Rewrite std::filesystem::path iterators and parser

This patch entirely rewrites the parsing logic for paths. Unlike the previous
implementation this one stores information about the current state; For example
if we are in a trailing separator or a root separator. This avoids the need for
extra lookahead (and extra work) when incrementing or decrementing an iterator.
Roughly this gives us a 15% speedup over the previous implementation.

Unfortunately this implementation is still a lot slower than libstdc++'s.
Because libstdc++ pre-parses and splits the path upon construction their
iterators are trivial to increment/decrement. This makes libc++ lazy parsing
100x slower than libstdc++. However the pre-parsing libstdc++ causes a ton
of extra and unneeded allocations when constructing the string. For example
`path("/foo/bar/")` would require at least 5 allocations with libstdc++
whereas libc++ uses only one. The non-allocating behavior is much preferable
when you consider filesystem usages like 'exists("/foo/bar/")'.

Even then libc++'s path seems to be twice as slow to simply construct compared
to libstdc++. More investigation is needed about this.


git-svn-id: https://llvm.org/svn/llvm-project/libcxx/trunk@285526 91177308-0d34-0410-b5e6-96231b3b80d8

benchmarks/CMakeLists.txt

@@ -115,7 +115,8 @@ macro(add_benchmark_test name source_file)
   if (LIBCXX_BENCHMARK_NATIVE_STDLIB)
     set(native_target ${name}_native)
     add_executable(${native_target} EXCLUDE_FROM_ALL ${source_file})
-    add_dependencies(${native_target} google-benchmark-native)
+    add_dependencies(${native_target} google-benchmark-native
+                                      google-benchmark-libcxx)
     target_link_libraries(${native_target} -lbenchmark)
     if (LIBCXX_BENCHMARK_NATIVE_STDLIB STREQUAL "libstdc++")
       target_link_libraries(${native_target} -lstdc++fs)

include/experimental/filesystem

@@ -1083,7 +1083,8 @@ public:
     typedef const path&                reference;
 public:
     _LIBCPP_INLINE_VISIBILITY
-    iterator() : __elem_(), __path_ptr_(nullptr), __pos_(0) {}
+    iterator() : __stashed_elem_(), __path_ptr_(nullptr),
+                 __entry_(), __state_(__singular) {}
 
     iterator(const iterator&) = default;
     ~iterator() = default;
@@ -1092,16 +1093,20 @@ public:
 
     _LIBCPP_INLINE_VISIBILITY
     reference operator*() const {
-        return __elem_;
+        return __stashed_elem_;
     }
 
     _LIBCPP_INLINE_VISIBILITY
     pointer operator->() const {
-        return &__elem_;
+        return &__stashed_elem_;
     }
 
     _LIBCPP_INLINE_VISIBILITY
     iterator& operator++() {
+        _LIBCPP_ASSERT(__state_ != __singular,
+                       "attempting to increment a singular iterator");
+        _LIBCPP_ASSERT(__state_ != __at_end,
+                      "attempting to increment the end iterator");
         return __increment();
     }
 
@@ -1114,6 +1119,10 @@ public:
 
     _LIBCPP_INLINE_VISIBILITY
     iterator& operator--() {
+        _LIBCPP_ASSERT(__state_ != __singular,
+                       "attempting to decrement a singular iterator");
+        _LIBCPP_ASSERT(__entry_.data() != __path_ptr_->native().data(),
+                       "attempting to decrement the begin iterator");
         return __decrement();
     }
 
@@ -1127,21 +1136,25 @@ public:
 private:
     friend class path;
 
+    static constexpr unsigned char __singular = 0;
+    static constexpr unsigned char __at_end = 6;
+
     inline _LIBCPP_INLINE_VISIBILITY
     friend bool operator==(const iterator&, const iterator&);
 
     _LIBCPP_FUNC_VIS iterator& __increment();
     _LIBCPP_FUNC_VIS iterator& __decrement();
 
