@BvB93 I have an unrelated question.

charris@fc [numpy.git (prepare-1.26.0-release)]$ git diff main numpy/f2py
diff --git a/numpy/f2py/__init__.pyi b/numpy/f2py/__init__.pyi
index 81b6a24f39..6e3a82cf8f 100644
--- a/numpy/f2py/__init__.pyi
+++ b/numpy/f2py/__init__.pyi
@@ -14,6 +14,7 @@ class _F2PyDict(_F2PyDictBase, total=False):
     ltx: list[str]
 
 __all__: list[str]
+__path__: list[str]
 test: PytestTester

I assume the difference is harmless. Is it?

I assume the difference is harmless. Is it?

Correct, it's harmless. I would recommend leaving __path__ out if it's not too much work (it's an attribute that's explicitly handled by type checkers these days), but either way is fine tbh.

The test failures look unrelated. The macos_arm64_test crash is new. Hmm ... Will rerun.

Thanks Bas.

@seiko2plus Any chance the aarch64 fixes could have accounted for the macos_arm64_test crash? I'm sorry that I reran the test, but wanted to see if it was repeatable.

Any chance the aarch64 fixes could have accounted for the macos_arm64_test crash?

I ran tests on clang 14 & 15, both were fine. If the partial load refactor (gh-24461) is the issue, then it should be consistent.