Merge pull request numpy#18999 from seberg/raise-futurewarning-string…

…-promotion

DEP: Ensure the string promotion FutureWarning is raised

doc/release/upcoming_changes/18116.future.rst

@@ -11,9 +11,8 @@ are:
   a string result.
 * `numpy.array` and related functions will start returning ``object``
   arrays because these functions use ``object`` as a fallback when
-  no common dtype can be found. (In this case setting the
-  ``FutureWarning`` to be raised will unfortunately lead to the new
-  behaviour)
+  no common dtype can be found.  However, it may happen that future
+  releases of NumPy will generally error in these cases.
 
 This will mainly affect code such as::
 
@@ -24,6 +23,7 @@ and::
     np.concatenate((['string'], [0]))
 
 in both cases adding ``dtype="U"`` or ``dtype="S"`` will give the
-previous (string) result.
+previous (string) result, while ``dtype=object`` will ensure an array with
+object dtype is returned.
 
 Comparisons, universal functions, and casting are not affected by this.

numpy/core/src/multiarray/array_coercion.c

@@ -622,8 +622,12 @@ handle_promotion(PyArray_Descr **out_descr, PyArray_Descr *descr,
     }
     PyArray_Descr *new_descr = PyArray_PromoteTypes(descr, *out_descr);
     if (NPY_UNLIKELY(new_descr == NULL)) {
-        if (fixed_DType != NULL) {
-            /* If a DType is fixed, promotion must not fail. */
+        if (fixed_DType != NULL || PyErr_ExceptionMatches(PyExc_FutureWarning)) {
+            /*
+             * If a DType is fixed, promotion must not fail. Do not catch
+             * FutureWarning (raised for string+numeric promotions). We could
+             * only catch TypeError here or even always raise the error.
+             */
             return -1;
         }
         PyErr_Clear();

numpy/core/tests/test_deprecations.py

@@ -1121,10 +1121,7 @@ def test_deprecated(self, dtype, string_dt):
         self.assert_deprecated(lambda: np.concatenate((arr1, arr2), axis=0))
         self.assert_deprecated(lambda: np.concatenate((arr1, arr2), axis=None))
 
-        # coercing to an array is similar, but will fall-back to `object`
-        # (when raising the FutureWarning, this already happens)
-        self.assert_deprecated(lambda: np.array([arr1[0], arr2[0]]),
-                               exceptions=())
+        self.assert_deprecated(lambda: np.array([arr1[0], arr2[0]]))
 
     @pytest.mark.parametrize("dtype", "?bhilqpBHILQPefdgFDG")
     @pytest.mark.parametrize("string_dt", ["S", "U"])

numpy/core/tests/test_regression.py

@@ -503,8 +503,8 @@ def test_swap_real(self):
         assert_equal(np.arange(4, dtype='<c8').real.max(), 3.0)
 
     def test_object_array_from_list(self):
-        # Ticket #270
-        assert_(np.array([1, 'A', None]).shape == (3,))
+        # Ticket #270 (gh-868)
+        assert_(np.array([1, None, 'A']).shape == (3,))
 
     def test_multiple_assign(self):
         # Ticket #273
@@ -2052,18 +2052,18 @@ def test_search_sorted_invalid_arguments(self):
 
     def test_string_truncation(self):
         # Ticket #1990 - Data can be truncated in creation of an array from a
-        # mixed sequence of numeric values and strings
+        # mixed sequence of numeric values and strings (gh-2583)
         for val in [True, 1234, 123.4, complex(1, 234)]:
-            for tostr in [asunicode, asbytes]:
-                b = np.array([val, tostr('xx')])
+            for tostr, dtype in [(asunicode, "U"), (asbytes, "S")]:
+                b = np.array([val, tostr('xx')], dtype=dtype)
                 assert_equal(tostr(b[0]), tostr(val))
-                b = np.array([tostr('xx'), val])
+                b = np.array([tostr('xx'), val], dtype=dtype)
                 assert_equal(tostr(b[1]), tostr(val))
 
                 # test also with longer strings
-                b = np.array([val, tostr('xxxxxxxxxx')])
+                b = np.array([val, tostr('xxxxxxxxxx')], dtype=dtype)
                 assert_equal(tostr(b[0]), tostr(val))
-                b = np.array([tostr('xxxxxxxxxx'), val])
+                b = np.array([tostr('xxxxxxxxxx'), val], dtype=dtype)
                 assert_equal(tostr(b[1]), tostr(val))
 
     def test_string_truncation_ucs2(self):

numpy/ma/tests/test_mrecords.py

@@ -405,7 +405,7 @@ def test_fromarrays(self):
         for (f, l) in zip(('a', 'b', 'c'), (_a, _b, _c)):
             assert_equal(getattr(mrec, f)._mask, l._mask)
         # One record only
-        _x = ma.array([1, 1.1, 'one'], mask=[1, 0, 0],)
+        _x = ma.array([1, 1.1, 'one'], mask=[1, 0, 0], dtype=object)
         assert_equal_records(fromarrays(_x, dtype=mrec.dtype), mrec[0])
 
     def test_fromrecords(self):