Conditional string indexing of Synapses objects raises uninformative KeyError

[denisalevi]

Hi there!

I just adapted the test_synapses_to_synapses.py test to find out if I need to take care of some form of subgroups when the targat of a SynapticPathway is a Synapses object. Here is the adapted test:

from brian2 import *
set_device('cpp_standalone')

source = SpikeGeneratorGroup(3, [0, 1, 2], [0, 0, 0]*ms, period=2*ms)
modulator = SpikeGeneratorGroup(3, [0, 2], [1, 3]*ms)
target = NeuronGroup(3, 'v : integer')
conn = Synapses(source, target, 'w : integer', on_pre='v += w')
conn.connect(j='i')
conn.w = 1
# XXX: I added the `['i > 0']` index
modulatory_conn = Synapses(modulator, conn['i > 0'], on_pre='w += 1')
modulatory_conn.connect(j='i')
run(5*ms)
# First group has its weight increased to 2 after the first spike
# Third group has its weight increased to 2 after the second spike
assert_array_equal(target.v, [5, 3, 4])

This fails with

KeyError: 'No template with name "group_get_indices" found.'

All I did compared to the original test is adding a ['i > 0'] index to the target Synapses object in the modulatory synapse:

modulatory_conn = Synapses(modulator, conn['i > 0'], on_pre='w += 1')

1. Should this be possible at all? Creating a Synapses subgroup by string indexing? It seems like this could be implemented (something like a simple version of the synapses_create_generator template that checks the condition for each created synapse). But I don't think it is, right? If not, this should probably raise a better error message?
2. While here with this test: Is there any other way that modulatory_conn could target a subgroup of conn in the line above. Or in other words, is it possible to create a subgroup of a Synapses object before running the network in standalone mode? If not, this would make the brian2cuda implementation much easier.
3. And to make sure I didn't miss anything: SynapsesPahtway objects have to have a spiking NeuronGroup as their source and having Synapses as source of a SynapticPathway object is not possible since Synapses can't define spikes / events on which the pathway is activated. Correct?

[denisalevi]

I just looked into the astrocyte examples and my question 3 is answered: Synapses object can be sources of SynapticPathways. OK.