-    path __elem_;
+    path __stashed_elem_;
     const path* __path_ptr_;
-    size_t __pos_;
+    path::__string_view __entry_;
+    unsigned char __state_;
 };
 
 inline _LIBCPP_INLINE_VISIBILITY
 bool operator==(const path::iterator& __lhs, const path::iterator& __rhs) {
     return __lhs.__path_ptr_ == __rhs.__path_ptr_ &&
-           __lhs.__pos_      == __rhs.__pos_;
+           __lhs.__entry_.data() == __rhs.__entry_.data();
 }
 
 inline _LIBCPP_INLINE_VISIBILITY

src/experimental/filesystem/path.cpp

@@ -9,126 +9,255 @@
 #include "experimental/filesystem"
 #include "string_view"
 #include "utility"
-
+#include "cassert"
-_LIBCPP_BEGIN_NAMESPACE_EXPERIMENTAL_FILESYSTEM
-
-_LIBCPP_CONSTEXPR path::value_type path::preferred_separator;
-
-
-using string_view_t = path::__string_view;
 
 namespace { namespace parser
 {
+using namespace std;
+using namespace std::experimental::filesystem;
 
-using value_type = path::value_type;
+using string_view_t = path::__string_view;
 using string_view_pair = pair<string_view_t, string_view_t>;
+using PosPtr = path::value_type const*;
 
-// status reporting
+struct PathParser {
-constexpr size_t npos = static_cast<size_t>(-1);
+  enum ParserState : unsigned char {
+    // Zero is a special sentinel value used by default constructed iterators.
+    PS_BeforeBegin = 1,
+    PS_InRootName,
+    PS_InRootDir,
+    PS_InFilenames,
+    PS_InTrailingSep,
+    PS_AtEnd
+  };
 
-inline bool good(size_t pos) { return pos != npos; }
+  const string_view_t Path;
+  string_view_t RawEntry;
+  ParserState State;
 
-// lexical elements
+private:
-constexpr value_type preferred_separator = path::preferred_separator;
+  PathParser(string_view_t P, ParserState State) noexcept
-constexpr value_type const * preferred_separator_str = "/";
+      : Path(P), State(State) {}
-constexpr value_type const * dot = ".";
 
-// forward //
+public:
-bool is_separator(string_view_t const &, size_t);
+  PathParser(string_view_t P, string_view_t E, unsigned char S)
-bool is_root_name(const string_view_t&, size_t);
+      : Path(P), RawEntry(E), State(static_cast<ParserState>(S)) {
-bool is_root_directory(string_view_t const &, size_t);
+    assert(S != 0);
-bool is_trailing_separator(string_view_t const &, size_t);
+    assert(S != PS_BeforeBegin);
+  }
 
-size_t start_of(string_view_t const &, size_t);
+  static PathParser CreateBegin(string_view_t P) noexcept {
-size_t end_of(string_view_t const &, size_t);
+    PathParser PP(P, PS_BeforeBegin);
+    PP.increment();
+    return PP;
+  }
 
-size_t root_name_start(const string_view_t& s);
+  static PathParser CreateEnd(string_view_t P) noexcept {
-size_t root_name_end(const string_view_t&);
+    PathParser PP(P, PS_AtEnd);
+    return PP;
+  }
 
-size_t root_directory_start(string_view_t const &);
+  PosPtr peek() const noexcept {
-size_t root_directory_end(string_view_t const &);
+    auto End = &Path.back() + 1;
+    auto TkEnd = getNextTokenStartPos();
+    return TkEnd == End ? nullptr : TkEnd;
+  }
 
-string_view_pair separate_filename(string_view_t const &);
+  void increment() noexcept {
-string_view extract_raw(string_view_t const &, size_t);
+    const PosPtr End = &Path.back() + 1;
-string_view extract_preferred(string_view_t const &, size_t);
+    const PosPtr Start = getNextTokenStartPos();
+    if (Start == End)
+      return makeState(PS_AtEnd);
 
