swev-id: astropy__astropy-13236 – keep structured ndarrays as Columns

astropy/table/table.py

@@ -1239,13 +1239,6 @@ def _convert_data_to_col(self, data, copy=True, default_name=None, dtype=None, n
                                 f'{fully_qualified_name} '
                                 'did not return a valid mixin column')
 
-        # Structured ndarray gets viewed as a mixin unless already a valid
-        # mixin class
-        if (not isinstance(data, Column) and not data_is_mixin
-                and isinstance(data, np.ndarray) and len(data.dtype) > 1):
-            data = data.view(NdarrayMixin)
-            data_is_mixin = True
-
         # Get the final column name using precedence.  Some objects may not
         # have an info attribute. Also avoid creating info as a side effect.
         if not name:

astropy/table/tests/test_mixin.py

@@ -698,11 +698,8 @@ def test_skycoord_representation():
 
 
 def test_ndarray_mixin():
-    """
-    Test directly adding a plain structured array into a table instead of the
-    view as an NdarrayMixin.  Once added as an NdarrayMixin then all the previous
-    tests apply.
-    """
+    """Structured ndarrays become plain Columns unless explicitly viewed."""
+
     a = np.array([(1, 'a'), (2, 'b'), (3, 'c'), (4, 'd')],
                  dtype='<i4,' + ('|U1'))
     b = np.array([(10, 'aa'), (20, 'bb'), (30, 'cc'), (40, 'dd')],
@@ -711,43 +708,27 @@ def test_ndarray_mixin():
                            names=['rx', 'ry'])
     d = np.arange(8, dtype='i8').reshape(4, 2).view(NdarrayMixin)
 
-    # Add one during initialization and the next as a new column.
+    # Add one during initialization and the next as new columns.
     t = Table([a], names=['a'])
     t['b'] = b
     t['c'] = c
     t['d'] = d
 
-    assert isinstance(t['a'], NdarrayMixin)
-
-    assert t['a'][1][1] == a[1][1]
-    assert t['a'][2][0] == a[2][0]
-
-    assert t[1]['a'][1] == a[1][1]
-    assert t[2]['a'][0] == a[2][0]
-
-    assert isinstance(t['b'], NdarrayMixin)
+    for name, arr in (('a', a), ('b', b), ('c', c)):
+        col = t[name]
+        assert isinstance(col, Column)
+        assert col.dtype == arr.dtype
+        assert np.array_equal(col, arr)
 
-    assert t['b'][1]['x'] == b[1]['x']
-    assert t['b'][1]['y'] == b[1]['y']
-
-    assert t[1]['b']['x'] == b[1]['x']
-    assert t[1]['b']['y'] == b[1]['y']
-
-    assert isinstance(t['c'], NdarrayMixin)
-
-    assert t['c'][1]['rx'] == c[1]['rx']
-    assert t['c'][1]['ry'] == c[1]['ry']
-
-    assert t[1]['c']['rx'] == c[1]['rx']
-    assert t[1]['c']['ry'] == c[1]['ry']
+        # Row-first access keeps the structured dtype
+        row = t[1][name]
+        for field in arr.dtype.names:
+            assert col[1][field] == arr[field][1]
+            assert row[field] == arr[field][1]
 
+    # Explicit NdarrayMixin view still produces a mixin column
     assert isinstance(t['d'], NdarrayMixin)
-
-    assert t['d'][1][0] == d[1][0]
-    assert t['d'][1][1] == d[1][1]
-
-    assert t[1]['d'][0] == d[1][0]
-    assert t[1]['d'][1] == d[1][1]
+    assert np.array_equal(t['d'], d)
 
     assert t.pformat(show_dtype=True) == [
         '  a [f0, f1]     b [x, y]      c [rx, ry]      d    ',
@@ -759,6 +740,78 @@ def test_ndarray_mixin():
         "     (4, 'd')    (40, 'dd')   (400., 'rdd')   6 .. 7"]
 
