for (auto it = m_program.lookarounds.begin() + index; *it != Lookaround::EndOfLookaround; ++it) { if (look_direction == MatchDirection::Forward and pos == config.subject_end) return false; Codepoint cp = utf8::codepoint(pos, config.subject_end); if (ignore_case) cp = to_lower(cp); const Lookaround op = *it; if (op == Lookaround::AnyChar) {} // any character matches else if (op == Lookaround::AnyCharExceptNewLine) { if (cp == '\n') return false; } else if (op >= Lookaround::CharacterClass and op < Lookaround::CharacterType) { auto index = to_underlying(op) - to_underlying(Lookaround::CharacterClass); if (not is_character_class(m_program.character_classes[index], cp)) return false; } else if (op >= Lookaround::CharacterType and op < Lookaround::OpEnd) { auto ctype = static_cast(to_underlying(op) & 0xFF); if (not is_ctype(ctype, cp)) return false; } else if (static_cast(op) != cp) return false; if (look_direction == MatchDirection::Backward and pos == config.subject_begin) return *++it == Lookaround::EndOfLookaround; (look_direction == MatchDirection::Forward) ? utf8::to_next(pos, config.subject_end) : utf8::to_previous(pos, config.subject_begin); } return true; } static bool is_line_start(const Iterator& pos, const ExecConfig& config) { if (pos == config.subject_begin) return not (config.flags & RegexExecFlags::NotBeginOfLine); return utf8::codepoint(utf8::previous(pos, config.subject_begin), config.subject_end) == '\n'; } static bool is_line_end(const Iterator& pos, const ExecConfig& config) { if (pos == config.subject_end) return not (config.flags & RegexExecFlags::NotEndOfLine); return utf8::codepoint(pos, config.subject_end) == '\n'; } static bool is_word_boundary(const Iterator& pos, const ExecConfig& config) { if (pos == config.subject_begin) return not (config.flags & RegexExecFlags::NotBeginOfWord); if (pos == config.subject_end) return not (config.flags & RegexExecFlags::NotEndOfWord); return is_word(utf8::codepoint(utf8::previous(pos, config.subject_begin), config.subject_end)) != is_word(utf8::codepoint(pos, config.subject_end)); } static Codepoint read_codepoint(Iterator& it, const ExecConfig& config) { if (forward) return utf8::read_codepoint(it, config.subject_end); else { utf8::to_previous(it, config.subject_begin); return utf8::codepoint(it, config.subject_end); } } static Codepoint codepoint(const Iterator& it, const ExecConfig& config) { return utf8::codepoint(forward ? it : utf8::previous(it, config.subject_begin), config.subject_end); } const CompiledRegex& m_program; struct DualThreadStack { DualThreadStack() = default; DualThreadStack(const DualThreadStack&) = delete; ~DualThreadStack() { delete[] m_data; } bool current_is_empty() const { return m_current == 0; } bool next_is_empty() const { return m_next == m_capacity; } void push_current(Thread thread) { kak_assert(m_current < m_next); m_data[m_current++] = thread; grow_ifn(); } Thread pop_current() { kak_assert(m_current > 0); return m_data[--m_current]; } void push_next(Thread thread) { kak_assert(m_current < m_next); m_data[--m_next] = thread; } void clear_next() { m_next = m_capacity; } ConstArrayView next_threads() const { return { m_data + m_next, m_data + m_capacity }; } void swap_next() { kak_assert(m_next < m_capacity); do { m_data[m_current++] = m_data[m_next++]; } while (m_next < m_capacity); } void grow_ifn() { if (m_current != m_next) return; const auto new_capacity = m_capacity ? m_capacity * 2 : 4; Thread* new_data = new Thread[new_capacity]; std::copy(m_data, m_data + m_current, new_data); const auto new_next = new_capacity - (m_capacity - m_next); std::copy(m_data + m_next, m_data + m_capacity, new_data + new_next); m_data = new_data; m_capacity = new_capacity; m_next = new_next; } private: Thread* m_data = nullptr; int32_t m_capacity = 0; // Maximum capacity should be 2*instruction count, so 65536 int32_t m_current = 0; int32_t m_next = 0; }; static constexpr bool forward = direction == MatchDirection::Forward; DualThreadStack m_threads; Vector m_saves; int16_t m_first_free = -1; int16_t m_captures = -1; bool m_found_match = false; bool m_find_next_start = false; }; } #endif // regex_impl_hh_INCLUDED