ENH: Add tests for linalg.getattr deprecation mechanism #29898
Description
Enhancement Proposal: Tests for linalg.__getattr__ Deprecation
Description:
The __getattr__ hook in numpy/linalg/linalg.py (around line 10) forwards attributes from the private _linalg module while issuing a DeprecationWarning to guide users to the public API. This is great for backward compatibility, but it currently lacks dedicated tests in numpy/linalg/tests/test_linalg.py for key behaviors like:
- Successful attribute forwarding (e.g.,
inv,det) with correct warning emission. - Raising
AttributeErrorfor missing attributes. - Accurate
stacklevel=3in warnings for proper traceback pointing.
Adding these would ensure reliability, especially for ML workflows relying on linalg (e.g., SVD in data augmentation or solving in regression, as in my projects using NumPy with TensorFlow/scikit-learn).
Proposed Changes:
- Add a new test class
TestLinalgGetattrDeprecationwith pytest functions:test_getattr_successful: Captures warning and asserts array equality (e.g., forinvon a sample matrix).test_getattr_missing: AssertsAttributeErrorfor non-existent attr likefoo.test_getattr_stacklevel: Verifies warning frame points to caller.
- Use
warnings.catch_warningsandnumpy.testing.assert_array_equalfor assertions. - Aim for 100% coverage of the
__getattr__logic.
Minimal Repro/Expected Behavior:
Here's a snippet demonstrating the current behavior (tested on NumPy 2.1.0, Python 3.12):
import numpy as np
import warnings
# Successful access
with warnings.catch_warnings(record=True) as w:
warnings.simplefilter("always")
A = np.array([[1, 2], [3, 4]], dtype=float)
result = np.linalg.linalg.inv(A) # Triggers warning
expected = np.linalg.inv(A)
# Assert result == expected and warning emitted
print("Warning emitted:", len(w) == 1)
print("Warning message:", str(w[0].message) if w else "No warning")
# Missing attr
try:
_ = np.linalg.linalg.foo
except AttributeError as e:
print("Error:", str(e))