-inline bool is_separator(const string_view_t& s, size_t pos) {
+    switch (State) {
-    return (pos < s.size() && s[pos] == preferred_separator);
+    case PS_BeforeBegin: {
-}
+      PosPtr TkEnd = consumeSeparator(Start, End);
-
+      // If we consumed exactly two separators we have a root name.
-inline bool is_root_name(const string_view_t& s, size_t pos) {
+      if (TkEnd && TkEnd == Start + 2) {
-  return good(pos) && pos == 0 ? root_name_start(s) == pos : false;
+        // FIXME Do we need to consume a name or is '//' a root name on its own?
-}
+        // what about '//.', '//..', '//...'?
-
+        auto NameEnd = consumeName(TkEnd, End);
-inline bool is_root_directory(const string_view_t& s, size_t pos) {
+        if (NameEnd)
-    return good(pos) ? root_directory_start(s) == pos : false;
+          TkEnd = NameEnd;
-}
+        return makeState(PS_InRootName, Start, TkEnd);
-
+      }
-inline bool is_trailing_separator(const string_view_t& s, size_t pos) {
+      else if (TkEnd)
-    return (pos < s.size() && is_separator(s, pos) &&
+        return makeState(PS_InRootDir, Start, TkEnd);
-            end_of(s, pos) == s.size()-1 &&
+      else
-            !is_root_directory(s, pos) && !is_root_name(s, pos));
+        return makeState(PS_InFilenames, Start, consumeName(Start, End));
-}
-
-size_t start_of(const string_view_t& s, size_t pos) {
-    if (pos >= s.size()) return npos;
-    bool in_sep = (s[pos] == preferred_separator);
-    while (pos - 1 < s.size() &&
-        (s[pos-1] == preferred_separator) == in_sep)
-    { --pos; }
-    if (pos == 2 && !in_sep && s[0] == preferred_separator &&
-        s[1] == preferred_separator)
-    { return 0; }
-    return pos;
-}
-
-size_t end_of(const string_view_t& s, size_t pos) {
-    if (pos >= s.size()) return npos;
-    // special case for root name
-    if (pos == 0 && is_root_name(s, pos)) return root_name_end(s);
-    bool in_sep = (s[pos] == preferred_separator);
-    while (pos + 1 < s.size() && (s[pos+1] == preferred_separator) == in_sep)
-    { ++pos; }
-    return pos;
-}
-
-inline size_t root_name_start(const string_view_t& s) {
-    return good(root_name_end(s)) ? 0 : npos;
-}
-
-size_t root_name_end(const string_view_t& s) {
-    if (s.size() < 2 || s[0] != preferred_separator
-        || s[1] != preferred_separator) {
-        return npos;
     }
-    if (s.size() == 2) {
-        return 1;
-    }
-    size_t index = 2; // current position
-    if (s[index] == preferred_separator) {
-        return npos;
-    }
-    while (index + 1 < s.size() && s[index+1] != preferred_separator) {
-        ++index;
-    }
-    return index;
-}
 
-size_t root_directory_start(const string_view_t& s) {
+    case PS_InRootName:
-    size_t e = root_name_end(s);
+      return makeState(PS_InRootDir, Start, consumeSeparator(Start, End));
-    if (!good(e))
+    case PS_InRootDir:
-    return is_separator(s, 0) ? 0 : npos;
+      return makeState(PS_InFilenames, Start, consumeName(Start, End));
-    return is_separator(s, e + 1) ? e + 1 : npos;
-}
 
