Mechanism nodes created by the EMComposition json_summary issue

[ddelpiano]

As per title, we noticed that some of the nodes created as boilerplate by the EMCompositon have some problems in the json summary, in details it seems that the input ports are not documented correctly in the json summary and instead these are projections linked to the node itself.

An example:

checking the node (ProcessingMechanism PREVIOUS_STATE [QUERY]) we can see that the json summary, at the input ports key has the following:

'PREVIOUS_STATE__QUERY__input_port__INPUT_CIM_PREVIOUS_STATE__QUERY__InputPort_0__to__PREVIOUS_STATE__QUERY__InputPort_0_'

and

'PREVIOUS_STATE__QUERY__input_port_MappingProjection_from_PREVIOUS_STATE_OutputPort_0__to_PREVIOUS_STATE__QUERY__InputPort_0_'

which, looking into the data does not seems to be input ports, also confirmed by checking node.input_ports which return the following:

[(InputPort InputPort-0)]

Follow the model used:
https://github.com/MetaCell/PsyNeuLinkView/tree/main/test_models/CSW

Thanks

[kmantel]

I wasn't able to run the full script there with current PNL versions, but it got far enough to see this.

The PNL and MDF/json "input port"s are slightly different. PNL InputPorts can accept multiple Projections, but MDF input ports can only accept one. So if there are multiple Projections to a PNL InputPort, we create one MDF input port for each, and then add a combination function to the MDF node. PNL InputPort handles this combination itself. So those two you're seeing correspond to InputPort-0 in this example.

I see how this is confusing because it wasn't obvious to me what was happening just by reading the port names

Will provide more info

[kmantel]

The only other thing I can say is that when one PNL InputPort corresponds to more than one MDF input port, the MDF input ports will be named by combining the Mechanism and Projection's names, and they should be in the same order as PNL InputPorts

But let me know if you need more info