get rid of dangerous "TypeError = not constant" mechanism

theano/__init__.py

@@ -171,7 +171,8 @@ def get_constant_value(v):
 
     If theano.sparse is also there, we will look over CSM op.
 
-    If `v` is not some view of constant data, then raise a TypeError.
+    If `v` is not some view of constant data, then raise a
+    tensor.basic.NotConstantError.
     """
     if hasattr(theano, 'sparse') and isinstance(v.type,
                                                 theano.sparse.SparseType):

theano/gradient.py

@@ -1618,7 +1618,7 @@ def _is_zero(x):
     try:
         constant_value = theano.get_constant_value(x)
         no_constant_value = False
-    except TypeError:
+    except theano.tensor.basic.NotConstantError:
         pass
 
     if no_constant_value:

theano/sandbox/linalg/ops.py

@@ -215,7 +215,7 @@ def is_positive(v):
     if v.owner and v.owner.op == tensor.pow:
         try:
             exponent = tensor.get_constant_value(v.owner.inputs[1])
-        except TypeError:
+        except tensor.basic.NotConstantError:
             return False
         if 0 == exponent % 2:
             return True

theano/sandbox/scan.py

@@ -136,7 +136,7 @@ def scan(fn,
     else:
         try:
             n_fixed_steps = opt.get_constant_value(n_steps)
-        except (TypeError, AttributeError):
+        except tensor.basic.NotConstantError:
             n_fixed_steps = None
 
     # Check n_steps is an int

theano/scan_module/scan.py

@@ -364,7 +364,7 @@ def scan(fn,
     else:
         try:
             n_fixed_steps = opt.get_constant_value(n_steps)
-        except (TypeError, AttributeError):
+        except tensor.basic.NotConstantError:
             n_fixed_steps = None
 
     # Check n_steps is an int

theano/tensor/basic.py

@@ -463,17 +463,10 @@ def _allclose(a, b, rtol=None, atol=None):
     return numpy.allclose(a, b, atol=atol_, rtol=rtol_)
 
 
-class NotConstantError(TypeError):
+class NotConstantError(Exception):
     """
     Raised by get_constant_value if called on something that is
     not constant.
-    For now it is a TypeError, to maintain the old interface
-    that get_constant_value should raise a TypeError in this
-    situation. However, this is unsafe because get_constant_value
-    could inadvertently raise a TypeError if it has a bug.
-    So we should eventually make NotConstantError derive
-    from Exception directly, and modify all code that uses
-    get_constant_value to catch this more specific exception.
     """
     pass
 
@@ -2364,7 +2357,7 @@ class SpecifyShape(Op):
                 s = get_constant_value(node.inputs[1][dim])
                 s = as_tensor_variable(s)
                 new_shape.append(s)
-            except TypeError:
+            except NotConstantError:
                 new_shape.append(node.inputs[1][dim])
 
         assert len(new_shape) == len(xshape)
@@ -2662,7 +2655,7 @@ def max(x, axis=None, keepdims=False):
         try:
             const = get_constant_value(axis)
             out = CAReduce(scal.maximum, list(const))(x)
-        except Exception:
+        except NotConstantError:
             out = max_and_argmax(x, axis)[0]
 
     if keepdims:
@@ -3272,7 +3265,7 @@ class Alloc(gof.Op):
             # if s is constant 1, then we're broadcastable in that dim
             try:
                 const_shp = get_constant_value(s)
-            except TypeError:
+            except NotConstantError:
                 const_shp = None
             bcast.append(numpy.all(1 == const_shp))
         otype = TensorType(dtype=v.dtype, broadcastable=bcast)
@@ -3820,7 +3813,7 @@ def extract_constant(x):
     '''
     try:
         x = get_constant_value(x)
-    except Exception:
+    except NotConstantError:
         pass
     if (isinstance(x, scal.ScalarVariable) or
         isinstance(x, scal.sharedvar.ScalarSharedVariable)):
@@ -5423,7 +5416,7 @@ class Join(Op):
                     # int
                     axis = int(get_constant_value(axis))
 