-size_t root_directory_end(const string_view_t& s) {
+    case PS_InFilenames: {
-    size_t st = root_directory_start(s);
+      PosPtr SepEnd = consumeSeparator(Start, End);
-    if (!good(st)) return npos;
+      assert(SepEnd);
-    size_t index = st;
+      if (SepEnd != End) {
-    while (index + 1 < s.size() && s[index + 1] == preferred_separator)
+        PosPtr TkEnd = consumeName(SepEnd, End);
-      { ++index; }
+        if (TkEnd)
-    return index;
+          return makeState(PS_InFilenames, SepEnd, TkEnd);
-}
+      }
+      return makeState(PS_InTrailingSep, Start, SepEnd);
+    }
+
+    case PS_InTrailingSep:
+      return makeState(PS_AtEnd);
+
+    case PS_AtEnd:
+      _LIBCPP_UNREACHABLE();
+    }
+  }
+
+  void decrement() noexcept {
+    const PosPtr REnd = &Path.front() - 1;
+    const PosPtr RStart = getCurrentTokenStartPos() - 1;
+    assert(RStart != REnd);
+
+    switch (State) {
+    case PS_AtEnd: {
+      // Try to consume a trailing separator or root directory first.
+      if (PosPtr SepEnd = consumeSeparator(RStart, REnd)) {
+        if (SepEnd == REnd)
+          return makeState((RStart == REnd + 2) ? PS_InRootName : PS_InRootDir,
+                           Path.data(), RStart + 1);
+        // Check if we're seeing the root directory separator
+        auto PP = CreateBegin(Path);
+        bool InRootDir = PP.State == PS_InRootName &&
+            &PP.RawEntry.back() == SepEnd;
+        return makeState(InRootDir ? PS_InRootDir : PS_InTrailingSep,
+                         SepEnd + 1, RStart + 1);
+      } else {
+        PosPtr TkStart = consumeName(RStart, REnd);
+        assert(TkStart);
+        if (TkStart == REnd + 2 && consumeSeparator(TkStart, REnd) == REnd)
+          return makeState(PS_InRootName, Path.data(), RStart + 1);
+        else
+          return makeState(PS_InFilenames, TkStart + 1, RStart + 1);
+      }
+    }
+    case PS_InTrailingSep:
+      return makeState(PS_InFilenames, consumeName(RStart, REnd) + 1, RStart + 1);
+    case PS_InFilenames: {
+      PosPtr SepEnd = consumeSeparator(RStart, REnd);
+      if (SepEnd == REnd)
+        return makeState((RStart == REnd + 2) ? PS_InRootName : PS_InRootDir,
+                         Path.data(), RStart + 1);
+      PosPtr TkEnd = consumeName(SepEnd, REnd);
+      if (TkEnd == REnd + 2 && consumeSeparator(TkEnd, REnd) == REnd)
+        return makeState(PS_InRootDir, SepEnd + 1, RStart + 1);
+      return makeState(PS_InFilenames, TkEnd + 1, SepEnd + 1);
+    }
+    case PS_InRootDir:
+      return makeState(PS_InRootName, Path.data(), RStart + 1);
+    case PS_InRootName:
+    case PS_BeforeBegin:
+      _LIBCPP_UNREACHABLE();
+    }
+  }
+
+  /// \brief Return a view with the "preferred representation" of the current
+  ///   element. For example trailing separators are represented as a '.'
+  string_view_t operator*() const noexcept {
+    switch (State) {
+    case PS_BeforeBegin:
+    case PS_AtEnd:
+      return "";
+    case PS_InRootDir:
+      return "/";
+    case PS_InTrailingSep:
+      return ".";
+    case PS_InRootName:
+    case PS_InFilenames:
+      return RawEntry;
+    }
+    _LIBCPP_UNREACHABLE();
+  }
+
+  explicit operator bool() const noexcept {
+    return State != PS_BeforeBegin && State != PS_AtEnd;
+  }
+
+  PathParser& operator++() noexcept {
+    increment();
+    return *this;
+  }
+
+  PathParser& operator--() noexcept {
+    decrement();
+    return *this;
+  }
+
+private:
+  void makeState(ParserState NewState, PosPtr Start, PosPtr End) noexcept {
+    assert(NewState != PS_BeforeBegin && NewState != PS_AtEnd);
+    State = NewState;
+    assert(Start < End);
+    assert(Start >= &Path.front() && End <= &Path.back() + 1);
+    RawEntry = string_view_t(Start, End - Start);
+  }
+  void makeState(ParserState NewState) noexcept {
+    assert(NewState == PS_BeforeBegin || NewState == PS_AtEnd);
+    State = NewState;
+    RawEntry = {};
+  }
+
+  /// \brief Return a pointer to the first character after the currently
+  ///   lexed element.
+  PosPtr getNextTokenStartPos() const noexcept {
+    switch (State) {
+    case PS_BeforeBegin:
+      return &Path.front();
+    case PS_InRootName:
+    case PS_InRootDir:
+    case PS_InFilenames:
+      return &RawEntry.back() + 1;
+    case PS_InTrailingSep:
+    case PS_AtEnd:
+      return &Path.back() + 1;
+    }
+    _LIBCPP_UNREACHABLE();
+  }
+
+  /// \brief Return a pointer to the first character in the currently lexed
+  ///   element.
+  PosPtr getCurrentTokenStartPos() const noexcept {
+    switch (State) {
+    case PS_BeforeBegin:
+    case PS_InRootName:
+      return &Path.front();
+    case PS_InRootDir:
+    case PS_InFilenames:
+    case PS_InTrailingSep:
+      return &RawEntry.front();
+    case PS_AtEnd:
+      return &Path.back() + 1;
+    }
+    _LIBCPP_UNREACHABLE();
+  }
+
+  PosPtr consumeSeparator(PosPtr P, PosPtr End) const noexcept {
+    if (P == End || *P != '/')
+      return nullptr;
+    const int Inc = P < End ? 1 : -1;
+    P += Inc;
+    while (P != End && *P == '/')
+      P += Inc;
+    return P;
+  }
+
+  PosPtr consumeName(PosPtr P, PosPtr End) const noexcept {
+    if (P == End || *P == '/')
+      return nullptr;
+    const int Inc = P < End ? 1 : -1;
+    P += Inc;
+    while (P != End && *P != '/')
+      P += Inc;
+    return P;
+  }
+};
 
 string_view_pair separate_filename(string_view_t const & s) {
     if (s == "." || s == ".." || s.empty()) return string_view_pair{s, ""};
@@ -137,112 +266,25 @@ string_view_pair separate_filename(string_view_t const & s) {
     return string_view_pair{s.substr(0, pos), s.substr(pos)};
 }
 