 
+def _structured_sample():
+    return np.array([(1, 2.0), (3, 4.5)], dtype=[('x', 'i4'), ('y', 'f8')])
+
+
+def test_structured_add_column_becomes_column():
+    t = Table()
+    struct = _structured_sample()
+    t.add_column(struct, name='a')
+
+    assert isinstance(t['a'], Column)
+    assert t['a'].dtype == struct.dtype
+    assert np.array_equal(t['a'], struct)
+
+
+def test_structured_setitem_becomes_column():
+    t = Table()
+    struct = _structured_sample()
+    t['b'] = struct
+
+    assert isinstance(t['b'], Column)
+    assert np.array_equal(t['b'], struct)
+
+
+def test_structured_replace_column_becomes_column():
+    t = Table()
+    t['base'] = Column([10, 20], name='base')
+    struct = _structured_sample()
+
+    t.replace_column('base', struct)
+
+    assert isinstance(t['base'], Column)
+    assert t['base'].dtype == struct.dtype
+    assert np.array_equal(t['base'], struct)
+
+
+def test_structured_constructor_list_of_columns():
+    a = _structured_sample()
+    b = np.array([(5, 'x'), (6, 'y')], dtype=[('col', 'i2'), ('label', 'U1')])
+
+    t = Table([a, b], names=['a', 'b'])
+
+    for name, arr in (('a', a), ('b', b)):
+        assert isinstance(t[name], Column)
+        assert t[name].dtype == arr.dtype
+        assert np.array_equal(t[name], arr)
+
+
+def test_structured_constructor_splits_fields():
+    struct = np.array([(1, 2.0), (3, 4.5)], dtype=[('x', 'i4'), ('y', 'f8')])
+
+    t = Table(struct)
+
+    assert list(t.colnames) == ['x', 'y']
+    assert np.array_equal(t['x'], struct['x'])
+    assert np.array_equal(t['y'], struct['y'])
+
+
+def test_structured_column_ecsv_roundtrip(tmp_path):
+    struct = _structured_sample()
+    t = Table()
+    t['a'] = struct
+
+    path = tmp_path / 'structured.ecsv'
+    t.write(path, format='ascii.ecsv', overwrite=True)
+
+    t2 = Table.read(path, format='ascii.ecsv')
+
+    assert isinstance(t2['a'], Column)
+    assert t2['a'].dtype == struct.dtype
+    assert np.array_equal(t2['a'], struct)
+
+
 def test_possible_string_format_functions():
     """
     The QuantityInfo info class for Quantity implements a

cextern/cfitsio/lib/putcol.c

@@ -1111,7 +1111,7 @@ int ffiter(int n_cols,
     long rept, rowrept, width, tnull, naxes[9] = {1,1,1,1,1,1,1,1,1}, groups;
     double zeros = 0.;
     char message[FLEN_ERRMSG], keyname[FLEN_KEYWORD], nullstr[FLEN_VALUE];
-    char **stringptr, *nullptr, *cptr;
+    char **stringptr, *null_ptr, *cptr;
 
     if (*status > 0)
         return(*status);
@@ -1885,24 +1885,24 @@ int ffiter(int n_cols,
           {
             stringptr = cols[jj].array;
             dataptr = stringptr + 1;
-            nullptr = *stringptr;
+            null_ptr = *stringptr;
             nbytes = 2;
           }
           else
           {
             dataptr = (char *) cols[jj].array + col[jj].nullsize;
-            nullptr = (char *) cols[jj].array;
+            null_ptr = (char *) cols[jj].array;
             nbytes = col[jj].nullsize;
           }
 
-          if (memcmp(nullptr, &zeros, nbytes) ) 
+          if (memcmp(null_ptr, &zeros, nbytes) ) 
           {
             /* null value flag not zero; must check for and write nulls */
             if (hdutype == IMAGE_HDU)   
             {
                 if (ffppn(cols[jj].fptr, cols[jj].datatype, 
                       felement, cols[jj].repeat * ntodo, dataptr,
-                      nullptr, &tstatus) > 0)
+                      null_ptr, &tstatus) > 0)
                 break;
             }
             else
@@ -1917,7 +1917,7 @@ int ffiter(int n_cols,
 
                 if (ffpcn(cols[jj].fptr, cols[jj].datatype, cols[jj].colnum, frow,
                       felement, cols[jj].repeat * ntodo, dataptr,
-                      nullptr, &tstatus) > 0)
+                      null_ptr, &tstatus) > 0)
                 break;
             }
           }

pyproject.toml

@@ -2,7 +2,7 @@
 requires = ["setuptools",
             "setuptools_scm>=6.2",
             "wheel",
-            "cython==0.29.22",
+            "cython>=0.29.37,<0.30",
             "oldest-supported-numpy",
             "extension-helpers"]
 build-backend = 'setuptools.build_meta'