-                except TypeError:
+                except NotConstantError:
                     pass
             if isinstance(axis, int):
                 # Basically, broadcastable -> length 1, but the
@@ -5792,7 +5785,7 @@ class Reshape(Op):
                 try:
                     bcasts[index] = (hasattr(y, 'get_constant_value') and
                                      y.get_constant_value() == 1)
-                except TypeError:
+                except NotConstantError:
                     pass
             return gof.Apply(self, [x, shp], [tensor(x.type.dtype, bcasts)])
 
@@ -5868,7 +5861,7 @@ class Reshape(Op):
                     os_i = get_constant_value(node.inputs[1][i]).item()
                     if os_i == -1:
                         os_i = default_os_i
-                except TypeError:
+                except NotConstantError:
                     os_i = default_os_i
                 oshape.append(os_i)
             return [tuple(oshape)]
@@ -6150,7 +6143,7 @@ class ARange(Op):
             try:
                 v = get_constant_value(var)
                 return numpy.all(v == value)
-            except Exception:
+            except NotConstantError:
                 pass
             return False
 

theano/tensor/nnet/sigm.py

@@ -138,7 +138,7 @@ def _is_1(expr):
     try:
         v = opt.get_constant_value(expr)
         return numpy.allclose(v, 1)
-    except TypeError:
+    except tensor.NotConstantError:
         return False
 
 log1msigm_to_softplus = gof.PatternSub(

theano/tensor/opt.py

@@ -33,7 +33,7 @@ from theano.gof.opt import (Optimizer, pre_constant_merge,
                             pre_greedy_local_optimizer)
 from theano.gof.opt import merge_optimizer
 from theano.gof import toolbox, DestroyHandler
-from basic import get_constant_value, ShapeError
+from basic import get_constant_value, ShapeError, NotConstantError
 
 
 theano.configparser.AddConfigVar('on_shape_error',
@@ -95,7 +95,7 @@ def scalarconsts_rest(inputs):
             v = get_constant_value(i)
             consts.append(v)
             origconsts.append(i)
-        except Exception:
+        except NotConstantError:
             nonconsts.append(i)
     return consts, origconsts, nonconsts
 
@@ -324,13 +324,13 @@ def local_0_dot_x(node):
     try:
         if get_constant_value(x) == 0:
             replace = True
-    except TypeError:
+    except NotConstantError:
         pass
 
     try:
         if get_constant_value(y) == 0:
             replace = True
-    except TypeError:
+    except NotConstantError:
         pass
 
     if replace:
@@ -1179,7 +1179,7 @@ def local_subtensor_make_vector(node):
                 try:
                     v = get_constant_value(idx)
                     return [x.owner.inputs[v]]
-                except Exception:
+                except NotConstantError:
                     pass
             else:
                 # it is a slice of ints and/or Variables
@@ -1321,7 +1321,7 @@ def local_remove_useless_assert(node):
                     #Should we raise an error here? How to be sure it
                     #is not catched?
                     cond.append(c)
-            except TypeError:
+            except NotConstantError:
                 cond.append(c)
 
         if len(cond) == 0:
@@ -1551,7 +1551,7 @@ def local_useless_subtensor(node):
             length_pos = shape_of[node.inputs[0]][pos]
             try:
                 length_pos_data = get_constant_value(length_pos)
-            except TypeError:
+            except NotConstantError:
                 pass
 
             if isinstance(idx.stop, int):
@@ -2034,7 +2034,7 @@ def local_incsubtensor_of_allocs(node):
         try:
             if get_constant_value(y) == 0:
                 replace = True
-        except TypeError:
+        except NotConstantError:
             pass
 
         if replace:
@@ -2060,12 +2060,12 @@ def local_setsubtensor_of_allocs(node):
 
         try:
             replace_x = get_constant_value(x)
-        except TypeError:
+        except NotConstantError:
             pass
 
         try:
             replace_y = get_constant_value(y)
-        except TypeError:
+        except NotConstantError:
             pass
 
         if (replace_x == replace_y and
@@ -2260,7 +2260,7 @@ def local_mul_switch_sink(node):
                     fct[0].values_eq_approx = fct[
                         0].type.values_eq_approx_remove_nan
                     return fct
-            except TypeError:
+            except NotConstantError:
                 pass
             try:
                 if get_constant_value(switch.inputs[2]) == 0.:
@@ -2270,7 +2270,7 @@ def local_mul_switch_sink(node):
                     fct[0].values_eq_approx = fct[
                         0].type.values_eq_approx_remove_nan
                     return fct
-            except TypeError:
+            except NotConstantError:
                 pass
     return False
 