-inline string_view extract_raw(const string_view_t& s, size_t pos) {
+string_view_t createView(PosPtr S, PosPtr E) noexcept {
-    size_t end_i = end_of(s, pos);
+  return {S, static_cast<size_t>(E - S) + 1};
-    if (!good(end_i)) return string_view{};
-    return string_view(s).substr(pos, end_i - pos + 1);
-}
-
-string_view extract_preferred(const string_view_t& s, size_t pos) {
-    string_view raw = extract_raw(s, pos);
-    if (raw.empty())
-        return raw;
-    if (is_trailing_separator(s, pos))
-        return string_view{dot};
-    if (is_separator(s, pos) && !is_root_name(s, pos))
-        return string_view(preferred_separator_str);
-    return raw;
 }
 
 }} // namespace parser
 
+_LIBCPP_BEGIN_NAMESPACE_EXPERIMENTAL_FILESYSTEM
 
-////////////////////////////////////////////////////////////////////////////////
+using parser::string_view_t;
-//                            path_view_iterator
+using parser::string_view_pair;
-////////////////////////////////////////////////////////////////////////////////
+using parser::PathParser;
-namespace {
+using parser::createView;
 
-struct path_view_iterator {
-  const string_view __s_;
-  size_t __pos_;
-
-  explicit path_view_iterator(string_view const& __s) : __s_(__s), __pos_(__s_.empty() ? parser::npos : 0) {}
-  explicit path_view_iterator(string_view const& __s, size_t __p) : __s_(__s), __pos_(__p) {}
-
-  string_view operator*() const {
-    return parser::extract_preferred(__s_, __pos_);
-  }
-
-  path_view_iterator& operator++() {
-    increment();
-    return *this;
-  }
-
-  path_view_iterator& operator--() {
-    decrement();
-    return *this;
-  }
-
-  void increment() {
-    if (__pos_ == parser::npos) return;
-    while (! set_position(parser::end_of(__s_, __pos_)+1))
-        ;
-    return;
-  }
-
-  void decrement() {
-    if (__pos_ == 0) {
-      set_position(0);
-    }
-    else if (__pos_ == parser::npos) {
-      auto const str_size = __s_.size();
-      set_position(parser::start_of(
-          __s_, str_size != 0 ? str_size - 1 : str_size));
-    } else {
-      while (!set_position(parser::start_of(__s_, __pos_-1)))
-        ;
-    }
-  }
-
-  bool set_position(size_t pos) {
-    if (pos >= __s_.size()) {
-      __pos_ = parser::npos;
-    } else {
-      __pos_ = pos;
-    }
-    return valid_iterator_position();
-  }
-
-  bool valid_iterator_position() const {
-    if (__pos_ == parser::npos) return true; // end position is valid
-    return (!parser::is_separator      (__s_, __pos_) ||
-          parser::is_root_directory    (__s_, __pos_) ||
-          parser::is_trailing_separator(__s_, __pos_) ||
-          parser::is_root_name         (__s_, __pos_));
-  }
-
-  bool is_end() const { return __pos_ == parser::npos; }
-
-  inline bool operator==(path_view_iterator const& __p) {
-      return __pos_ == __p.__pos_;
-  }
-};
-
-path_view_iterator pbegin(path const& p) {
-    return path_view_iterator(p.native());
-}
-
-path_view_iterator pend(path const& p) {
-    path_view_iterator __p(p.native());
-    __p.__pos_ = parser::npos;
-    return __p;
-}
-
-} // end namespace
 ///////////////////////////////////////////////////////////////////////////////
 //                            path definitions
 ///////////////////////////////////////////////////////////////////////////////
 
