Mailing List Archive

bpo-45703: Invalidate NamespacePath cache on importlib.invalidate ca… (GH-29384)
https://github.com/python/cpython/commit/ae1965ccb4b1fad63fab40fe8805d1b8247668d3
commit: ae1965ccb4b1fad63fab40fe8805d1b8247668d3
branch: main
author: Miro Hron?ok <miro@hroncok.cz>
committer: miss-islington <31488909+miss-islington@users.noreply.github.com>
date: 2021-11-23T07:38:02-08:00
summary:

bpo-45703: Invalidate _NamespacePath cache on importlib.invalidate_ca… (GH-29384)



Consider the following directory structure:

.
??? PATH1
??? namespace
??? sub1
??? __init__.py

And both PATH1 and PATH2 in sys path:

$ PYTHONPATH=PATH1:PATH2 python3.11
>>> import namespace
>>> import namespace.sub1
>>> namespace.__path__
_NamespacePath(['.../PATH1/namespace'])
>>> ...

While this interpreter still runs, PATH2/namespace/sub2 is created:

.
??? PATH1
? ??? namespace
? ??? sub1
? ??? __init__.py
??? PATH2
??? namespace
??? sub2
??? __init__.py

The newly created module cannot be imported:

>>> ...
>>> namespace.__path__
_NamespacePath(['.../PATH1/namespace'])
>>> import namespace.sub2
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ModuleNotFoundError: No module named 'namespace.sub2'

Calling importlib.invalidate_caches() now newly allows to import it:

>>> import importlib
>>> importlib.invalidate_caches()
>>> namespace.__path__
_NamespacePath(['.../PATH1/namespace'])
>>> import namespace.sub2
>>> namespace.__path__
_NamespacePath(['.../PATH1/namespace', '.../PATH2/namespace'])

This was not previously possible.

files:
A Misc/NEWS.d/next/Library/2021-11-03-13-41-49.bpo-45703.35AagL.rst
M Doc/library/importlib.rst
M Lib/importlib/_bootstrap_external.py
M Lib/test/test_importlib/test_namespace_pkgs.py

diff --git a/Doc/library/importlib.rst b/Doc/library/importlib.rst
index 6b71e64bcbfa7..59c8c64acdc3b 100644
--- a/Doc/library/importlib.rst
+++ b/Doc/library/importlib.rst
@@ -145,6 +145,10 @@ Functions

.. versionadded:: 3.3

+ .. versionchanged:: 3.10
+ Namespace packages created/installed in a different :data:`sys.path`
+ location after the same namespace was already imported are noticed.
+
.. function:: reload(module)

Reload a previously imported *module*. The argument must be a module object,
diff --git a/Lib/importlib/_bootstrap_external.py b/Lib/importlib/_bootstrap_external.py
index 85c5193407d80..6970e9f0a94d4 100644
--- a/Lib/importlib/_bootstrap_external.py
+++ b/Lib/importlib/_bootstrap_external.py
@@ -1231,10 +1231,15 @@ class _NamespacePath:
using path_finder. For top-level modules, the parent module's path
is sys.path."""

+ # When invalidate_caches() is called, this epoch is incremented
+ # https://bugs.python.org/issue45703
+ _epoch = 0
+
def __init__(self, name, path, path_finder):
self._name = name
self._path = path
self._last_parent_path = tuple(self._get_parent_path())
+ self._last_epoch = self._epoch
self._path_finder = path_finder

def _find_parent_path_names(self):
@@ -1254,7 +1259,7 @@ def _get_parent_path(self):
def _recalculate(self):
# If the parent's path has changed, recalculate _path
parent_path = tuple(self._get_parent_path()) # Make a copy
- if parent_path != self._last_parent_path:
+ if parent_path != self._last_parent_path or self._epoch != self._last_epoch:
spec = self._path_finder(self._name, parent_path)
# Note that no changes are made if a loader is returned, but we
# do remember the new parent path
@@ -1262,6 +1267,7 @@ def _recalculate(self):
if spec.submodule_search_locations:
self._path = spec.submodule_search_locations
self._last_parent_path = parent_path # Save the copy
+ self._last_epoch = self._epoch
return self._path

def __iter__(self):
@@ -1355,6 +1361,9 @@ def invalidate_caches():
del sys.path_importer_cache[name]
elif hasattr(finder, 'invalidate_caches'):
finder.invalidate_caches()
+ # Also invalidate the caches of _NamespacePaths
+ # https://bugs.python.org/issue45703
+ _NamespacePath._epoch += 1

@staticmethod
def _path_hooks(path):
diff --git a/Lib/test/test_importlib/test_namespace_pkgs.py b/Lib/test/test_importlib/test_namespace_pkgs.py
index f80283233f977..2ea41b7a4c5c3 100644
--- a/Lib/test/test_importlib/test_namespace_pkgs.py
+++ b/Lib/test/test_importlib/test_namespace_pkgs.py
@@ -4,6 +4,7 @@
import importlib.machinery
import os
import sys
+import tempfile
import unittest
import warnings

@@ -130,6 +131,40 @@ def test_imports(self):
self.assertEqual(foo.two.attr, 'portion2 foo two')


+class SeparatedNamespacePackagesCreatedWhileRunning(NamespacePackageTest):
+ paths = ['portion1']
+
+ def test_invalidate_caches(self):
+ with tempfile.TemporaryDirectory() as temp_dir:
+ # we manipulate sys.path before anything is imported to avoid
+ # accidental cache invalidation when changing it
+ sys.path.append(temp_dir)
+
+ import foo.one
+ self.assertEqual(foo.one.attr, 'portion1 foo one')
+
+ # the module does not exist, so it cannot be imported
+ with self.assertRaises(ImportError):
+ import foo.just_created
+
+ # util.create_modules() manipulates sys.path
+ # so we must create the modules manually instead
+ namespace_path = os.path.join(temp_dir, 'foo')
+ os.mkdir(namespace_path)
+ module_path = os.path.join(namespace_path, 'just_created.py')
+ with open(module_path, 'w', encoding='utf-8') as file:
+ file.write('attr = "just_created foo"')
+
+ # the module is not known, so it cannot be imported yet
+ with self.assertRaises(ImportError):
+ import foo.just_created
+
+ # but after explicit cache invalidation, it is importable
+ importlib.invalidate_caches()
+ import foo.just_created
+ self.assertEqual(foo.just_created.attr, 'just_created foo')
+
+
class SeparatedOverlappingNamespacePackages(NamespacePackageTest):
paths = ['portion1', 'both_portions']

diff --git a/Misc/NEWS.d/next/Library/2021-11-03-13-41-49.bpo-45703.35AagL.rst b/Misc/NEWS.d/next/Library/2021-11-03-13-41-49.bpo-45703.35AagL.rst
new file mode 100644
index 0000000000000..9fa9be56b8327
--- /dev/null
+++ b/Misc/NEWS.d/next/Library/2021-11-03-13-41-49.bpo-45703.35AagL.rst
@@ -0,0 +1,5 @@
+When a namespace package is imported before another module from the same
+namespace is created/installed in a different :data:`sys.path` location
+while the program is running, calling the
+:func:`importlib.invalidate_caches` function will now also guarantee the new
+module is noticed.

_______________________________________________
Python-checkins mailing list
Python-checkins@python.org
https://mail.python.org/mailman/listinfo/python-checkins