Duplicate decision logic in a node: Decision tree optimisation?

[m0rjc]

The numbers nodes produce multiple transitions which share the 0 to 9 condition. It would be nice to be able to render nodes with such common code as decision trees so reducing the duplication of code.

This would make the model a little more complex, or make the assembly builder a little more complex, depending on where the optimisation is done.

Cutting out the 0-9 test would save 6 instructions in each Numbers node.

The Numbers Nodes also duplicate commands, though optimising these would require re-ordering of the logic with care of side effects so would be challenging to automate! manual re-writing of these nodes, or generation of custom logic using a special subclass of Node, could save some space on the microcontroller.

For example, at time of writing, see "step_gps_State_0015" in M0RJCTracker's gps.asm

[m0rjc]

Other ideas I'm looking at would involve creating a Node that performs its transition logic immediately on entry. The "Shared Entry Code" feature makes this possible, and the Builder is already capable of outputting the code. This feature, even if not exposed to the user, could be used by the Numbers Node to optimise out the duplicate logic.

[m0rjc]

The work to implement subroutines gives a good foundation for this. All that is needed is improvement of the numbers node to take advantage of it. This can be deferred until I know I have a fully working system.