Fix an error in the synapse generation with numpy, if "int" is used a…

…s part of the condition

Fixes #835

brian2/codegen/runtime/numpy_rt/templates/synapses_create_generator.py_

@@ -52,7 +52,7 @@ except:
                 # however it can't therefore give a jump size of 0, but we could have rand()<p for
                 # for the first array. Therefore we set start to the actual first place we'd like
                 # to see a hit minus one, allowing us to have start as the first hit.
-                jump = _numpy.array(_numpy.ceil(_numpy.log(_np_rand(m))*pconst), dtype=int)
+                jump = _numpy.array(_numpy.ceil(_numpy.log(_np_rand(m))*pconst), dtype=_numpy.int32)
                 jump[0] += start-1
                 samples = _numpy.cumsum(jump)
                 all_samples.append(samples)

brian2/tests/test_synapses.py

@@ -312,6 +312,16 @@ def test_connection_random_with_condition():
     S17 = Synapses(G, G, 'w:1', 'v+=w')
     S17.connect('i!=j', p='j*0.1')
 
+    # Forces the use of the "jump algorithm"
+    big_group = NeuronGroup(10000, 'v: 1', threshold='False')
+    S18 = Synapses(big_group, big_group, 'w:1', 'v+=w')
+    S18.connect('i != j', p=0.001)
+
+    # See github issue #835 -- this failed when using numpy
+    S19 = Synapses(big_group, big_group, 'w:1', 'v+=w')
+    S19.connect('i < int(N_post*0.5)', p=0.001)
+
+
     with catch_logs() as _:  # Ignore warnings about empty synapses
         run(0*ms)  # for standalone