Instantiate Models in Matlab causes vtp warning

[jimmyDunne]

In Matlab, when creating a new model and assigning display geometry, the OpenSim API must not know where the Geometry folder is and prints warnings to the console. The easiest way to see this is by running OpenSimCreateCreateTugOfWarModel.m

The Model Geometry is viewable when opened in the GUI, however this should be considered an isuue since;

1. If I try to view a simulation with this model in Matlab, the display geometry doesn't show.
2. The warning gets printed to console at every addComponent (addBody(), addForce(), addController()) and can be (is) distracting.

I tried adding the Geometry folder path to Matlab path and my terminal path to no avail ☹️

[chrisdembia]

We should emit those warnings only once, and we might be able to read the GUI's Geometry Path preference using Windows' registry.

For now, you should be able to set the OPENSIM_HOME environment variable to a directory that contains a Geometry folder (that is not the geometry folder itself, but one level up).

[aymanhab]

Alternatively you can do what the GUI does which is call ModelVisualizer.addDirToGeometrySearchPaths(path)

[aymanhab]

@jimmyDunne For first example e.g. tug-of-war, we probably don't want to use vtp files and have users deal with this issue altogether. The model has built-in shapes mostly. It would be better to use Bricks and Spheres rather than vtp files.

[jimmyDunne]

Sure— but doesn't change that this needs will need to be altered. Users are already (rightly) complaining (forum)

[clnsmith]

I'm also running into this in C++ code as well. It is annoying since it throws the warnings everytime you call initSystem.

Another option is to add a Property to the Model (and thus .osim file) that is GeometrySearchPath. Currently when I am developing a new model, I often have different model versions in different folders. Right now I have to duplicate the "Geometry" folder into each of these model version folders. The other option is to use full system paths to the geometry files in the model file, but this is annoying if you are using multiple computers/sharing models. I guess you would still need to change the GeometrySearchPath, so maybe its not that great of an idea, but atleast that is only one line.

[chrisdembia]

Good idea. We could also consider allowing some sort of per-model environment variable or config file in the user's home directory.

We should also emit the warning only once.

Does using model.updVisualizer().addDirToGeometrySearchPaths() help for now?

[aymanhab]

Thanks for reporting @clnsmith I think the right fix is to have a clear "Stage" where some checks are made at this Stage e.g. deserialization but not all the time (e.g. finalizeXXXXX). The essence of the problem as I see it is that we have a big :hammer: and we use it all the time because there's no granularity in the tricky initialization sequence. My 2 cents.

[jimmyDunne]

@aseth1 suggested adding a method that disables reading in geometry.

Visualizing the geometry could be set off by default and the Visualize() could set the model geometry to true

[aymanhab]

My investigation so far:

• The attempted loading of the meshes happens during construction (Mesh::extendFinalizeFromProperties invoked from componentsFinalizeFromProperties ) so we have no place to control this loading unless we pass a flag to the constructor, or include it as a property in the file which could lead to more confusion downstream. Would like to hear your thoughts @jimmyDunne or mention anybody else who would be interested. Thanks

[jimmyDunne]

I think everyone is in agreement that we should find a way to stop these warnings, so whatever the best system is for dealing with that would be good. If it is to add properties, then we should explore that.