Adapt unit test to test both paths of fix for direct outputs

theano/scan_module/tests/test_scan.py

@@ -2526,27 +2526,35 @@ class T_Scan(unittest.TestCase):
 
         x = tensor.scalar()
         seq = tensor.vector()
-        out, updates = theano.scan(lambda a, b : a + b, sequences=seq,
-                                   outputs_info=x)
-
-        # Obtain a reference to the scan output before the subtensor and
-        # compile a function with it as output
-        assert isinstance(out.owner.op, tensor.subtensor.Subtensor)
-        out = out.owner.inputs[0]
-        fct = theano.function([x, seq], out[:-1])
+        outputs_info=[x, tensor.zeros_like(x)]
+        (out1, out2), updates = theano.scan(lambda a, b, c : (a + b, b + c),
+                                            sequences=seq,
+                                            outputs_info=outputs_info)
+
+        # Obtain a reference to the scan outputs before the subtensor and
+        # compile a function with them as outputs
+        assert isinstance(out1.owner.op, tensor.subtensor.Subtensor)
+        assert isinstance(out2.owner.op, tensor.subtensor.Subtensor)
+        out1_direct = out1.owner.inputs[0]
+        out2_direct = out2.owner.inputs[0]
+        fct = theano.function([x, seq], [out1_direct[:-1], out2_direct[:-1]])
 
         # Test the function to ensure valid outputs
         floatX = theano.config.floatX
 
         init_value = 5.0
         seq_value = numpy.arange(4, dtype=floatX)
-        output = fct(init_value, seq_value)
+        output1, output2 = fct(init_value, seq_value)
 
-        expected_output = [init_value]
+        expected_output1 = [init_value]
+        expected_output2 = [0]
         for i in seq_value[:-1]:
-            expected_output.append(expected_output[-1] + i)
+            expected_output2.append(expected_output1[-1] +
+                                    expected_output2[-1])
+            expected_output1.append(expected_output1[-1] + i)
 
-        utt.assert_allclose(output, expected_output)
+        utt.assert_allclose(output1, expected_output1)
+        utt.assert_allclose(output2, expected_output2)
 
     def test_infer_shape(self):
         # Test for a crash in scan.infer_shape when using both