GH-117586: Speed up pathlib.Path.glob() by working with strings (#1…

…17589)

Move pathlib globbing implementation into a new private class: `glob._Globber`. This class implements fast string-based globbing. It's called by `pathlib.Path.glob()`, which then converts strings back to path objects.

In the private pathlib ABCs, add a `pathlib._abc.Globber` subclass that works with `PathBase` objects rather than strings, and calls user-defined path methods like `PathBase.stat()` rather than `os.stat()`.

This sets the stage for two more improvements:

- GH-115060: Query non-wildcard segments with `lstat()`
- GH-116380: Unify `pathlib` and `glob` implementations of globbing.

No change to the implementations of `glob.glob()` and `glob.iglob()`.

Lib/glob.py

@@ -4,7 +4,9 @@
 import os
 import re
 import fnmatch
+import functools
 import itertools
+import operator
 import stat
 import sys
 
@@ -256,7 +258,9 @@ def escape(pathname):
     return drive + pathname
 
 
+_special_parts = ('', '.', '..')
 _dir_open_flags = os.O_RDONLY | getattr(os, 'O_DIRECTORY', 0)
+_no_recurse_symlinks = object()
 
 
 def translate(pat, *, recursive=False, include_hidden=False, seps=None):
@@ -312,3 +316,185 @@ def translate(pat, *, recursive=False, include_hidden=False, seps=None):
                 results.append(any_sep)
     res = ''.join(results)
     return fr'(?s:{res})\Z'
+
+
+@functools.lru_cache(maxsize=512)
+def _compile_pattern(pat, sep, case_sensitive, recursive=True):
+    """Compile given glob pattern to a re.Pattern object (observing case
+    sensitivity)."""
+    flags = re.NOFLAG if case_sensitive else re.IGNORECASE
+    regex = translate(pat, recursive=recursive, include_hidden=True, seps=sep)
+    return re.compile(regex, flags=flags).match
+
+
+class _Globber:
+    """Class providing shell-style pattern matching and globbing.
+    """
+
+    def __init__(self,  sep, case_sensitive, recursive=False):
+        self.sep = sep
+        self.case_sensitive = case_sensitive
+        self.recursive = recursive
+
+    # Low-level methods
+
+    lstat = staticmethod(os.lstat)
+    scandir = staticmethod(os.scandir)
+    parse_entry = operator.attrgetter('path')
+    concat_path = operator.add
+
+    if os.name == 'nt':
+        @staticmethod
+        def add_slash(pathname):
+            tail = os.path.splitroot(pathname)[2]
+            if not tail or tail[-1] in '\\/':
+                return pathname
+            return f'{pathname}\\'
+    else:
+        @staticmethod
+        def add_slash(pathname):
+            if not pathname or pathname[-1] == '/':
+                return pathname
+            return f'{pathname}/'
+
+    # High-level methods
+
+    def compile(self, pat):
+        return _compile_pattern(pat, self.sep, self.case_sensitive, self.recursive)
+
+    def selector(self, parts):
+        """Returns a function that selects from a given path, walking and
+        filtering according to the glob-style pattern parts in *parts*.
+        """
+        if not parts:
+            return self.select_exists
+        part = parts.pop()
+        if self.recursive and part == '**':
+            selector = self.recursive_selector
+        elif part in _special_parts:
+            selector = self.special_selector
+        else:
+            selector = self.wildcard_selector
+        return selector(part, parts)
+
+    def special_selector(self, part, parts):
+        """Returns a function that selects special children of the given path.
+        """
+        select_next = self.selector(parts)
+
+        def select_special(path, exists=False):
+            path = self.concat_path(self.add_slash(path), part)
+            return select_next(path, exists)
+        return select_special
+
+    def wildcard_selector(self, part, parts):
+        """Returns a function that selects direct children of a given path,
+        filtering by pattern.
+        """
+
+        match = None if part == '*' else self.compile(part)
+        dir_only = bool(parts)
+        if dir_only:
+            select_next = self.selector(parts)
+
+        def select_wildcard(path, exists=False):
+            try:
+                # We must close the scandir() object before proceeding to
+                # avoid exhausting file descriptors when globbing deep trees.
+                with self.scandir(path) as scandir_it:
+                    entries = list(scandir_it)
+            except OSError:
+                pass
+            else:
+                for entry in entries:
+                    if match is None or match(entry.name):
+                        if dir_only:
+                            try:
+                                if not entry.is_dir():
+                                    continue
+                            except OSError:
+                                continue
+                        entry_path = self.parse_entry(entry)
+                        if dir_only:
+                            yield from select_next(entry_path, exists=True)
+                        else:
+                            yield entry_path
+        return select_wildcard
+
+    def recursive_selector(self, part, parts):
+        """Returns a function that selects a given path and all its children,
+        recursively, filtering by pattern.
+        """
+        # Optimization: consume following '**' parts, which have no effect.
+        while parts and parts[-1] == '**':
+            parts.pop()
+
+        # Optimization: consume and join any following non-special parts here,
+        # rather than leaving them for the next selector. They're used to
+        # build a regular expression, which we use to filter the results of
+        # the recursive walk. As a result, non-special pattern segments
+        # following a '**' wildcard don't require additional filesystem access
+        # to expand.
+        follow_symlinks = self.recursive is not _no_recurse_symlinks
+        if follow_symlinks:
+            while parts and parts[-1] not in _special_parts:
+                part += self.sep + parts.pop()
+
+        match = None if part == '**' else self.compile(part)
+        dir_only = bool(parts)
+        select_next = self.selector(parts)
+
+        def select_recursive(path, exists=False):
+            path = self.add_slash(path)
+            match_pos = len(str(path))
+            if match is None or match(str(path), match_pos):
+                yield from select_next(path, exists)
+            stack = [path]
+            while stack:
+                yield from select_recursive_step(stack, match_pos)
+
+        def select_recursive_step(stack, match_pos):
+            path = stack.pop()
+            try:
+                # We must close the scandir() object before proceeding to
+                # avoid exhausting file descriptors when globbing deep trees.
+                with self.scandir(path) as scandir_it:
+                    entries = list(scandir_it)
+            except OSError:
+                pass
+            else:
+                for entry in entries:
+                    is_dir = False
+                    try:
+                        if entry.is_dir(follow_symlinks=follow_symlinks):
+                            is_dir = True
+                    except OSError:
+                        pass
+
+                    if is_dir or not dir_only:
+                        entry_path = self.parse_entry(entry)
+                        if match is None or match(str(entry_path), match_pos):
+                            if dir_only:
+                                yield from select_next(entry_path, exists=True)
+                            else:
+                                # Optimization: directly yield the path if this is
+                                # last pattern part.
+                                yield entry_path
+                        if is_dir:
+                            stack.append(entry_path)
+
+        return select_recursive
+
+    def select_exists(self, path, exists=False):
+        """Yields the given path, if it exists.
+        """
+        if exists:
+            # Optimization: this path is already known to exist, e.g. because
+            # it was returned from os.scandir(), so we skip calling lstat().
+            yield path
+        else:
+            try:
+                self.lstat(path)
+                yield path
+            except OSError:
+                pass

