pythonGH-101362 - Optimise pathlib by deferring path normalisation

`PurePath` now normalises and splits paths only when necessary, e.g. when
`.name` or `.parent` is accessed. The result is cached. This speeds up path
object construction by around 4x.

`PurePath.__fspath__()` now returns an unnormalised path, which should be
transparent to filesystem APIs (else pathlib's normalisation is broken!).
This extends the earlier performance improvement to most impure `Path`
methods, and also speeds up pickling, `p.joinpath('bar')` and `p / 'bar'`.

This also fixes pythonGH-76846 and pythonGH-85281 by unifying path constructors and
adding an `__init__()` method.

Lib/pathlib.py

@@ -216,8 +216,8 @@ class _PathParents(Sequence):
     def __init__(self, path):
         # We don't store the instance to avoid reference cycles
         self._pathcls = type(path)
-        self._drv = path._drv
-        self._root = path._root
+        self._drv = path.drive
+        self._root = path.root
         self._parts = path._parts
 
     def __len__(self):
@@ -251,33 +251,45 @@ class PurePath(object):
     directly, regardless of your system.
     """
     __slots__ = (
-        '_drv', '_root', '_parts',
+        '_fspath', '_drv', '_root', '_parts_cached',
         '_str', '_hash', '_parts_tuple', '_parts_normcase_cached',
     )
     _flavour = os.path
 
-    def __new__(cls, *args):
+    def __new__(cls, *args, **kwargs):
         """Construct a PurePath from one or several strings and or existing
         PurePath objects.  The strings and path objects are combined so as
         to yield a canonicalized path, which is incorporated into the
         new PurePath object.
         """
         if cls is PurePath:
             cls = PureWindowsPath if os.name == 'nt' else PurePosixPath
-        return cls._from_parts(args)
+        return super().__new__(cls)
 
     def __reduce__(self):
-        # Using the parts tuple helps share interned path parts
-        # when pickling related paths.
-        return (self.__class__, tuple(self._parts))
+        return (self.__class__, (self._fspath,))
+
+    def __init__(self, *args):
+        if not args:
+            path = ''
+        elif len(args) == 1:
+            path = os.fspath(args[0])
+        else:
+            path = self._flavour.join(*args)
+
+        if not isinstance(path, str):
+            raise TypeError(
+                "argument should be a str object or an os.PathLike "
+                "object returning str, not %r"
+                % type(path))
+        self._fspath = path
 
     @classmethod
-    def _parse_parts(cls, parts):
-        if not parts:
+    def _parse_path(cls, path):
+        if not path:
             return '', '', []
         sep = cls._flavour.sep
         altsep = cls._flavour.altsep
-        path = cls._flavour.join(*parts)
         if altsep:
             path = path.replace(altsep, sep)
         drv, root, rel = cls._flavour.splitroot(path)
@@ -288,43 +300,18 @@ def _parse_parts(cls, parts):
         parsed = [sys.intern(x) for x in unfiltered_parsed if x and x != '.']
         return drv, root, parsed
 
-    @classmethod
-    def _parse_args(cls, args):
-        # This is useful when you don't want to create an instance, just
-        # canonicalize some constructor arguments.
-        parts = []
-        for a in args:
-            if isinstance(a, PurePath):
-                parts += a._parts
-            else:
-                a = os.fspath(a)
-                if isinstance(a, str):
-                    # Force-cast str subclasses to str (issue #21127)
-                    parts.append(str(a))
-                else:
-                    raise TypeError(
-                        "argument should be a str object or an os.PathLike "
-                        "object returning str, not %r"
-                        % type(a))
-        return cls._parse_parts(parts)
-
-    @classmethod
-    def _from_parts(cls, args):
-        # We need to call _parse_args on the instance, so as to get the
-        # right flavour.
-        self = object.__new__(cls)
-        drv, root, parts = self._parse_args(args)
+    def _load_parts(self):
+        drv, root, parts = self._parse_path(self._fspath)
         self._drv = drv
         self._root = root
-        self._parts = parts
-        return self
+        self._parts_cached = parts
 
     @classmethod
     def _from_parsed_parts(cls, drv, root, parts):
-        self = object.__new__(cls)
+        self = cls(cls._format_parsed_parts(drv, root, parts))
         self._drv = drv
         self._root = root
-        self._parts = parts
+        self._parts_cached = parts
         return self
 
     @classmethod
@@ -340,12 +327,12 @@ def __str__(self):
         try:
             return self._str
         except AttributeError:
-            self._str = self._format_parsed_parts(self._drv, self._root,
+            self._str = self._format_parsed_parts(self.drive, self.root,
                                                   self._parts) or '.'
             return self._str
 
     def __fspath__(self):
-        return str(self)
+        return self._fspath or '.'
 
