Implement unstack and fix repeat test

ndonnx/_logic_in_data/_array.py

@@ -196,6 +196,17 @@ def __dlpack__(
     def __dlpack_device__(self) -> tuple[Enum, int]:
         raise ValueError("ONNX provides no control over the used device")
 
+    def __iter__(self):
+        try:
+            n, *_ = self.shape
+        except IndexError:
+            raise ValueError("iteration over 0-d array")
+        if isinstance(n, int):
+            return (self[i, ...] for i in range(n))
+        raise ValueError(
+            "iteration requires dimension of static length, but dimension 0 is dynamic."
+        )
+
     def __getitem__(self, key: GetitemIndex, /) -> Array:
         idx = normalize_getitem_key(key)
         data = self._data[idx]

ndonnx/_logic_in_data/_funcs.py

@@ -565,7 +565,12 @@ def unique_values(x: Array, /) -> Array:
 
 
 def unstack(x: Array, /, *, axis: int = 0) -> tuple[Array, ...]:
-    raise NotImplementedError
+    # Only possible for statically known dimensions
+    if not isinstance(x.shape[axis], int):
+        raise ValueError(
+            f"'unstack' can only be applied to statically known dimensions, but axis `{axis}` has dynamic length."
+        )
+    return tuple(el for el in moveaxis(x, axis, 0))
 
 
 def vecdot(x1: Array, x2: Array, /, *, axis: int = -1) -> Array:

tests/test_logic_in_data.py

@@ -352,6 +352,13 @@ def test_indexing_value_prop_tuple_index():
         np.testing.assert_equal(el.unwrap_numpy(), np_arr[idx])
 
 
+def test_iteration():
+    np_arr = np.asarray([1, 2])
+    arr = ndx.asarray(np_arr)
+    for npa, nda in zip(np_arr, arr):
+        np.testing.assert_array_equal(npa, nda.unwrap_numpy())
+
+
 @pytest.mark.parametrize("idx", [(0, 1), (-1, ...), (..., 1), (-1, ..., 1)])
 @pytest.mark.parametrize(
     "np_array",