Path.rglob fails with *.* on Python 3.11.0a7

[domdfcoding]

Bug report

On Python 3.11.0a7, calling Path.rglob("*.*") returns an iterator over no files, whereas on previous versions (including alpha 6) it would correctly return an iterator over all files in the directory and all of its children.

The following code illustrates this problem:

import pathlib, tempfile, pprint

with tempfile.TemporaryDirectory() as tmpdir:
	tmp_path = pathlib.Path(tmpdir)
	(tmp_path / "a.txt").touch()
	(tmp_path / "b.txt").touch()
	(tmp_path / "c").mkdir()
	(tmp_path / "c" / "d.txt").touch()
	pprint.pprint(list(tmp_path.rglob("*.*")))
	pprint.pprint(list(tmp_path.rglob("*.txt")))

Save this as pathlib_rglob.py.

Output:

$ python3.11 pathlib_rglob.py 
[]
[PosixPath('/tmp/tmpngjo2cyn/a.txt'),
 PosixPath('/tmp/tmpngjo2cyn/b.txt'),
 PosixPath('/tmp/tmpngjo2cyn/c/d.txt')]
$ python3.10 pathlib_rglob.py 
[PosixPath('/tmp/tmpodfbqyhx/a.txt'),
 PosixPath('/tmp/tmpodfbqyhx/b.txt'),
 PosixPath('/tmp/tmpodfbqyhx/c/d.txt')]
[PosixPath('/tmp/tmpodfbqyhx/a.txt'),
 PosixPath('/tmp/tmpodfbqyhx/b.txt'),
 PosixPath('/tmp/tmpodfbqyhx/c/d.txt')]

$ # This is Python 3.11.0a6 built from source
$ ~/python311a6/bin/python3.11 pathlib_rglob.py 
[PosixPath('/tmp/tmpobellcrw/a.txt'),
 PosixPath('/tmp/tmpobellcrw/b.txt'),
 PosixPath('/tmp/tmpobellcrw/c/d.txt')]
[PosixPath('/tmp/tmpobellcrw/a.txt'),
 PosixPath('/tmp/tmpobellcrw/b.txt'),
 PosixPath('/tmp/tmpobellcrw/c/d.txt')]

---

After much debugging I have tracked the issue to #32029
With that change, fnmatch.translate("*.*") gives (?s:(?>.*?\.).*)\Z.
With the change reverted (or on 3.11.0a6) its (?s:(?=(?P<g0>.*?\.))(?P=g0).*)\Z.
I don't know what the difference between those amounts to, but the former doesn't work while the latter does:

>>> import re
>>> re.fullmatch('(?s:(?>.*?\\.).*)\\Z', "a.txt")
>>> re.fullmatch('(?s:(?=(?P<g0>.*?\\.))(?P=g0).*)\\Z', "a.txt")
<re.Match object; span=(0, 5), match='a.txt'>

Your environment

• CPython versions tested on: 3.10.4, 3.11.0a6, 3.11.0a7, built from git checkouts of the relevant tag.
• Operating system and architecture: Ubuntu 20.04 amd64

[domdfcoding]

@tim-one as you made the original change in #32029 could you take a look at this?

[ghost]

It's a bug of re module.

@serhiy-storchaka , the creator of PR #31982 (Add support of atomic grouping in re).
This patch fixes the problem. I will submit a PR after thoroughly understanding the problem.

 Modules/_sre/sre_lib.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/Modules/_sre/sre_lib.h b/Modules/_sre/sre_lib.h
index 3472e65b87..c73e116143 100644
--- a/Modules/_sre/sre_lib.h
+++ b/Modules/_sre/sre_lib.h
@@ -1350,8 +1350,8 @@ SRE(match)(SRE_STATE* state, const SRE_CODE* pattern, int toplevel)
                when the end of the group, represented by a SUCCESS op
                code, is reached. */
             /* Group Pattern begins at an offset of 1 code. */
-            DO_JUMP(JUMP_ATOMIC_GROUP, jump_atomic_group,
-                    &pattern[1]);
+            DO_JUMP0(JUMP_ATOMIC_GROUP, jump_atomic_group,
+                     &pattern[1]);
 
             /* Test Exit Condition */
             RETURN_ON_ERROR(ret);

[serhiy-storchaka]

Ah, I see.

