Merge pull request #2416 from neutrinoceros/numpy2_compat

ENH: avoid deprecation warnings from numpy 2

h5py/_hl/dataset.py

@@ -213,10 +213,16 @@ def __len__(self):
         """
         return len(self._dset)
 
-    def __array__(self, dtype=None):
+    def __array__(self, dtype=None, copy=True):
+        if copy is False:
+            raise ValueError(
+                f"AstypeWrapper.__array__ received {copy=} "
+                f"but memory allocation cannot be avoided on read"
+            )
+
         data = self[:]
         if dtype is not None:
-            data = data.astype(dtype)
+            return data.astype(dtype, copy=False)
         return data
 
 
@@ -251,7 +257,16 @@ def __len__(self):
         """
         return len(self._dset)
 
-    def __array__(self):
+    def __array__(self, dtype=None, copy=True):
+        if dtype not in (None, object):
+            raise TypeError(
+                "AsStrWrapper.__array__ doesn't support the dtype argument"
+            )
+        if copy is False:
+            raise ValueError(
+                f"AsStrWrapper.__array__ received {copy=} "
+                f"but memory allocation cannot be avoided on read"
+            )
         return numpy.array([
             b.decode(self.encoding, self.errors) for b in self._dset
         ], dtype=object).reshape(self._dset.shape)
@@ -268,11 +283,17 @@ def __init__(self, dset, prior_dtype, names):
             names = [names]
         self.read_dtype = readtime_dtype(prior_dtype, names)
 
-    def __array__(self, dtype=None):
+    def __array__(self, dtype=None, copy=True):
+        if copy is False:
+            raise ValueError(
+                f"FieldsWrapper.__array__ received {copy=} "
+                f"but memory allocation cannot be avoided on read"
+            )
         data = self[:]
         if dtype is not None:
-            data = data.astype(dtype)
-        return data
+            return data.astype(dtype, copy=False)
+        else:
+            return data
 
     def __getitem__(self, args):
         data = self._dset.__getitem__(args, new_dtype=self.read_dtype)
@@ -1049,11 +1070,16 @@ def write_direct(self, source, source_sel=None, dest_sel=None):
                 self.id.write(mspace, fspace, source, dxpl=self._dxpl)
 
     @with_phil
-    def __array__(self, dtype=None):
+    def __array__(self, dtype=None, copy=True):
         """ Create a Numpy array containing the whole dataset.  DON'T THINK
         THIS MEANS DATASETS ARE INTERCHANGEABLE WITH ARRAYS.  For one thing,
         you have to read the whole dataset every time this method is called.
         """
+        if copy is False:
+            raise ValueError(
+                f"Dataset.__array__ received {copy=} "
+                f"but memory allocation cannot be avoided on read"
+            )
         arr = numpy.zeros(self.shape, dtype=self.dtype if dtype is None else dtype)
 
         # Special case for (0,)*-shape datasets

h5py/tests/test_dataset.py

@@ -1785,6 +1785,13 @@ def test_non_contiguous_arrays(self):
 
         assert all(self.f['nc2'][0] == y[::2]), f"{self.f['nc2'][0]} != {y[::2]}"
 
+    def test_asstr_array_dtype(self):
+        dt = h5py.string_dtype(encoding='ascii')
+        fill_value = b'bar'
+        ds = self.f.create_dataset('x', (100,), dtype=dt, fillvalue=fill_value)
+        with pytest.raises(TypeError):
+            np.array(ds.asstr(), dtype=int)
+
 
 class TestLowOpen(BaseDataset):
 
@@ -2013,3 +2020,34 @@ def test_virtual_prefix_require(self):
         self.assertEqual(virtual_prefix, virtual_prefix_readback)
         self.assertIsInstance(dset, Dataset)
         self.assertEqual(dset.shape, (10, 3))
+
+
+
+COPY_IF_NEEDED = False if np.__version__.startswith("1.") else None
+
+VIEW_GETTERS = {
+    "ds": lambda ds: ds,
+    "astype": lambda ds: ds.astype(dtype=object),
+    "asstr": lambda ds: ds.asstr(),
+}
+
+@pytest.mark.parametrize("copy", [True, COPY_IF_NEEDED])
+@pytest.mark.parametrize("view_getter", VIEW_GETTERS.values(), ids=VIEW_GETTERS.keys())
+def test_array_copy(view_getter, copy, writable_file):
+    dt = h5py.string_dtype(encoding='ascii')
+    fill_value = b'bar'
+    ds = writable_file.create_dataset('x', (10,), dtype=dt, fillvalue=fill_value)
+    np.array(view_getter(ds), copy=copy)
+
+@pytest.mark.skipif(
+    np.__version__.startswith("1."),
+    reason="forbidding copies requires numpy 2",
+)
+@pytest.mark.parametrize("view_getter", VIEW_GETTERS.values(), ids=VIEW_GETTERS.keys())
+def test_array_copy_false(view_getter, writable_file):
+    dt = h5py.string_dtype(encoding='ascii')
+    fill_value = b'bar'
+    ds = writable_file.create_dataset('x', (10,), dtype=dt, fillvalue=fill_value)
+    view = view_getter(ds)
+    with pytest.raises(ValueError):
+        np.array(view, copy=False)

h5py/tests/test_dtype.py

@@ -131,7 +131,7 @@ class TestEmptyVlen(TestCase):
     def test_write_empty_vlen(self):
         fname = self.mktemp()
         with h5py.File(fname, 'w') as f:
-            d = np.core.records.fromarrays([[], []], names='a,b', formats='|V16,O')
+            d = np.rec.fromarrays([[], []], names='a,b', formats='|V16,O')
             dset = f.create_dataset('test', data=d, dtype=[('a', '|V16'), ('b', h5py.special_dtype(vlen=np.float64))])
             self.assertEqual(dset.size, 0)