WIP: add compatibility shims for {eig,eigvals} #379

ev-br · 2026-01-10T19:45:54Z

WIP until 2025.12 is released.

dask does not have eig, thus xfail the tests
numpy needs a shim for return dtype stability
cupy needs an xfail until the min cupy version is <=13

data-apis/array-api-tests#404 is the matching test PR

We need a wrapper because numpy currently returns `float|complex`. Implementation-wise, follow `linalg.solve` and copy-paste relevant numpy code with minimal required modifications.

Copilot

Pull request overview

This PR adds compatibility shims for the eig and eigvals linear algebra functions to support the 2025.12 array API standard. The shims ensure dtype stability for NumPy by always returning complex dtypes, as required by the Array API specification, while also handling xfails for libraries that don't yet support these functions.

Changes:

Added EigResult namedtuple class to common/_linalg.py
Implemented eig() and eigvals() shim functions in numpy/linalg.py with dtype conversion logic
Added xfail entries for dask test suite (dask doesn't have these functions yet)

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 4 comments.

File	Description
dask-xfails.txt	Added xfail entries for test_eig and test_eigvals under 2025.12 support section
array_api_compat/common/_linalg.py	Added EigResult namedtuple to match the pattern of other result types
array_api_compat/numpy/linalg.py	Implemented eig() and eigvals() shims with dtype stability fixes and imported EigResult

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-01-10T19:51:11Z

array_api_compat/numpy/linalg.py

+def eig(x: Array, /) -> tuple[Array, Array]:
+    try:
+        from numpy.linalg._linalg import (  # type: ignore[attr-defined]
+            _assert_stacked_square,
+            _assert_finite,
+            _commonType,
+            _makearray,
+            _raise_linalgerror_eigenvalues_nonconvergence,
+            isComplexType,
+            _complexType,
+        )
+    except ImportError:
+        from numpy.linalg.linalg import (  # type: ignore[attr-defined]
+            _assert_stacked_square,
+            _assert_finite,
+            _commonType,
+            _makearray,
+            _raise_linalgerror_eigenvalues_nonconvergence,
+            isComplexType,
+            _complexType,
+        )
+    from numpy.linalg import _umath_linalg
+
+    x, wrap = _makearray(x)
+    _assert_stacked_square(x)
+    _assert_finite(x)
+    t, result_t = _commonType(x)
+
+    signature = 'D->DD' if isComplexType(t) else 'd->DD'
+    with np.errstate(call=_raise_linalgerror_eigenvalues_nonconvergence,
+                  invalid='call', over='ignore', divide='ignore',
+                  under='ignore'):
+        w, vt = _umath_linalg.eig(x, signature=signature)
+
+    result_t = _complexType(result_t)
+    vt = vt.astype(result_t, copy=False)
+    return EigResult(w.astype(result_t, copy=False), wrap(vt))


The return type annotation for the eig function indicates it returns a tuple[Array, Array], but the implementation returns an EigResult (a NamedTuple). This inconsistency should be corrected to return EigResult instead of tuple[Array, Array] to match the actual behavior and to be consistent with similar functions like eigh which properly returns EighResult.

Copilot · 2026-01-10T19:51:11Z

array_api_compat/numpy/linalg.py

+        )
+    from numpy.linalg import _umath_linalg
+
+    x, wrap = _makearray(x)


The variable 'wrap' returned from _makearray is not used in the eigvals function, but is assigned. This suggests either it should be used (similar to how it's used in the eig function to wrap the result) or should be replaced with an underscore to indicate it's intentionally unused.

