theano_test_value_size.patch

diff --git a/theano/configdefaults.py b/theano/configdefaults.py
index 58ed2e9..97d6564 100644
--- a/theano/configdefaults.py
+++ b/theano/configdefaults.py
@@ -470,6 +470,14 @@ AddConfigVar('compute_test_value',
         EnumStr('off', 'ignore', 'warn', 'raise', 'pdb'),
         in_c_key=False)
 
+AddConfigVar('store_test_value_maxsize',
+        ("Maximum size for test values that are kept.  If compute_test_value "
+         "is enabled, keeps test values smaller than the given size (in "
+         "number of entries).  Beyond that, only the shape is stored; a "
+         "an array with the same shape and type is created on demand, filled "
+         "with a single random entry from the array."),
+        IntParam(sys.maxint),
+        in_c_key=False)
 
 AddConfigVar('compute_test_value_opt',
              ("For debugging Theano optimization only."
diff --git a/theano/gof/op.py b/theano/gof/op.py
index ac85eec..a306077 100644
--- a/theano/gof/op.py
+++ b/theano/gof/op.py
@@ -18,6 +18,7 @@ import numpy
 import os
 import sys
 import warnings
+import numpy
 
 import theano
 from theano import config
@@ -461,6 +462,10 @@ class PureOp(object):
         elif isinstance(v, graph.Variable) and hasattr(v.tag, 'test_value'):
             # ensure that the test value is correct
             return v.type.filter(v.tag.test_value)
+        elif isinstance(v, graph.Variable) and hasattr(v.tag, 'test_shape'):
+            test_value = numpy.empty(v.tag.test_shape, dtype=v.type.dtype)
+            test_value.fill(v.tag.test_value_fill)
+            return v.type.filter(test_value, strict=False, allow_downcast=True)
 
         raise AttributeError('%s has no test value' % v)
 
@@ -552,7 +557,14 @@ class PureOp(object):
 
                     # add 'test_value' to output tag, so that downstream ops can use these
                     # numerical values as inputs to their perform method.
-                    output.tag.test_value = storage_map[output][0]
+                    test_value = storage_map[output][0]
+                    if not hasattr(test_value, 'size') or \
+                       test_value.size < config.store_test_value_maxsize:
+                        output.tag.test_value = test_value
+                    elif hasattr(test_value, 'shape'):
+                        test_value = numpy.asarray(test_value)
+                        output.tag.test_shape = test_value.shape
+                        output.tag.test_value_fill = test_value.flat[0]
 
         if self.default_output is not None:
             rval = node.outputs[self.default_output]