Don't rely on the module cache when "importing self"

[DanielNoord]

Steps

• For new features or bug fixes, add a ChangeLog entry describing what your PR does.
• Write a good description on what the PR does.

Description

Relevant historic commits are:
6d59ad0
9684d0f

Basically, by not relying on the cache we trigger a new import system and check and rightfully end up with the stdlib math module instead of the current module. I tested with pylint and test suite there still passes.

Refs: pylint-dev/pylint#7289 and pylint-dev/pylint#5151, but mostly the latter.

Type of Changes

	Type
✓	🐛 Bug fix

[DanielNoord]

Note that the old math stays in place in the cache. This is how we handle all cache clashes so I think this is also good in this case?

[coveralls]

Pull Request Test Coverage Report for Build 2911196102

• 7 of 7 (100.0%) changed or added relevant lines in 3 files are covered.
• 21 unchanged lines in 2 files lost coverage.
• Overall coverage increased (+0.09%) to 92.345%

---

Files with Coverage Reduction	New Missed Lines	%
astroid/interpreter/objectmodel.py	6	92.63%
astroid/brain/brain_dataclasses.py	15	91.63%

Totals
Change from base Build 2891370842:	0.09%
Covered Lines:	9711
Relevant Lines:	10516

---

💛 - Coveralls

[jacobtylerwalls]

Had some time to do a quick first pass.

[Pierre-Sassoulas]

Nice MR, I like the typing update too :)

[Pierre-Sassoulas]

Suggested change

	def find_in_testdata(name: str) -> Path:
	def construct_absolute_path_in_testdata(name: str) -> Path:

For me the verb "find" imply searching for it by walking the directory tree. I would expect the give "foo.py" and get back the file in testdata/bar/ called foo.py

[DanielNoord]

I changed it to use get_testdata_path. The other one is a bit long.

[Pierre-Sassoulas]

Suggested change

	data_dir = resources.find_in_testdata("import_conflicting_names")
	math_file = data_dir / "math.py"
	math_file = resources.construct_absolute_path_in_testdata("import_conflicting_names/math.py")

I think this would work ?

[DanielNoord]

Ready for review again! @Pierre-Sassoulas @jacobtylerwalls 😄

[jacobtylerwalls]

The test passes for me on main. Would you have the opportunity to create a test that fails before the fix?

[DanielNoord]

The test passes for me on main. Would you have the opportunity to create a test that fails before the fix?

Are you sure?

diff --git a/tests/unittest_manager.py b/tests/unittest_manager.py
index 9cf02aff8..fe1f22642 100644
--- a/tests/unittest_manager.py
+++ b/tests/unittest_manager.py
@@ -325,6 +325,22 @@ class AstroidManagerTest(
             with self.assertRaises(AstroidBuildingError):
                 self.manager.ast_from_module_name("foo.bar.baz")
 
+    def test_same_name_import_module(self) -> None:
+        """Test inference of an import statement with the same name as the module.
+        See https://github.com/PyCQA/pylint/issues/5151.
+        """
+        math_file = resources.find("data/import_conflicting_names/math.py")
+        module = self.manager.ast_from_file(math_file)
+
+        # Change the cache key and module name to mimic importing the test file
+        # from the root/top level. This creates a clash between math.py and stdlib math.
+        self.manager.astroid_cache["math"] = self.manager.astroid_cache.pop(module.name)
+        module.name = "math"
+
+        # Infer the 'import math' statement
+        stdlib_math = next(module.body[1].value.args[0].infer())
+        assert self.manager.astroid_cache["math"] != stdlib_math
+
 
 class BorgAstroidManagerTC(unittest.TestCase):
     def test_borg(self) -> None:

together with the math.py file

Fails for me with pytest -k test_same_name_import_module on:

>       assert self.manager.astroid_cache["math"] != stdlib_math
E       assert <Module.math l.0 at 0x110670210> != <Module.math l.0 at 0x110670210>

[DanielNoord]

Oh never mind. That doesn't actually fail. You just can't Module != Module. I'll have a look today.

Edit: @jacobtylerwalls I should have taken my coffee. That assertion fails correctly. On main self.manager.astroid_cache["math"] == stdlib_math for me, which shows that we are still using the cache as stdlib_math is the math.py file Module object.

[jacobtylerwalls]

Yeah, I did something horribly wrong when testing it, I guess. Sorry!

[emcd]

EDIT: This PR may cause breakage with the use of importlib.import_module on PyPy 3.8. Please see #1755 for reference. The breakage actually comes from earlier - this commit is not the most recent release. Apologies.