Fix diagonal() strides calculation for empty results

CHANGELOG.md

@@ -76,6 +76,7 @@ Also, that release drops support for Python 3.9, making Python 3.10 the minimum
 * Resolved an issue causing `dpnp.linspace` to return an incorrect output shape when inputs were passed as arrays [#2712](https://github.com/IntelPython/dpnp/pull/2712)
 * Resolved an issue where `dpnp` always returns the base allocation pointer, when the view start is expected [#2651](https://github.com/IntelPython/dpnp/pull/2651)
 * Fixed an issue causing an exception in `dpnp.geomspace` and `dpnp.logspace` when called with explicit `device` keyword but any input array is allocated on another device [#2723](https://github.com/IntelPython/dpnp/pull/2723)
+* Resolved an issue with strides calculation in `dpnp.diagonal` to return correct values for empty diagonals [#2814](https://github.com/IntelPython/dpnp/pull/2814)
 
 ### Security
 

dpnp/dpnp_iface_indexing.py

@@ -719,24 +719,18 @@ def diagonal(a, offset=0, axis1=0, axis2=1):
         offset = -offset
 
     a_shape = a.shape
-    a_straides = a.strides
+    a_strides = a.strides
     n, m = a_shape[-2:]
-    st_n, st_m = a_straides[-2:]
-
-    # Compute shape, strides and offset of the resulting diagonal array
-    # based on the input offset
-    if offset == 0:
-        out_shape = a_shape[:-2] + (min(n, m),)
-        out_strides = a_straides[:-2] + (st_n + st_m,)
-        out_offset = 0
-    elif 0 < offset < m:
-        out_shape = a_shape[:-2] + (min(n, m - offset),)
-        out_strides = a_straides[:-2] + (st_n + st_m,)
-        out_offset = st_m // a.itemsize * offset
-    else:
-        out_shape = a_shape[:-2] + (0,)
-        out_strides = a_straides[:-2] + (a.itemsize,)
-        out_offset = 0
+    st_n, st_m = a_strides[-2:]
+
+    # Compute the diagonal array as a view:
+    # - stride: sum of row and column strides (diag advances in both dimensions)
+    # - shape: determined by diagonal size using max(0, min(n, m - offset))
+    # - offset: starting position in buffer for non-zero offsets
+    diag_size = max(0, min(n, m - offset))
+    out_shape = a_shape[:-2] + (diag_size,)
+    out_strides = a_strides[:-2] + (st_n + st_m,)
+    out_offset = st_m // a.itemsize * offset
 
     return dpnp_array(
         out_shape, buffer=a, strides=out_strides, offset=out_offset

dpnp/tests/test_indexing.py

@@ -18,6 +18,7 @@
 from dpnp.exceptions import AxisError, ExecutionPlacementError
 
 from .helper import (
+    generate_random_numpy_array,
     get_abs_array,
     get_all_dtypes,
     get_array,
@@ -44,7 +45,9 @@ def wrapped(a, axis, **kwargs):
 
 
 class TestDiagonal:
-    @pytest.mark.parametrize("dtype", get_all_dtypes(no_bool=True))
+    @pytest.mark.parametrize(
+        "dtype", get_all_dtypes(no_none=True, no_bool=True)
+    )
     @pytest.mark.parametrize("offset", [-3, -1, 0, 1, 3])
     @pytest.mark.parametrize(
         "shape",
@@ -58,7 +61,7 @@ class TestDiagonal:
             "(2, 2, 2, 3)",
         ],
     )
-    def test_diagonal_offset(self, shape, dtype, offset):
+    def test_offset(self, shape, dtype, offset):
         a = numpy.arange(numpy.prod(shape), dtype=dtype).reshape(shape)
         a_dp = dpnp.array(a)
         expected = numpy.diagonal(a, offset)
@@ -74,7 +77,7 @@ def test_diagonal_offset(self, shape, dtype, offset):
             ((4, 3, 5, 2), [(0, 1), (1, 2), (2, 3), (0, 3)]),
         ],
     )
-    def test_diagonal_axes(self, shape, axis_pairs, dtype):
+    def test_axes(self, shape, axis_pairs, dtype):
         a = numpy.arange(numpy.prod(shape), dtype=dtype).reshape(shape)
         a_dp = dpnp.array(a)
         for axis1, axis2 in axis_pairs:
@@ -91,7 +94,7 @@ def test_linalg_diagonal(self, offset):
         result = dpnp.linalg.diagonal(a_dp, offset=offset)
         assert_array_equal(expected, result)
 
-    def test_diagonal_errors(self):
+    def test_errors(self):
         a = dpnp.arange(12).reshape(3, 4)
 
         # unsupported type
@@ -115,6 +118,64 @@ def test_diagonal_errors(self):
         assert_raises(ValueError, a.diagonal, axis1=1, axis2=1)
         assert_raises(ValueError, a.diagonal, axis1=1, axis2=-1)
 
+    @pytest.mark.parametrize("dt", get_all_dtypes(no_none=True))
+    @pytest.mark.parametrize(
+        "shape, offset",
+        [
+            ((2, 5), 5),  # offset >= m
+            ((2, 5), 10),  # offset >> m
+            ((4, 5), 6),  # offset >= m
+            ((2, 5), -5),  # negative offset >= n
+            ((3, 3, 4), 5),  # 3D array, offset >= m
+        ],
+    )
+    def test_empty_strides(self, dt, shape, offset):
+        a = generate_random_numpy_array(shape=shape, dtype=dt)
+        ia = dpnp.array(a)
+
+        expected = numpy.diagonal(a, offset)
+        result = dpnp.diagonal(ia, offset)
+
+        # Check both shape and strides match NumPy
+        assert expected.shape == result.shape
+        assert expected.strides == result.strides
+        assert_array_equal(expected, result)
+
+    @pytest.mark.parametrize("dt", get_all_dtypes(no_none=True))
+    def test_view(self, dt):
+        a = generate_random_numpy_array(shape=(3, 4), dtype=dt)
+        a = dpnp.array(a)
+        ia = a.copy()
+
+        diag = dpnp.diagonal(a)
+        diag[1] = 17  # modify a diagonal element
+        ia[1, 1] = 17  # do the same in original copy of the array
+
+        assert (a == ia).all()
+
+    @pytest.mark.parametrize("dt", get_all_dtypes(no_none=True))
+    @pytest.mark.parametrize(
+        "slice_spec, offset",
+        [
+            ((slice(None), slice(None, None, 2)), 0),  # skip columns
+            ((slice(None, None, 2), slice(None)), 1),  # skip rows
+            ((slice(None, None, 2), slice(None, None, 2)), 0),  # skip both
+        ],
+    )
+    def test_noncontiguous(self, dt, slice_spec, offset):
+        a = generate_random_numpy_array(shape=(4, 6), dtype=dt)
+        a_sliced = a[slice_spec]
+        ia = dpnp.array(a)
+        ia_sliced = ia[slice_spec]
+
+        expected = numpy.diagonal(a_sliced, offset=offset)
+        result = dpnp.diagonal(ia_sliced, offset=offset)
+
+        # Check strides match for non-contiguous arrays
+        assert expected.shape == result.shape
+        assert expected.strides == result.strides
+        assert_array_equal(expected, result)
+
 
 class TestExtins:
     @pytest.mark.parametrize("dt", get_all_dtypes(no_none=True))