Fix the inconsistent integral property in np_sum (#71)

* Fix the integral inconsistency in np_sum
* Add integral property tests for numpy functions min, max, amin, amax, argmin, argmax, sgn, absolute
* Solves pycodestyle warning in the tests
* Use a style with fewer awaits.
---------
Co-authored-by: lschoe <l.a.m.schoenmakers@tue.nl>

mpyc/runtime.py

@@ -2909,9 +2909,16 @@ async def np_sum(self, a, axis=None, keepdims=False, initial=0):
             else:
                 shape = tuple(s for i, s in enumerate(a.shape) if i not in axes)
         if shape == ():
-            await self.returnType(sectype)
+            if isinstance(a, self.SecureFixedPointArray):
+                rettype = (sectype, a.integral)
+            else:
+                rettype = sectype
         else:
-            await self.returnType((type(a), shape))
+            if isinstance(a, self.SecureFixedPointArray):
+                rettype = (type(a), a.integral, shape)
+            else:
+                rettype = (type(a), shape)
+        await self.returnType(rettype)
         a, initial = await self.gather(a, initial)
         return np.sum(a, axis=axis, keepdims=keepdims, initial=initial.value)
         # TODO: handle switch from initial (field elt) to initial.value inside finfields.py

tests/test_runtime.py

@@ -325,6 +325,41 @@ def test_secfxp_array(self):
         self.assertEqual(np.vstack((c1, c1)).integral, False)
         self.assertEqual(np.vstack((c2, c2)).integral, True)
 
+        self.assertEqual(np.sum(c2).integral, True)
+        self.assertEqual(np.sum(c2, axis=0).integral, True)
+        self.assertEqual(np.sum(c1).integral, False)
+        self.assertEqual(np.sum(c1, axis=0).integral, False)
+
+        self.assertEqual(mpc.np_sgn(c1).integral, True)
+        self.assertEqual(mpc.np_sgn(c2).integral, True)
+        self.assertEqual(np.absolute(c1).integral, False)
+        self.assertEqual(np.absolute(c2).integral, True)
+
+        self.assertEqual(np.minimum(c1, c2).integral, False)
+        self.assertEqual(np.minimum(c1, c1).integral, False)
+        self.assertEqual(np.minimum(c2, c2).integral, True)
+        self.assertEqual(np.maximum(c1, c2).integral, False)
+        self.assertEqual(np.maximum(c1, c1).integral, False)
+        self.assertEqual(np.maximum(c2, c2).integral, True)
+
+        self.assertEqual(np.amin(c2).integral, True)
+        self.assertEqual(np.amin(c2, axis=0).integral, True)
+        self.assertEqual(np.amin(c1).integral, False)
+        self.assertEqual(np.amin(c1, axis=0).integral, False)
+        self.assertEqual(np.amax(c2).integral, True)
+        self.assertEqual(np.amax(c2, axis=0).integral, True)
+        self.assertEqual(np.amax(c1).integral, False)
+        self.assertEqual(np.amax(c1, axis=0).integral, False)
+
+        self.assertEqual(np.argmin(c2).integral, True)
+        self.assertEqual(np.argmin(c2, axis=0).integral, True)
+        self.assertEqual(np.argmin(c1).integral, True)
+        self.assertEqual(np.argmin(c1, axis=0).integral, True)
+        self.assertEqual(np.argmax(c2).integral, True)
+        self.assertEqual(np.argmax(c2, axis=0).integral, True)
+        self.assertEqual(np.argmax(c1).integral, True)
+        self.assertEqual(np.argmax(c1, axis=0).integral, True)
+
     @unittest.skipIf(not np, 'NumPy not available or inside MPyC disabled')
     def test_secfld_array(self):
         np.assertEqual = np.testing.assert_array_equal