     def as_posix(self):
         """Return the string representation of the path with forward (/)
@@ -356,7 +343,7 @@ def as_posix(self):
     def __bytes__(self):
         """Return the bytes representation of the path.  This is only
         recommended to use under Unix."""
-        return os.fsencode(self)
+        return os.fsencode(str(self))
 
     def __repr__(self):
         return "{}({!r})".format(self.__class__.__name__, self.as_posix())
@@ -366,7 +353,7 @@ def as_uri(self):
         if not self.is_absolute():
             raise ValueError("relative path can't be expressed as a file URI")
 
-        drive = self._drv
+        drive = self.drive
         if len(drive) == 2 and drive[1] == ':':
             # It's a path on a local drive => 'file:///c:/a/b'
             prefix = 'file:///' + drive
@@ -422,23 +409,43 @@ def __ge__(self, other):
             return NotImplemented
         return self._parts_normcase >= other._parts_normcase
 
-    drive = property(attrgetter('_drv'),
-                     doc="""The drive prefix (letter or UNC path), if any.""")
+    @property
+    def drive(self):
+        """The drive prefix (letter or UNC path), if any."""
+        try:
+            return self._drv
+        except AttributeError:
+            self._load_parts()
+            return self._drv
 
-    root = property(attrgetter('_root'),
-                    doc="""The root of the path, if any.""")
+    @property
+    def root(self):
+        """The root of the path, if any."""
+        try:
+            return self._root
+        except AttributeError:
+            self._load_parts()
+            return self._root
+
+    @property
+    def _parts(self):
+        try:
+            return self._parts_cached
+        except AttributeError:
+            self._load_parts()
+            return self._parts_cached
 
     @property
     def anchor(self):
         """The concatenation of the drive and root, or ''."""
-        anchor = self._drv + self._root
+        anchor = self.drive + self.root
         return anchor
 
     @property
     def name(self):
         """The final path component, if any."""
         parts = self._parts
-        if len(parts) == (1 if (self._drv or self._root) else 0):
+        if len(parts) == (1 if (self.drive or self.root) else 0):
             return ''
         return parts[-1]
 
@@ -487,7 +494,7 @@ def with_name(self, name):
         drv, root, tail = f.splitroot(name)
         if drv or root or not tail or f.sep in tail or (f.altsep and f.altsep in tail):
             raise ValueError("Invalid name %r" % (name))
-        return self._from_parsed_parts(self._drv, self._root,
+        return self._from_parsed_parts(self.drive, self.root,
                                        self._parts[:-1] + [name])
 
     def with_stem(self, stem):
@@ -512,7 +519,7 @@ def with_suffix(self, suffix):
             name = name + suffix
         else:
             name = name[:-len(old_suffix)] + suffix
-        return self._from_parsed_parts(self._drv, self._root,
+        return self._from_parsed_parts(self.drive, self.root,
                                        self._parts[:-1] + [name])
 
     def relative_to(self, other, /, *_deprecated, walk_up=False):
@@ -571,22 +578,7 @@ def joinpath(self, *args):
         paths) or a totally different path (if one of the arguments is
         anchored).
         """
-        drv1, root1, parts1 = self._drv, self._root, self._parts
-        drv2, root2, parts2 = self._parse_args(args)
-        if root2:
-            if not drv2 and drv1:
-                return self._from_parsed_parts(drv1, root2, [drv1 + root2] + parts2[1:])
-            else:
-                return self._from_parsed_parts(drv2, root2, parts2)
-        elif drv2:
-            if drv2 == drv1 or self._flavour.normcase(drv2) == self._flavour.normcase(drv1):
-                # Same drive => second path is relative to the first.
-                return self._from_parsed_parts(drv1, root1, parts1 + parts2[1:])
-            else:
-                return self._from_parsed_parts(drv2, root2, parts2)
-        else:
-            # Second path is non-anchored (common case).
-            return self._from_parsed_parts(drv1, root1, parts1 + parts2)
+        return type(self)(self, *args)
 
     def __truediv__(self, key):
         try:
@@ -596,15 +588,15 @@ def __truediv__(self, key):
 
     def __rtruediv__(self, key):
         try:
-            return self._from_parts([key] + self._parts)
+            return type(self)(key, self)
         except TypeError:
             return NotImplemented
 
