New Agent getID outside of an init function

[ptheywood]

The current documentation for host functions & conditions > Agent Creation states that:

newAgent() returns an instance of HostNewAgentAPI, this can be used like other objects to set and get a new agent’s variables via setVariable() and getVariable(). Additionally, getID() can be used to retrieve the ID which will be assigned to the new agent.

I.e. this is what you will find if you want to find how to get the ID of a new agent.

If instead of doing this in an init function, but instead doing it outside of / prior to the Simulation existing, on a flamegpu::AgentVector_Agent, the `getID() method exists, but always returns 0 (because it has not been set yet, and can't be until it is associated with the Simulation).

This could either be documented, or possible an exception thrown if this method is used prior to the instance having it's actual ID set?

[Robadob]

Should be documented as exception. I don't think it's possible to assume agent IDs outside of the CUDASimulation object, as this could lead to an ID collision (which would throw an exception), particularly given agent state vectors can be combined etc.

The exception to that rule is agent state input files, which may contain IDs but purge all agent state on load (iirc).

[ptheywood]

It didn't throw when I did this unintentionally (started with init func, but switched to pre-init to avoid having to pass too many objects into an init func via file scoped variables), all agents just returned 0.

I agree it should throw as it can't be accurate, and be documented as throwing, plus probably mentioning in the user guide too.

[Robadob]

It didn't throw when I did this unintentionally (started with init func, but switched to pre-init to avoid having to pass too many objects into an init func via file scoped variables), all agents just returned 0.

https://github.com/FLAMEGPU/FLAMEGPU2/blob/master/src/flamegpu/simulation/detail/CUDAAgent.cu#L179

Well that seems like a bug, when agents are passed to CUDAAgent, validation is (supposed to be) performed, which would be redundant if we were force zeroing them.