+constexpr path::value_type path::preferred_separator;
+
 path & path::replace_extension(path const & replacement)
 {
     path p = extension();
@@ -263,58 +305,65 @@ path & path::replace_extension(path const & replacement)
 
 string_view_t path::__root_name() const
 {
-    return parser::is_root_name(__pn_, 0)
+    auto PP = PathParser::CreateBegin(__pn_);
-      ? parser::extract_preferred(__pn_, 0)
+    if (PP.State == PathParser::PS_InRootName)
-      : string_view_t{};
+      return *PP;
+    return {};
 }
 
 string_view_t path::__root_directory() const
 {
-    auto start_i = parser::root_directory_start(__pn_);
+    auto PP = PathParser::CreateBegin(__pn_);
-    if(!parser::good(start_i)) {
+    if (PP.State == PathParser::PS_InRootName)
-        return {};
+      ++PP;
-    }
+    if (PP.State == PathParser::PS_InRootDir)
-    return parser::extract_preferred(__pn_, start_i);
+      return *PP;
+    return {};
 }
 
 string_view_t path::__root_path_raw() const
 {
-    size_t e = parser::root_directory_end(__pn_);
+    auto PP = PathParser::CreateBegin(__pn_);
-    if (!parser::good(e))
+    if (PP.State == PathParser::PS_InRootName) {
-      e = parser::root_name_end(__pn_);
+      auto NextCh = PP.peek();
-    if (parser::good(e))
+      if (NextCh && *NextCh == '/') {
-      return string_view_t(__pn_).substr(0, e + 1);
+        ++PP;
+        assert(PP.State == PathParser::PS_InRootDir);
+        return createView(__pn_.data(), &PP.RawEntry.back());
+      }
+      return PP.RawEntry;
+    }
+    if (PP.State == PathParser::PS_InRootDir)
+      return *PP;
     return {};
 }
 
 string_view_t path::__relative_path() const
 {
-    if (empty()) {
+    auto PP = PathParser::CreateBegin(__pn_);
-        return __pn_;
+    while (PP.State <= PathParser::PS_InRootDir)
-    }
+      ++PP;
-    auto end_i = parser::root_directory_end(__pn_);
+    if (PP.State == PathParser::PS_AtEnd)
-    if (not parser::good(end_i)) {
+      return {};
-        end_i = parser::root_name_end(__pn_);
+    return createView(PP.RawEntry.data(), &__pn_.back());
-    }
-    if (not parser::good(end_i)) {
-        return __pn_;
-    }
-    return string_view_t(__pn_).substr(end_i+1);
 }
 
 string_view_t path::__parent_path() const
 {
-    if (empty() || pbegin(*this) == --pend(*this)) {
+    if (empty())
-        return {};
+      return {};
-    }
+    auto PP = PathParser::CreateEnd(__pn_);
-    auto end_it = --(--pend(*this));
+    --PP;
-    auto end_i = parser::end_of(__pn_, end_it.__pos_);
+    if (PP.RawEntry.data() == __pn_.data())
-    return string_view_t(__pn_).substr(0, end_i+1);
+      return {};
+    --PP;
+    return createView(__pn_.data(), &PP.RawEntry.back());
 }
 
 string_view_t path::__filename() const
 {
-    return empty() ? string_view_t{} : *--pend(*this);
+    if (empty()) return {};
+    return *(--PathParser::CreateEnd(__pn_));
 }
 
 string_view_t path::__stem() const
@@ -330,24 +379,24 @@ string_view_t path::__extension() const
 ////////////////////////////////////////////////////////////////////////////
 // path.comparisons
 int path::__compare(string_view_t __s) const {
-    path_view_iterator thisIter(this->native());
+    auto PP = PathParser::CreateBegin(__pn_);
-    path_view_iterator sIter(__s);
+    auto PP2 = PathParser::CreateBegin(__s);
-    while (!thisIter.is_end() && !sIter.is_end()) {
+    while (PP && PP2) {
-        int res = (*thisIter).compare(*sIter);
+        int res = (*PP).compare(*PP2);
         if (res != 0) return res;
-        ++thisIter; ++sIter;
+        ++PP; ++PP2;
     }
-    if (thisIter.is_end() && sIter.is_end())
+    if (PP.State == PP2.State && PP.State == PathParser::PS_AtEnd)
         return 0;
-    if (thisIter.is_end())
+    if (PP.State == PathParser::PS_AtEnd)
         return -1;
     return 1;
 }
 
