Boolean is_active variable not handled correctly in Python code

[brian-team]

Boolean is_active variable not handled correctly in Python code

[thesamovar]

This is a bit weird - shouldn't is_active be allocated once and then assigned to by is_active[:] = ...?

[mstimberg]

This is a bit weird - shouldn't is_active be allocated once and then assigned to by is_active[:] = ...?

I guess this is a consequence of my previous optimisation... The is_active calculation is now contained in the abstract code of the state update, so the state updater abstract code looks something like:

is_active = (t >= refractory_until)
_x = is_active*dt*(-x/tau) + x
x = _x
_array_x[:] = x
_array_is_active[:] = is_active

[thesamovar]

Ah yes, I see. OK, so that's unavoidable indeed. Well fixed in that case! ;)

[mstimberg]

I'll reopen this issue since I ran into it again... For variables like not_refractory we can mostly work around this problem, but we need a general solution. An example use case that currently fails (in Python): The user defines a boolean parameter channel_active in the model and uses this as part of the equations. During the equation rewriting done by sympy for the state update step, the expression -channel_active might appear which will not work as expected. I think we need to fix this in the numpy language by always loading boolean variables as 1 * variable.

[mstimberg]

I've dealt with this by explicitly introducing an int_ function (it was easier not calling it int, we can change this later if we want to) that should be used whenever a boolean variable or expression has to be interpreted in a 0/1 sense (i.e. in particular in model equations). The refractory variable is used in the correct way now.