Workaround for optimization failure

theano/tensor/opt.py

@@ -2235,8 +2235,20 @@ def local_reshape_chain(node):
         return False
 
     # TODO: this can permit a failing program to run by eliminating
-    #       the the lower reshape
-    return [node.op(node.inputs[0].owner.inputs[0], node.inputs[1])]
+    #       the lower reshape
+    rval = node.op(node.inputs[0].owner.inputs[0], node.inputs[1])
+    # It might happen that the desired output of this node has a broadcastable
+    # pattern that does not match that of 'rval'. This is when originally, we
+    # were able to figure out that one of the dimensions of the reshape is one,
+    # but some other transformation replaced the shape by one for which this
+    # cannot be guessed.
+    # We should try to figure out why we lost the information about this
+    # constant value... but in the meantime, better not apply this
+    # optimization.
+    if rval.broadcastable == node.outputs[0].broadcastable:
+        return [rval]
+    else:
+        return False
 register_canonicalize(local_reshape_chain)
 
 if 0: