Add assert about the shape in the graph.

theano/tensor/opt.py

@@ -370,10 +370,16 @@ def local_0_dot_x(node):
     if replace:
         constant_zero = T.constant(0, dtype=node.outputs[0].type.dtype)
         if x.ndim == 2 and y.ndim == 2:
+            constant_zero = assert_(constant_zero,
+                                    T.eq(x.shape[1], y.shape[0]))
             return [T.alloc(constant_zero, x.shape[0], y.shape[1])]
         elif x.ndim == 1 and y.ndim == 2:
+            constant_zero = assert_(constant_zero,
+                                    T.eq(x.shape[0], y.shape[0]))
             return [T.alloc(constant_zero, y.shape[1])]
         elif x.ndim == 2 and y.ndim == 1:
+            constant_zero = assert_(constant_zero,
+                                    T.eq(x.shape[1], y.shape[1]))
             return [T.alloc(constant_zero, x.shape[0])]
         elif x.ndim == 1 and y.ndim == 1:
             return [constant_zero]

theano/tensor/tests/test_opt.py

@@ -2395,6 +2395,16 @@ class Test_alloc_zero(unittest.TestCase):
                     assert numpy.all([ not isinstance(x.op, tensor.Dot) for x in
                                       f.maker.fgraph.toposort() ])
 
+    def test_dot_allocs_0_err(self):
+        #test that we don't remove errors
+        m1 = tensor.matrix('m1')
+        vm = numpy.asarray([[1, 2, 3], [4, 5, 6]],
+                           dtype=theano.config.floatX)
+        z = numpy.zeros((3, 3), dtype=theano.config.floatX)
+        o = tensor.dot(z, m1)
+        f = theano.function([m1], o, mode=self.mode)
+        self.assertRaises((ValueError, AssertionError), f, vm)
+
 
 def test_local_subtensor_of_alloc():