 ////////////////////////////////////////////////////////////////////////////
 // path.nonmembers
-size_t hash_value(const path& __p) _NOEXCEPT {
+size_t hash_value(const path& __p) noexcept {
-  path_view_iterator thisIter(__p.native());
+  auto PP = PathParser::CreateBegin(__p.native());
   struct HashPairT {
     size_t first;
     size_t second;
@@ -355,10 +404,10 @@ size_t hash_value(const path& __p) _NOEXCEPT {
   HashPairT hp = {0, 0};
   std::hash<string_view> hasher;
   std::__scalar_hash<decltype(hp)> pair_hasher;
-  while (!thisIter.is_end()) {
+  while (PP) {
-    hp.second = hasher(*thisIter);
+    hp.second = hasher(*PP);
     hp.first = pair_hasher(hp);
-    ++thisIter;
+    ++PP;
   }
   return hp.first;
 }
@@ -367,35 +416,39 @@ size_t hash_value(const path& __p) _NOEXCEPT {
 // path.itr
 path::iterator path::begin() const
 {
-    path_view_iterator pit = pbegin(*this);
+    auto PP = PathParser::CreateBegin(__pn_);
     iterator it;
     it.__path_ptr_ = this;
-    it.__pos_ = pit.__pos_;
+    it.__state_ = PP.State;
-    it.__elem_.__assign_view(*pit);
+    it.__entry_ = PP.RawEntry;
+    it.__stashed_elem_.__assign_view(*PP);
     return it;
 }
 
 path::iterator path::end() const
 {
     iterator it{};
+    it.__state_ = PathParser::PS_AtEnd;
     it.__path_ptr_ = this;
-    it.__pos_ = parser::npos;
     return it;
 }
 
 path::iterator& path::iterator::__increment() {
-  path_view_iterator it(__path_ptr_->native(), __pos_);
+  static_assert(__at_end == PathParser::PS_AtEnd, "");
-  it.increment();
+  PathParser PP(__path_ptr_->native(), __entry_, __state_);
-  __pos_ = it.__pos_;
+  ++PP;
-  __elem_.__assign_view(*it);
+  __state_ = PP.State;
+  __entry_ = PP.RawEntry;
+  __stashed_elem_.__assign_view(*PP);
   return *this;
 }
 
 path::iterator& path::iterator::__decrement() {
-  path_view_iterator it(__path_ptr_->native(), __pos_);
+  PathParser PP(__path_ptr_->native(), __entry_, __state_);
-  it.decrement();
+  --PP;
-  __pos_ = it.__pos_;
+  __state_ = PP.State;
-  __elem_.__assign_view(*it);
+  __entry_ = PP.RawEntry;
+  __stashed_elem_.__assign_view(*PP);
   return *this;
 }
 

test/libcxx/experimental/filesystem/class.path/path.itr/iterator_db.pass.cpp

@@ -0,0 +1,74 @@
+//===----------------------------------------------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+// UNSUPPORTED: c++98, c++03
+// UNSUPPORTED: libcpp-no-exceptions
+
+// <experimental/filesystem>
+
+// class path
+
+#define _LIBCPP_DEBUG 0
+#define _LIBCPP_ASSERT(cond, msg) ((cond) ? ((void)0) : throw 42)
+
+#include <experimental/filesystem>
+#include <iterator>
+#include <type_traits>
+#include <cassert>
+
+#include "test_macros.h"
+#include "filesystem_test_helper.hpp"
+
+namespace fs = std::experimental::filesystem;
+
+int main() {
+  using namespace fs;
+  // Test incrementing/decrementing a singular iterator
+  {
+    path::iterator singular;
+    try {
+      ++singular;
+      assert(false);
+    } catch (int) {}
+    try {
+      --singular;
+      assert(false);
+    } catch (int) {}
+  }
+  // Test decrementing the begin iterator
+  {
+    path p("foo/bar");
+    auto it = p.begin();
+    try {
+      --it;
+      assert(false);
+    } catch (int) {}
+    ++it;
+    ++it;
+    try {
+      ++it;
+      assert(false);
+    } catch (int) {}
+  }
+  // Test incrementing the end iterator
+  {
+    path p("foo/bar");
+    auto it = p.end();
+    try {
+      ++it;
+      assert(false);
+    } catch (int) {}
+    --it;
+    --it;
+    try {
+      --it;
+      assert(false);
+    } catch (int) {}
+  }
+}