GH-115060: Speed up `pathlib.Path.glob()` by not scanning literal parts (#117732)

Don't bother calling `os.scandir()` to scan for literal pattern segments,
like `foo` in `foo/*.py`. Instead, append the segment(s) as-is and call
through to the next selector with `exists=False`, which signals that the
path might not exist. Subsequent selectors will call `os.scandir()` or
`os.lstat()` to filter out missing paths as needed.
This commit is contained in:
Barney Gale 2024-04-12 22:19:21 +01:00 committed by GitHub
parent 069de14cb9
commit 0eb52f5f26
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 42 additions and 11 deletions

View File

@ -331,9 +331,10 @@ class _Globber:
"""Class providing shell-style pattern matching and globbing.
"""
def __init__(self, sep, case_sensitive, recursive=False):
def __init__(self, sep, case_sensitive, case_pedantic=False, recursive=False):
self.sep = sep
self.case_sensitive = case_sensitive
self.case_pedantic = case_pedantic
self.recursive = recursive
# Low-level methods
@ -373,6 +374,8 @@ class _Globber:
selector = self.recursive_selector
elif part in _special_parts:
selector = self.special_selector
elif not self.case_pedantic and magic_check.search(part) is None:
selector = self.literal_selector
else:
selector = self.wildcard_selector
return selector(part, parts)
@ -387,6 +390,23 @@ class _Globber:
return select_next(path, exists)
return select_special
def literal_selector(self, part, parts):
"""Returns a function that selects a literal descendant of a path.
"""
# Optimization: consume and join any subsequent literal parts here,
# rather than leaving them for the next selector. This reduces the
# number of string concatenation operations and calls to add_slash().
while parts and magic_check.search(parts[-1]) is None:
part += self.sep + parts.pop()
select_next = self.selector(parts)
def select_literal(path, exists=False):
path = self.concat_path(self.add_slash(path), part)
return select_next(path, exists=False)
return select_literal
def wildcard_selector(self, part, parts):
"""Returns a function that selects direct children of a given path,
filtering by pattern.

View File

@ -686,8 +686,14 @@ class PathBase(PurePathBase):
def _glob_selector(self, parts, case_sensitive, recurse_symlinks):
if case_sensitive is None:
case_sensitive = _is_case_sensitive(self.parser)
case_pedantic = False
else:
# The user has expressed a case sensitivity choice, but we don't
# know the case sensitivity of the underlying filesystem, so we
# must use scandir() for everything, including non-wildcard parts.
case_pedantic = True
recursive = True if recurse_symlinks else glob._no_recurse_symlinks
globber = self._globber(self.parser.sep, case_sensitive, recursive)
globber = self._globber(self.parser.sep, case_sensitive, case_pedantic, recursive)
return globber.selector(parts)
def glob(self, pattern, *, case_sensitive=None, recurse_symlinks=True):

View File

@ -1429,10 +1429,10 @@ class DummyPath(PathBase):
return "{}({!r})".format(self.__class__.__name__, self.as_posix())
def stat(self, *, follow_symlinks=True):
if follow_symlinks:
path = str(self.resolve())
if follow_symlinks or self.name in ('', '.', '..'):
path = str(self.resolve(strict=True))
else:
path = str(self.parent.resolve() / self.name)
path = str(self.parent.resolve(strict=True) / self.name)
if path in self._files:
st_mode = stat.S_IFREG
elif path in self._directories:
@ -1741,8 +1741,9 @@ class DummyPathTest(DummyPurePathTest):
def test_glob_posix(self):
P = self.cls
p = P(self.base)
q = p / "FILEa"
given = set(p.glob("FILEa"))
expect = set()
expect = {q} if q.exists() else set()
self.assertEqual(given, expect)
self.assertEqual(set(p.glob("FILEa*")), set())
@ -1753,8 +1754,6 @@ class DummyPathTest(DummyPurePathTest):
self.assertEqual(set(p.glob("FILEa")), { P(self.base, "fileA") })
self.assertEqual(set(p.glob("*a\\")), { P(self.base, "dirA/") })
self.assertEqual(set(p.glob("F*a")), { P(self.base, "fileA") })
self.assertEqual(set(map(str, p.glob("FILEa"))), {f"{p}\\fileA"})
self.assertEqual(set(map(str, p.glob("F*a"))), {f"{p}\\fileA"})
def test_glob_empty_pattern(self):
P = self.cls
@ -1857,8 +1856,9 @@ class DummyPathTest(DummyPurePathTest):
def test_rglob_posix(self):
P = self.cls
p = P(self.base, "dirC")
q = p / "dirD" / "FILEd"
given = set(p.rglob("FILEd"))
expect = set()
expect = {q} if q.exists() else set()
self.assertEqual(given, expect)
self.assertEqual(set(p.rglob("FILEd*")), set())
@ -1868,7 +1868,6 @@ class DummyPathTest(DummyPurePathTest):
p = P(self.base, "dirC")
self.assertEqual(set(p.rglob("FILEd")), { P(self.base, "dirC/dirD/fileD") })
self.assertEqual(set(p.rglob("*\\")), { P(self.base, "dirC/dirD/") })
self.assertEqual(set(map(str, p.rglob("FILEd"))), {f"{p}\\dirD\\fileD"})
@needs_symlinks
def test_rglob_recurse_symlinks_common(self):
@ -1931,7 +1930,11 @@ class DummyPathTest(DummyPurePathTest):
self.assertEqual(set(p.glob("dirA/../file*")), { P(self.base, "dirA/../fileA") })
self.assertEqual(set(p.glob("dirA/../file*/..")), set())
self.assertEqual(set(p.glob("../xyzzy")), set())
self.assertEqual(set(p.glob("xyzzy/..")), set())
if self.cls.parser is posixpath:
self.assertEqual(set(p.glob("xyzzy/..")), set())
else:
# ".." segments are normalized first on Windows, so this path is stat()able.
self.assertEqual(set(p.glob("xyzzy/..")), { P(self.base, "xyzzy", "..") })
self.assertEqual(set(p.glob("/".join([".."] * 50))), { P(self.base, *[".."] * 50)})
@needs_symlinks

View File

@ -0,0 +1,2 @@
Speed up :meth:`pathlib.Path.glob` by not scanning directories for
non-wildcard pattern segments.