Integrate new solver

[stfnp]

For the past year or so I have been working on a complete rewrite (and redesign) of the VirtualBow solver, i.e. the part that does the numerical computations. Unlike the old solver, the new one is written in Rust instead of C++ (see also #215). Even though the new solver is not yet finished, it has reached a state where it can be integrated into the main application in order to make testing and filling in missing features easier. This is currently happening on the branch integrate-new-solver (not usable yet).

This issue is for tracking the progress and the remaining tasks and will be updated frequently.

Remaining tasks for the integration

• Include Rust compilation in the main build process
• Make CI/CD pipeline work
• Run solver tests in the pipeline
• Call new solver from main application
• Finalize new solver's output format
• Load new solver output from the main application

Remaining tasks/missing functionality of the new solver

• Add back test test_load_output
• Evaluation and output of layer stresses
• Dynamic simulation, including output of correct arrow position
• Damping of limb and string
• Manually test new string damping to verify that it does something
• New string model without penalty contacts
• Split solver into two crates: virtualbow-lib and virtualbow-cli and add back integration tests for the command line interface.

More detailed solver issues

• Document test example of nonlinear pendulum
• Implement and document static test for the string model
• Implement test of linear beam dynamics
• Extend static tests to bows with recurves
• Extend static tests to cross section stresses and strains
• Extend dynamic tests (energy, basic properties)
• Complete last missing outputs
  • Minimum and maximum layer stresses
  • Kinetic energy of string and arrow
  • Limb velocity
  • String velocity
  • Characteristic dynamic numbers
  • Bonus: Force/draw stiffness
• Release string force gradually, add setting for release time
• String is not placed at the correct distance from the reference line
• Bracing angle incorrect for recurves, probably wrong measurement of string angle. Fix and check in tests.
• Reduce stiffness of the string in compression for dynamic analysis, document this in the theory manual
• Consistent use of either SVector<f64, N> or [f64; N] in the input and output structs
• Reference materials and layers by their name instead of their index
• Check for self-intersecting geometry (curvature too high for thickness)
• Progress estimation for dynamic simulation
• Check for/fix duplicate time point in the dynamic output
• Add new/changed numerical parameters to the bow settings
• Timestep estimation or control for the dynamic simulation
• Add missing documentation for eigenvalue solver
• Implement Error trait for all custom errors
• In solvers, check if states are finite and abort early if not
• Better method than trapezoidal rule for segment stiffness matrices?
• Add tests for cubic spline extrapolation
• Consider replacing soa_derive with https://github.com/tim-harding/soa-rs
• Proper error handling for computation of natural frequencies

GUI/Integration issues

• Add tests for the C++ part for loading the new solver formats correcly

Final details/cleanup

• Check all bow examples that don't pass the tests yet
• Use Rustdoc to generate a documentation for the public API
• Revisit all string parameters in public functions, make them usable for &str as well as String
• Go over all error types and check if there is a test for them
• Document evaluation of stresses and strains for the new beam elements
• Include source for choice of corotational reference frame https://apps.dtic.mil/sti/tr/pdf/ADA358178.pdf
• Completely revise the theoretical documentation
• Have a look at all the TODO comments in the code
• Document that files older than 0.7 are no longer supported
• Document old settings that currently have no effect anymore
• Have a look at all the unwrap() and expect() calls in the code
• Run Clippy and maybe include in the CI build as well
• Make warnings errors in the CI build
• Check for symmetry errors in the model (*2, /2)

Keep for later

• Move verification examples into their own LaTex document
• assert_abs_diff_eq does not seem to check the length of the inputs. Verify and open an issue.
• Use the new input format in the GUI, delegate conversion from older formats to the solver
• Use GXBeam as a reference for the beam verification examples. Also pick beam length != 1.
• Re-simulate user bows where the results didn't match well
• Find a way to compute dissipated energy of each component
• Implement stiffness of the force/draw curve
• Add string angle to the output results
• More advanced static algorithm (line search, modified Newton, step size control, ...)

Issues that need to be revisited after integrating the new solver

• Restrictions for layer height distribution #244
• Better error message when the sum of all layers is zero #234
• Consider using JSON Schema for validating input files #230
• Better error messages for file input/output #227
• Hypothetical use of the Rust programming language #215
• Implicit time integration #211
• Parallelize the simulation for better performance #207
• Consider using beam elements in absolute nodal coordinate formulation #189
• Include shear deformation #179
• New output result: Force that the string exerts on the limb tip #157
• Try different stiffness matrix decompositions for the static solver #156
• Abort simulation (static, dynamic) early when the solution diverges #154
• Consider using a coordinate system in which the limb curvature is positive #138
• Determine stiffness coefficients of the beam elements by numerical integration #137
• Improve performance of the contact algorithm and/or make contact handling optional #135
• Add some derivatives of static results to the output #128
• Write tests for all numerical utilities (splines, root-finding, geometry) #119
• Possibility to associate a specific layer with the profile curve #114
• Consider adopting the IMPACT contact model from MSC Adams #112
• Add a command line option for json output (instead of msgpack) #111
• Use heaptrack to ensure there are no dynamic allocations in the simulation code #110
• Handle a decreasing draw curve (snap-through) #84
• Revisit the use of sparse matrices #82
• Improve strain interpolation to avoid nick in stress curves #70
• Implement some sort of caching for element properties #63
• Do regression tests with the output of the bow model #41
• Enforce that cross section height < radius of curvature #30