Add a test and comments, update changelog

docs/project/changelog.md

@@ -69,6 +69,9 @@ myst:
 - {{ Fix }} Fix sessionStorage-related crash when running in sandboxed iframe.
   {pr}`5186`
 
+- {{ Fix }} `JsFinder.find_spec()` no longer crashes when called during pytest test collection.
+  {pr}`5170`
+
 ### Packages
 
 - Upgraded `crc32c` to 2.7.1 {pr}`5169`

src/py/_pyodide/_importhook.py

@@ -25,6 +25,9 @@ def find_spec(
             try:
                 parent_module = sys.modules[parent]
             except KeyError:
+                # Note: This will never happen when we're called from importlib,
+                # but pytest hits this codepath. See
+                # `test_importhook_called_from_pytest`.
                 return None
             if not isinstance(parent_module, JsProxy):
                 # Not one of us.

src/tests/test_pyodide.py

@@ -1516,6 +1516,27 @@ def test_module_not_found_note(selenium_standalone):
     assert len(e.value.__notes__) == 1
 
 
+@run_in_pyodide
+def test_importhook_called_from_pytest(selenium):
+    """
+    Whenever importlib itself resolves `import a.b`, it splits on the . and
+    first imports `a` and then `a.b`. However, pytest does not, it calls
+    `find_spec("a.b")` directly here:
+    https://github.com/pytest-dev/pytest/blob/ea0fa639445ae08616edd2c15189a1a76168f018/src/_pytest/pathlib.py#L693-L698
+
+    This previously could lead to crashes in `JsFinder`.
+    """
+    import sys
+
+    def _import_module_using_spec(module_name):
+        """Modeled on a fragment of _pytest.pathlib._import_module_using_spec"""
+        for meta_importer in sys.meta_path:
+            spec = meta_importer.find_spec(module_name, path=[])
+
+    # Assertion: This should not raise KeyError.
+    _import_module_using_spec("a.b")
+
+
 def test_args(selenium_standalone_noload):
     selenium = selenium_standalone_noload
     assert selenium.run_js(