re_init_rng is not considering axons & it is calling the legacy/compatibility hoc function for SONATA cells

[anilbey]

Hello, there seems to be potentially 2 issues here.

This python function below

neurodamus/neurodamus/metype.py

Line 214 in 4b8fd79

Nd.rng123ForStochKvInit(self._ccell)

is calling this hoc function below although that is annotated as legacy. If the cell has the re_init_rng attribute - which is true for the some (maybe even all) SONATA cells (bluepyopt generates the hocs with re_init_rng function in it).
https://github.com/BlueBrain/neurodamus/blob/publication/nbS1-2023/core/hoc/RNGSettings.hoc#L155

the hoc code is iterating over somatic, basal and apical but not axon. AFAIK some axons may also have StochKvs. (cc @wvangeit)
There is also myelin, though I am not sure if it should be taken into account here.

If this is legacy, shouldn't this be disabled for SONATA cells? (bluecellulab disables it by default).
If it is to be enabled, the axons containing stochkv are currently missed.

[ferdonline]

The logic around the conditions for calling re_init_rng precede me but I can mention that, as a general rule, we don't instantiate axonal sections, hence this is likely the desired behavior. @jamesgking anything we might be misssing?

[wvangeit]

@ferdonline I'm not sure I understand the "we don't instantiate axonal sections". Did something change recently?

[ferdonline]

Not recently. In simulation, I believe for performance reasons, the axon is reduced to a stub one with only two sections. Still, I do believe it gets the right mechanisms. Should the RNG then still consider them?