Add ComponentPath overrides to Component interface

[adamkewley]

[perf-win]

… and integrate into Millard muscles and scalar actuators

Brief summary of changes

• Adds ComponentPath overrides for path-dependent parts of Component's public interface
• Changes Millard muscles, and scalar actuators to use ComponentPath, rather than std::string, so that the path is pre-parsed ahead of time, which can improve performance
• Performance numbers (i.e. biggest benefit when used with models that need to retrieve state variables this way):

Test Name	lhs [secs]	σ [secs]	rhs [secs]	σ [secs]	Speedup
ToyDropLanding_nomuscles	0.04	0.00	0.04	0.00	1.07
RajagopalModel	3.69	0.00	3.45	0.00	1.07
Gait2354	0.19	0.00	0.19	0.00	1.02
passive_dynamic_noanalysis	1.21	0.00	1.21	0.00	1.00
passive_dynamic	1.81	0.00	1.81	0.00	1.00
Arm26	0.16	0.00	0.16	0.00	1.02
ToyDropLanding	0.04	0.00	0.04	0.00	1.01

Testing I've completed

• Existing test suite
• Performance testing

Looking for feedback on...

• Whether it makes sense (that there's a performance upside to not constructing an OpenSim::CompnentPath to traverse a string when it's just the name of the value)

CHANGELOG.md (choose one)

• Updated

Performance analysis

Platform: Windows, self-hosted runner

Test Name	lhs [secs]	stderr [secs]	rhs [secs]	stderr [secs]	Speedup
Arm26	0.37	0.00	0.34	0.00	1.11
ellipsoid_wrap	4.10	0.01	4.10	0.00	1.00
ellipsoid_wrap_function_based_paths	3.44	0.01	3.45	0.01	1.00
Gait2354	0.41	0.00	0.42	0.00	0.97
MocoSlidingMass	1.48	0.01	1.45	0.01	1.02
MocoSquatToStand	14.19	0.09	13.47	0.04	1.05
passive_dynamic	5.67	0.02	5.60	0.00	1.01
passive_dynamic_noanalysis	3.42	0.01	3.43	0.00	1.00
RajagopalModel	8.89	0.02	8.19	0.00	1.09
ToyDropLanding	13.45	0.01	12.31	0.06	1.09
ToyDropLanding_fbp_stepwisereg	12.32	0.01	11.32	0.02	1.09
ToyDropLanding_function_based_paths	12.65	0.09	11.42	0.03	1.11
ToyDropLanding_nomuscles	0.57	0.00	0.58	0.00	0.99

---

This change is 

[adamkewley]

Speedup appears to be roughly 5 % on Rajagopal, which uses MillardEquilibriumMuscle:

Test Name	lhs [secs]	σ [secs]	rhs [secs]	σ [secs]	Speedup
ToyDropLanding_nomuscles	0.04	0.00	0.04	0.00	1.11
RajagopalModel	3.82	0.00	3.59	0.00	1.06
Gait2354	0.20	0.00	0.19	0.00	1.03
passive_dynamic_noanalysis	1.22	0.00	1.23	0.00	1.00
passive_dynamic	1.83	0.00	1.82	0.00	1.00
Arm26	0.17	0.00	0.16	0.00	1.03
ToyDropLanding	0.04	0.00	0.04	0.00	1.02

[adamkewley]

New approach: check if it's a path/name at runtime

Doing it this way means that no downstream (from OpenSim::Component) code needs to change - for the cost of having to check if the string contains a directory separator.

If the string check is too expensive then we'd need to add overrides like getStateVariableValueByName so that code can explicitly state that the thing being provided is a name, not a path. However, I have avoided doing this here because it makes the overall API more confusing (some methods would be byName others would be suffix-less, but implicitly require a name (e.g. getOutput), and so on).

[adamkewley]

Test Name	lhs [secs]	σ [secs]	rhs [secs]	σ [secs]	Speedup
ToyDropLanding_nomuscles	0.04	0.00	0.04	0.00	1.07
RajagopalModel	3.69	0.00	3.45	0.00	1.07
Gait2354	0.19	0.00	0.19	0.00	1.02
passive_dynamic_noanalysis	1.21	0.00	1.21	0.00	1.00
passive_dynamic	1.81	0.00	1.81	0.00	1.00
Arm26	0.16	0.00	0.16	0.00	1.02
ToyDropLanding	0.04	0.00	0.04	0.00	1.01

[adamkewley]

If you get a chance @nickbianco / @pepbos then this is ready for review.

It's mostly an isolated shortcut that handles the situation where (e.g.) getStateVariableValue is called with something like "somevar". The previous implementation assumes that all incoming strings might be paths to other components, so it spends a fair bit of CPU time (5-10 % when using Rajagopal model) converting to a ComponentPath, checking how many elements there are in the path, etc.

Instead, this version just checks if the path contains a ComponentPath::separator() (added in this PR - it's why the diffs are bigger than necessary, but I was getting sick of seeing "/" everywhere) and, if there's no separator, it does a direct lookup.

Note: the biggest gains for RajagopalModel are actually on getModelingInfo, which is called a lot more, and suffers from the same logic.

[nickbianco]

@adamkewley, I should have clarified: the perf suite will run on the next commit after adding [perf-win].