>>> re.match(r'(?s:(?>.*?\.).*)\Z', 'a.txt')
<re.Match object; span=(0, 5), match='a.txt'>
>>> re.fullmatch(r'(?s:(?>.*?\.).*)\Z', 'a.txt')

I thought that issues with fullmatch() and atomic grouping were solved, but it is not so.

Use the following tests in your PR:

diff --git a/Lib/test/test_re.py b/Lib/test/test_re.py
index 959582e2f1..a5aa1b358d 100644
--- a/Lib/test/test_re.py
+++ b/Lib/test/test_re.py
@@ -2173,6 +2173,10 @@ def test_fullmatch_possessive_quantifiers(self):
         self.assertIsNone(re.fullmatch(r'a*+', 'ab'))
         self.assertIsNone(re.fullmatch(r'a?+', 'ab'))
         self.assertIsNone(re.fullmatch(r'a{1,3}+', 'ab'))
+        self.assertTrue(re.fullmatch(r'a++b', 'ab'))
+        self.assertTrue(re.fullmatch(r'a*+b', 'ab'))
+        self.assertTrue(re.fullmatch(r'a?+b', 'ab'))
+        self.assertTrue(re.fullmatch(r'a{1,3}+b', 'ab'))
 
         self.assertTrue(re.fullmatch(r'(?:ab)++', 'ab'))
         self.assertTrue(re.fullmatch(r'(?:ab)*+', 'ab'))
@@ -2182,6 +2186,10 @@ def test_fullmatch_possessive_quantifiers(self):
         self.assertIsNone(re.fullmatch(r'(?:ab)*+', 'abc'))
         self.assertIsNone(re.fullmatch(r'(?:ab)?+', 'abc'))
         self.assertIsNone(re.fullmatch(r'(?:ab){1,3}+', 'abc'))
+        self.assertTrue(re.fullmatch(r'(?:ab)++c', 'abc'))
+        self.assertTrue(re.fullmatch(r'(?:ab)*+c', 'abc'))
+        self.assertTrue(re.fullmatch(r'(?:ab)?+c', 'abc'))
+        self.assertTrue(re.fullmatch(r'(?:ab){1,3}+c', 'abc'))
 
     def test_findall_possessive_quantifiers(self):
         self.assertEqual(re.findall(r'a++', 'aab'), ['aa'])
@@ -2217,6 +2225,10 @@ def test_fullmatch_atomic_grouping(self):
         self.assertIsNone(re.fullmatch(r'(?>a*)', 'ab'))
         self.assertIsNone(re.fullmatch(r'(?>a?)', 'ab'))
         self.assertIsNone(re.fullmatch(r'(?>a{1,3})', 'ab'))
+        self.assertTrue(re.fullmatch(r'(?>a+)b', 'ab'))
+        self.assertTrue(re.fullmatch(r'(?>a*)b', 'ab'))
+        self.assertTrue(re.fullmatch(r'(?>a?)b', 'ab'))
+        self.assertTrue(re.fullmatch(r'(?>a{1,3})b', 'ab'))
 
         self.assertTrue(re.fullmatch(r'(?>(?:ab)+)', 'ab'))
         self.assertTrue(re.fullmatch(r'(?>(?:ab)*)', 'ab'))
@@ -2226,6 +2238,10 @@ def test_fullmatch_atomic_grouping(self):
         self.assertIsNone(re.fullmatch(r'(?>(?:ab)*)', 'abc'))
         self.assertIsNone(re.fullmatch(r'(?>(?:ab)?)', 'abc'))
         self.assertIsNone(re.fullmatch(r'(?>(?:ab){1,3})', 'abc'))
+        self.assertTrue(re.fullmatch(r'(?>(?:ab)+)c', 'abc'))
+        self.assertTrue(re.fullmatch(r'(?>(?:ab)*)c', 'abc'))
+        self.assertTrue(re.fullmatch(r'(?>(?:ab)?)c', 'abc'))
+        self.assertTrue(re.fullmatch(r'(?>(?:ab){1,3})c', 'abc'))
 
     def test_findall_atomic_grouping(self):
         self.assertEqual(re.findall(r'(?>a+)', 'aab'), ['aa'])

[serhiy-storchaka]

Make the same change in POSSESSIVE_REPEAT.

[ghost]

A draft: https://github.com/animalize/cpython/commit/8e00544280f054b663c056c6a99973749efe4b97
I'm carefully reviewing #31982 and match_all/must_advance code.

[ghost]

@serhiy-storchaka , please review #91681 .