Copilot · 2026-01-10T19:51:11Z

array_api_compat/numpy/linalg.py

+def eig(x: Array, /) -> tuple[Array, Array]:
+    try:
+        from numpy.linalg._linalg import (  # type: ignore[attr-defined]
+            _assert_stacked_square,
+            _assert_finite,
+            _commonType,
+            _makearray,
+            _raise_linalgerror_eigenvalues_nonconvergence,
+            isComplexType,
+            _complexType,
+        )
+    except ImportError:
+        from numpy.linalg.linalg import (  # type: ignore[attr-defined]
+            _assert_stacked_square,
+            _assert_finite,
+            _commonType,
+            _makearray,
+            _raise_linalgerror_eigenvalues_nonconvergence,
+            isComplexType,
+            _complexType,
+        )
+    from numpy.linalg import _umath_linalg
+
+    x, wrap = _makearray(x)
+    _assert_stacked_square(x)
+    _assert_finite(x)
+    t, result_t = _commonType(x)
+
+    signature = 'D->DD' if isComplexType(t) else 'd->DD'
+    with np.errstate(call=_raise_linalgerror_eigenvalues_nonconvergence,
+                  invalid='call', over='ignore', divide='ignore',
+                  under='ignore'):
+        w, vt = _umath_linalg.eig(x, signature=signature)
+
+    result_t = _complexType(result_t)
+    vt = vt.astype(result_t, copy=False)
+    return EigResult(w.astype(result_t, copy=False), wrap(vt))
+
+
+def eigvals(x: Array, /) -> Array:
+    try:
+        from numpy.linalg._linalg import (  # type: ignore[attr-defined]
+            _assert_stacked_square,
+            _assert_finite,
+            _commonType,
+            _makearray,
+            _raise_linalgerror_eigenvalues_nonconvergence,
+            isComplexType,
+            _complexType,
+        )
+    except ImportError:
+        from numpy.linalg.linalg import (  # type: ignore[attr-defined]
+            _assert_stacked_square,
+            _assert_finite,
+            _commonType,
+            _makearray,
+            _raise_linalgerror_eigenvalues_nonconvergence,
+            isComplexType,
+            _complexType,
+        )
+    from numpy.linalg import _umath_linalg
+
+    x, wrap = _makearray(x)
+    _assert_stacked_square(x)
+    _assert_finite(x)
+    t, result_t = _commonType(x)
+
+    signature = 'D->D' if isComplexType(t) else 'd->D'
+    with np.errstate(call=_raise_linalgerror_eigenvalues_nonconvergence,
+                  invalid='call', over='ignore', divide='ignore',
+                  under='ignore'):
+        w = _umath_linalg.eigvals(x, signature=signature)
+
+    result_t = _complexType(result_t)
+    return w.astype(result_t, copy=False)


There is significant code duplication between the eig and eigvals functions. Both functions import the same set of helper functions from numpy.linalg, perform the same validation steps (_assert_stacked_square, _assert_finite), and use similar dtype handling logic. Consider extracting the common logic into a shared helper function to improve maintainability and reduce duplication.

Copilot · 2026-01-10T19:51:11Z

dask-xfails.txt

+# 2025.12 support
+array_api_tests/test_linalg.py::test_eig
+array_api_tests/test_linalg.py::test_eigvals
+


According to the PR description, CuPy needs xfail entries for eig and eigvals tests until the minimum CuPy version is >=13. However, these xfail entries are missing from cupy-xfails.txt. The dask-xfails.txt file includes the necessary entries (lines 133-134), but cupy-xfails.txt should also have corresponding entries under a "# 2025.12 support" section similar to the pattern used in dask-xfails.txt.

ev-br added 2 commits January 10, 2026 19:32

TST: dask.linalg.eig is not a thing

bc7b7b9

ENH: numpy: add linalg.{eig,eigvals}

d8d1e51

We need a wrapper because numpy currently returns `float|complex`. Implementation-wise, follow `linalg.solve` and copy-paste relevant numpy code with minimal required modifications.

Copilot AI review requested due to automatic review settings January 10, 2026 19:45

ev-br marked this pull request as draft January 10, 2026 19:45

Copilot started reviewing on behalf of ev-br January 10, 2026 19:46 View session

TST: cupy: xfail eig tests

9d2ec8e

ev-br mentioned this pull request Jan 10, 2026

WIP: add tests for eig and eigvals data-apis/array-api-tests#404

Draft

5 tasks

Copilot AI reviewed Jan 10, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

WIP: add compatibility shims for {eig,eigvals} #379

WIP: add compatibility shims for {eig,eigvals} #379

ev-br commented Jan 10, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Jan 10, 2026

Uh oh!

Copilot AI Jan 10, 2026

Uh oh!

Copilot AI Jan 10, 2026

Uh oh!

Copilot AI Jan 10, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

WIP: add compatibility shims for {eig,eigvals} #379

Are you sure you want to change the base?

WIP: add compatibility shims for {eig,eigvals} #379

Conversation

ev-br commented Jan 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Jan 10, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 10, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 10, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 10, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

ev-br commented Jan 10, 2026 •

edited

Loading