@@ -2301,7 +2301,7 @@ def local_div_switch_sink(node):
                 fct[0].values_eq_approx = fct[
                     0].type.values_eq_approx_remove_nan
                 return fct
-        except TypeError:
+        except NotConstantError:
             pass
         try:
             if get_constant_value(switch.inputs[2]) == 0.:
@@ -2310,7 +2310,7 @@ def local_div_switch_sink(node):
                 fct[0].values_eq_approx = fct[
                     0].type.values_eq_approx_remove_nan
                 return fct
-        except TypeError:
+        except NotConstantError:
             pass
     return False
 
@@ -2703,7 +2703,7 @@ class Canonizer(gof.LocalOptimizer):
         if isinstance(v, Variable):
             try:
                 return get_constant_value(v)
-            except TypeError:
+            except NotConstantError:
                 return None
         else:
             return v
@@ -3208,7 +3208,7 @@ def local_sum_alloc(node):
                     assert val.size == 1
                     val = val.reshape(1)[0] * T.mul(*shapes)
                     return [T.cast(val, dtype=node.outputs[0].dtype)]
-                except TypeError:
+                except NotConstantError:
                     pass
             else:
                 try:
@@ -3222,7 +3222,7 @@ def local_sum_alloc(node):
                     return [T.alloc(T.cast(val, dtype=node.outputs[0].dtype),
                                     *[shapes[i] for i in xrange(len(shapes))
                                       if i not in node.op.axis])]
-                except TypeError:
+                except NotConstantError:
                     pass
 
 
@@ -3283,7 +3283,7 @@ def local_mul_zero(node):
         for i in node.inputs:
             try:
                 value = get_constant_value(i)
-            except TypeError:
+            except NotConstantError:
                 continue
             #print 'MUL by value', value, node.inputs
             if N.all(value == 0):
@@ -3521,7 +3521,7 @@ def local_add_specialize(node):
         for input in node.inputs:
             try:
                 y = get_constant_value(input)
-            except TypeError:
+            except NotConstantError:
                 y = input
             if numpy.all(y == 0.0):
                 continue
@@ -3882,7 +3882,7 @@ def _is_1(expr):
     try:
         v = get_constant_value(expr)
         return numpy.allclose(v, 1)
-    except TypeError:
+    except NotConstantError:
         return False
 
 
@@ -3892,7 +3892,7 @@ def _is_minus1(expr):
     try:
         v = get_constant_value(expr)
         return numpy.allclose(v, -1)
-    except TypeError:
+    except NotConstantError:
         return False
 
 #1+erf(x)=>erfc(-x)
@@ -4133,7 +4133,7 @@ def local_grad_log_erfc_neg(node):
 
         try:
             cst2 = get_constant_value(mul_neg.owner.inputs[0])
-        except TypeError:
+        except NotConstantError:
             return False
 
         if len(mul_neg.owner.inputs) == 2:
@@ -4160,7 +4160,7 @@ def local_grad_log_erfc_neg(node):
             x = erfc_x
             try:
                 cst = get_constant_value(erfc_x.owner.inputs[0])
-            except TypeError:
+            except NotConstantError:
                 return False
             if cst2 != -cst * 2:
                 return False

theano/tensor/tests/test_basic.py

@@ -6176,7 +6176,7 @@ class T_get_constant_value(unittest.TestCase):
 
         b = tensor.iscalar()
         a = tensor.stack(b, 2, 3)
-        self.assertRaises(TypeError, get_constant_value, a[0])
+        self.assertRaises(tensor.basic.NotConstantError, get_constant_value, a[0])
         assert get_constant_value(a[1]) == 2
         assert get_constant_value(a[2]) == 3