Lib/pathlib/__init__.py

@@ -5,8 +5,10 @@
 operating systems.
 """
 
+import glob
 import io
 import ntpath
+import operator
 import os
 import posixpath
 import sys
@@ -111,6 +113,7 @@ class PurePath(_abc.PurePathBase):
         '_hash',
     )
     parser = os.path
+    _globber = glob._Globber
 
     def __new__(cls, *args, **kwargs):
         """Construct a PurePath from one or several strings and or existing
@@ -253,14 +256,17 @@ def _format_parsed_parts(cls, drv, root, tail):
         return cls.parser.sep.join(tail)
 
     def _from_parsed_parts(self, drv, root, tail):
-        path_str = self._format_parsed_parts(drv, root, tail)
-        path = self.with_segments(path_str)
-        path._str = path_str or '.'
+        path = self._from_parsed_string(self._format_parsed_parts(drv, root, tail))
         path._drv = drv
         path._root = root
         path._tail_cached = tail
         return path
 
+    def _from_parsed_string(self, path_str):
+        path = self.with_segments(path_str)
+        path._str = path_str or '.'
+        return path
+
     @classmethod
     def _parse_path(cls, path):
         if not path:
@@ -453,21 +459,6 @@ def as_uri(self):
         from urllib.parse import quote_from_bytes
         return prefix + quote_from_bytes(os.fsencode(path))
 
-    @property
-    def _pattern_stack(self):
-        """Stack of path components, to be used with patterns in glob()."""
-        parts = self._tail.copy()
-        pattern = self._raw_path
-        if self.anchor:
-            raise NotImplementedError("Non-relative patterns are unsupported")
-        elif not parts:
-            raise ValueError("Unacceptable pattern: {!r}".format(pattern))
-        elif pattern[-1] in (self.parser.sep, self.parser.altsep):
-            # GH-65238: pathlib doesn't preserve trailing slash. Add it back.
-            parts.append('')
-        parts.reverse()
-        return parts
-
     @property
     def _pattern_str(self):
         """The path expressed as a string, for use in pattern-matching."""
@@ -576,6 +567,17 @@ def write_text(self, data, encoding=None, errors=None, newline=None):
         encoding = io.text_encoding(encoding)
         return _abc.PathBase.write_text(self, data, encoding, errors, newline)
 
+    _remove_leading_dot = operator.itemgetter(slice(2, None))
+    _remove_trailing_slash = operator.itemgetter(slice(-1))
+
+    def _filter_trailing_slash(self, paths):
+        sep = self.parser.sep
+        anchor_len = len(self.anchor)
+        for path_str in paths:
+            if len(path_str) > anchor_len and path_str[-1] == sep:
+                path_str = path_str[:-1]
+            yield path_str
+
     def iterdir(self):
         """Yield path objects of the directory contents.
 
@@ -587,13 +589,9 @@ def iterdir(self):
     def _scandir(self):
         return os.scandir(self)
 
-    def _direntry_str(self, entry):
-        # Transform an entry yielded from _scandir() into a path string.
-        return entry.name if str(self) == '.' else entry.path
-
     def _make_child_direntry(self, entry):
         # Transform an entry yielded from _scandir() into a path object.
-        path_str = self._direntry_str(entry)
+        path_str = entry.name if str(self) == '.' else entry.path
         path = self.with_segments(path_str)
         path._str = path_str
         path._drv = self.drive
@@ -626,8 +624,30 @@ def glob(self, pattern, *, case_sensitive=None, recurse_symlinks=False):
         sys.audit("pathlib.Path.glob", self, pattern)
         if not isinstance(pattern, PurePath):
             pattern = self.with_segments(pattern)
-        return _abc.PathBase.glob(
-            self, pattern, case_sensitive=case_sensitive, recurse_symlinks=recurse_symlinks)
+        if pattern.anchor:
+            raise NotImplementedError("Non-relative patterns are unsupported")
+        parts = pattern._tail.copy()
+        if not parts:
+            raise ValueError("Unacceptable pattern: {!r}".format(pattern))
+        raw = pattern._raw_path
+        if raw[-1] in (self.parser.sep, self.parser.altsep):
+            # GH-65238: pathlib doesn't preserve trailing slash. Add it back.
+            parts.append('')
+        if not self.is_dir():
+            return iter([])
+        select = self._glob_selector(parts[::-1], case_sensitive, recurse_symlinks)
+        root = str(self)
+        paths = select(root, exists=True)
+
+        # Normalize results
+        if root == '.':
+            paths = map(self._remove_leading_dot, paths)
+        if parts[-1] == '':
+            paths = map(self._remove_trailing_slash, paths)
+        elif parts[-1] == '**':
+            paths = self._filter_trailing_slash(paths)
+        paths = map(self._from_parsed_string, paths)
+        return paths
 
     def rglob(self, pattern, *, case_sensitive=None, recurse_symlinks=False):
         """Recursively yield all existing files (of any kind, including
@@ -638,8 +658,7 @@ def rglob(self, pattern, *, case_sensitive=None, recurse_symlinks=False):
         if not isinstance(pattern, PurePath):
             pattern = self.with_segments(pattern)
         pattern = '**' / pattern
-        return _abc.PathBase.glob(
-            self, pattern, case_sensitive=case_sensitive, recurse_symlinks=recurse_symlinks)
+        return self.glob(pattern, case_sensitive=case_sensitive, recurse_symlinks=recurse_symlinks)
 
     def walk(self, top_down=True, on_error=None, follow_symlinks=False):
         """Walk the directory tree from this directory, similar to os.walk()."""
@@ -669,9 +688,7 @@ def absolute(self):
             # of joining, and we exploit the fact that getcwd() returns a
             # fully-normalized string by storing it in _str. This is used to
             # implement Path.cwd().
-            result = self.with_segments(cwd)
-            result._str = cwd
-            return result
+            return self._from_parsed_string(cwd)
         drive, root, rel = os.path.splitroot(cwd)
         if not rel:
             return self._from_parsed_parts(drive, root, self._tail)