Fix dpnp.linspace returning NaN for equal infinite endpoints

[antonwolfy]

dpnp.linspace returned NaN instead of the endpoint value when start and stop are equal and infinite — e.g. dpnp.linspace(inf, inf, 4) gave [nan inf inf nan].

The bug had two independent sources:

• The array path in dpnp_linspace computed delta = stop - start, so inf - inf = NaN propagated through the whole result.
• The scalar path delegates to dpnp.tensor.linspace, whose affine kernel LinearSequenceAffineFunctor evaluates start * w + stop * wc; at the endpoints one weight is 0, and inf * 0 = NaN.

The PR proposes to fix all the places.

Note, the tests compare against NumPy on versions that include the fix (NumPy >= 2.6.0) and fall back to explicit expected values on older NumPy.

• Have you provided a meaningful PR description?
• Have you added a test, reproducer or referred to an issue with a reproducer?
• Have you tested your changes locally for CPU and GPU devices?
• Have you made sure that new changes do not introduce compiler warnings?
• Have you checked performance impact of proposed changes?
• Have you added documentation for your changes, if necessary?
• Have you added your changes to the changelog?

[github-actions]

Array API standard conformance tests for dpnp=0.21.0dev7=py314ha0e2e8e_15 ran successfully.
Passed: 1375
Failed: 0
Skipped: 7

[github-actions]

View rendered docs @ https://intelpython.github.io/dpnp/pull/3043/index.html

[coveralls]

coverage: 78.492%. remained the same — fix-linspace-inf-endpoints into master

[vlad-perevezentsev]

LGTM
Thank you @antonwolfy

One separate issue for a follow-up
dpnp.linspace(-dpnp.inf, dpnp.inf, num) returns NaN for the last value instead of stop

In [4]: dpnp.linspace(-dpnp.inf, dpnp.inf, 4)
Out[4]: array([nan, nan, nan, nan])

In [6]: numpy.linspace(-numpy.inf, numpy.inf, 4)
/home/sdp/miniforge3/envs/dpnp_conda/lib/python3.14/site-packages/numpy/_core/function_base.py:163: RuntimeWarning: invalid value encountered in multiply
  y *= step
/home/sdp/miniforge3/envs/dpnp_conda/lib/python3.14/site-packages/numpy/_core/function_base.py:173: RuntimeWarning: invalid value encountered in add
  y += start
Out[6]: array([nan, nan, nan, inf])

[antonwolfy]

Just to note, NumPy behavior also doesn't seem correct, since the result should be:

[-inf, nan, nan, inf]

Since linspace(start, stop, num, endpoint=True) should return evenly spaced points on [−∞, +∞]. And MatLab returns exactly the above values.

I believe the only way to do that in dpnp/dpnp.tensor now is to have a special handling for the first and last elements of the interval.
But I wonder if that is needed for such edge case with no real use case.

[antonwolfy]

It seems the same is applicable to similar use cases:

np.linspace(0, np.inf, 4)
# Out: array([nan, inf, inf, inf])
# Expected:  [0.,  inf, inf, inf]

# np.linspace(-np.inf, 0, 4)
# Out: array([nan,   nan,  nan, 0.])
# Expected:  [-inf, -inf, -inf, 0.]