Fix chained pofs or sdfs in model

[adamkewley]

Fixes an issue found during #3694 where chaining PhysicalOffsetFrames or StationDefinedFrames does not work.

The reason it doesn't work is because the previous implementation does things like dynamic_casting to PhysicalOffsetFrame followed by hard-setting its parent's MobilizedBodyIndex. That approach doesn't take into account that (e.g.) the parent may, itself, be a PhysicalOffsetFrame (a correct implementation should also iterate through the whole chain). It also doesn't account for other dependent frame types (e.g. StationDefinedFrame).

To work around this, the following changes were necessary:

• Modify PhysicalOffsetFrame's extendAddToSystem to instead set its own body index based on the base frame it's attached to (previously, it would use getParentFrame, which doesn't account for chains of frames)
• Modify PhysicalOffsetFrame with a new extension member method, extendSetMobilizedBodyIndex:
  • Lets downstream Frame implementations transitively set body indices based on what they know about their frame topology (e.g. a PhysicalOffsetFrame "knows" it has a parent)
• Remove dynamic_casting to PhysicalOffsetFrame from joint:
  • The joint now just calls setMobilizedBodyIndex
  • Which has its old behavior
  • But PhysicalOffsetFrame/StationDefinedFrame now use the extension point to (e.g.) ensure that the MobilizedBodyIndex is set on frames that they are transitively attached to
  • This new approach is automatically recursive (e.g. calling it on a chain of PoFs will call setMobilizedBodyIndex on each element in the chain, rather than just the parent)

API CHANGES (NONE)

setMobilizedBodyIndexis nowpublic, rather thanprotected` (it is still labelled as advanced, though).

A new method was added, setMobilizedBodyIndexOf, which keeps the existing API backwards-compatible but lets frames set eachovers indices (required)

Testing I've completed

• Added a unit test that exercises the problem (that you can't chain POFs together when joints are involved). The previous PoF-chaining tests don't include joints
• The tests failed
• I wrote this proposed fix
• The new test passed, but a "known bug" test in StationDefinedFrame then failed
• The StationDefinedFrame tests were re-enabled/reversed (as documented)
• All tests now pass

CHANGELOG.md (choose one)

• no need to update because it's a patch to a recent feature, broken out into this PR so it can be treated in isolation

---

This change is 

[adamkewley]

Downstream PR of opensim-creator with this change, for basic sanity checks:

• https://github.com/ComputationalBiomechanicsLab/opensim-creator/pull/823

Binary builds available via Actions

[adamkewley]

StationDefinedFrames_Advanced.osim is a much more comprehensive test of how chains of PhysicalOffsetFrames and StationDefinedFrames can interplay:

Not obvious from this screenshot, but the topology here is something like:

• ground <-- pof
• pof <-- pof2
• pof2 <-- FreeJoint
• FreeJoint --> StationDefinedFrame
• StationDefinedFrame defined via stations on 3 RGB spheres as stationX --> mesh_pof --> body`
• StationDefinedFrame <-- FreeJoint2
• FreeJoint2 --> StationDefinedFrame2
• StationDefinedFrame2 defined via a mixture of stations defined on RG cubes defined via chains of PoFs onto a cylnder model and a station that's directly defined on the body frame

Little bit pathological, but the idea is that users might be combining mesh data, landmarks, etc. to build frames dynamically, so the engine needs to handle mixed-, chained-, etc. frames.

[adamkewley]

The implementation was updated to mean that there's now no user-facing API changes at all: what was protected is still protected. The only changes are additions extendSetMobilizedBodyIndex and setMobilizedBodyIndexOf(someotherframe), which are the bare-minimum necessary to ensure that downstream PhysicalFrame implementations can customize their own index behavior and also set indices on other frames when it makes sense.

[adamkewley]

Thanks @pepbos