     @property
     def parent(self):
         """The logical parent of the path."""
-        drv = self._drv
-        root = self._root
+        drv = self.drive
+        root = self.root
         parts = self._parts
         if len(parts) == 1 and (drv or root):
             return self
@@ -620,7 +612,7 @@ def is_absolute(self):
         a drive)."""
         # ntpath.isabs() is defective - see GH-44626 .
         if self._flavour is ntpath:
-            return bool(self._drv and self._root)
+            return bool(self.drive and self.root)
         return self._flavour.isabs(self)
 
     def is_reserved(self):
@@ -644,12 +636,12 @@ def match(self, path_pattern):
         Return True if this path matches the given pattern.
         """
         path_pattern = self._flavour.normcase(path_pattern)
-        drv, root, pat_parts = self._parse_parts((path_pattern,))
+        drv, root, pat_parts = self._parse_path(path_pattern)
         if not pat_parts:
             raise ValueError("empty pattern")
-        elif drv and drv != self._flavour.normcase(self._drv):
+        elif drv and drv != self._flavour.normcase(self.drive):
             return False
-        elif root and root != self._root:
+        elif root and root != self.root:
             return False
         parts = self._parts_normcase
         if drv or root:
@@ -702,24 +694,26 @@ class Path(PurePath):
     """
     __slots__ = ()
 
-    def __new__(cls, *args, **kwargs):
+    def __init__(self, *args, **kwargs):
         if kwargs:
             msg = ("support for supplying keyword arguments to pathlib.PurePath "
                    "is deprecated and scheduled for removal in Python {remove}")
             warnings._deprecated("pathlib.PurePath(**kwargs)", msg, remove=(3, 14))
+        super().__init__(*args)
+
+    def __new__(cls, *args, **kwargs):
         if cls is Path:
             cls = WindowsPath if os.name == 'nt' else PosixPath
-        self = cls._from_parts(args)
-        if self._flavour is not os.path:
+        elif cls._flavour is not os.path:
             raise NotImplementedError("cannot instantiate %r on your system"
                                       % (cls.__name__,))
-        return self
+        return super().__new__(cls)
 
     def _make_child_relpath(self, part):
         # This is an optimization used for dir walking.  `part` must be
         # a single part relative to this path.
         parts = self._parts + [part]
-        return self._from_parsed_parts(self._drv, self._root, parts)
+        return self._from_parsed_parts(self.drive, self.root, parts)
 
     def __enter__(self):
         # In previous versions of pathlib, __exit__() marked this path as
@@ -789,7 +783,7 @@ def glob(self, pattern):
         sys.audit("pathlib.Path.glob", self, pattern)
         if not pattern:
             raise ValueError("Unacceptable pattern: {!r}".format(pattern))
-        drv, root, pattern_parts = self._parse_parts((pattern,))
+        drv, root, pattern_parts = self._parse_path(pattern)
         if drv or root:
             raise NotImplementedError("Non-relative patterns are unsupported")
         if pattern[-1] in (self._flavour.sep, self._flavour.altsep):
@@ -804,7 +798,7 @@ def rglob(self, pattern):
         this subtree.
         """
         sys.audit("pathlib.Path.rglob", self, pattern)
-        drv, root, pattern_parts = self._parse_parts((pattern,))
+        drv, root, pattern_parts = self._parse_path(pattern)
         if drv or root:
             raise NotImplementedError("Non-relative patterns are unsupported")
         if pattern and pattern[-1] in (self._flavour.sep, self._flavour.altsep):
@@ -821,7 +815,7 @@ def absolute(self):
         """
         if self.is_absolute():
             return self
-        return self._from_parts([os.getcwd()] + self._parts)
+        return type(self)(os.getcwd(), *self._parts)
 
     def resolve(self, strict=False):
         """
@@ -839,7 +833,7 @@ def check_eloop(e):
         except OSError as e:
             check_eloop(e)
             raise
-        p = self._from_parts((s,))
+        p = type(self)(s)
 
         # In non-strict mode, realpath() doesn't raise on symlink loops.
         # Ensure we get an exception by calling stat()
@@ -929,7 +923,7 @@ def readlink(self):
         """
         if not hasattr(os, "readlink"):
             raise NotImplementedError("os.readlink() not available on this system")
-        return self._from_parts((os.readlink(self),))
+        return type(self)(os.readlink(self))
 
     def touch(self, mode=0o666, exist_ok=True):
         """
@@ -1198,12 +1192,12 @@ def expanduser(self):
         """ Return a new path with expanded ~ and ~user constructs
         (as returned by os.path.expanduser)
         """
-        if (not (self._drv or self._root) and
+        if (not (self.drive or self.root) and
             self._parts and self._parts[0][:1] == '~'):
             homedir = self._flavour.expanduser(self._parts[0])
             if homedir[:1] == "~":
                 raise RuntimeError("Could not determine home directory.")
-            return self._from_parts([homedir] + self._parts[1:])
+            return type(self)(homedir, *self._parts[1:])
 
         return self
 

Lib/test/test_pathlib.py

@@ -26,7 +26,9 @@
 class _BaseFlavourTest(object):
 
     def _check_parse_parts(self, arg, expected):
-        f = self.cls._parse_parts
+        def f(parts):
+            path = self.flavour.join(*parts) if parts else ''
+            return self.cls._parse_path(path)
         sep = self.flavour.sep
         altsep = self.flavour.altsep
         actual = f([x.replace('/', sep) for x in arg])