kmolan/multicalc-rust

What: the type carries mass, centre and rotational inertia, and offers inertia_about for a point. What the articulated-dynamics recursions will need is the 6x6 spatial matrix form, a transform to another frame combining the parallel-axis shift with a rotation, and the sum of two bodies' inertias. The last one already exists as open-coded arithmetic inside the MJCF loader, where synthesized_inertia weights the geom centres and sums the shifted tensors by hand — that duplication is the signal the operation belongs on the core type, and moving it also puts the parallel-axis term under the core crate's own tests. Where: crates/multicalc/src/spatial/inertia.rs, crates/multicalc-mjcf/src/body.rs. How to test: cargo test -p multicalc and cargo test -p multicalc-mjcf; assert transforming a body to a frame and back is the identity, that summing two halves of a split body reproduces the whole, and that the MJCF loader's results are unchanged after it switches to the shared operation.

What: the module docs say a filter is configured once, with the configuration checked up front, and that every call after that is total. The construction half is true, but none of MovingAverage, RunningMedian, SavitzkyGolay, Deadband, Hysteresis or SlewRateLimiter looks at the sample it is handed, so "total" here means "never panics", not "always returns a number" — and a reader takes the stronger meaning. The bar asks for the policy stated per module: which calls propagate silently, which check, and what a checked call returns. State it, and add the checked entry points issues 8 and 18 need. Where: crates/multicalc/src/signal_processing/mod.rs, and the doc comments on each filter. How to test: cargo test -p multicalc; the doctests carry the claims, so assert each stated behaviour with a non-finite sample in the module's own examples.

What: MultirotorMixer::new verifies that the allocation and its pseudo-inverse round-trip to the identity, comparing each entry against a hardcoded 1e-4. That is an absolute bar on a dimensionless quantity whose rounding scales with the allocation matrix's own magnitude, so a mixer with long arms and a small torque ratio is judged by a different standard than a compact one, and the same layout can pass at f64 and fail at f32. #229 makes the same argument for the finite-difference and Rk45 defaults. Express the bar in T::EPSILON terms scaled by the matrix norm. Where: crates/multicalc/src/plant/multirotor_mixing.rs. How to test: cargo test -p multicalc; assert the same set of layouts is accepted and rejected at f32 and f64, including one scaled up by a large factor.

What: solve_discrete_riccati and solve_discrete_lyapunov fix MAXIMUM_PASSES = 64 as a private constant and symmetric_eigendecomposition fixes max_sweeps = 60 as a local, so none of the three can be tuned and none states its budget in its signature. An iterative API should take a runtime budget with a documented default and a hard cap, which is what makes a worst-case execution time computable for a caller running these at startup on a target. Take the budget as an argument with the current values as the defaults, and report the passes taken on success so a marginal design is distinguishable from a comfortable one. Where: crates/multicalc/src/linear_algebra/riccati.rs, lyapunov.rs, symmetric_eigendecomposition.rs. How to test: cargo test -p multicalc; assert a problem needing more than a deliberately small budget reports DidNotConverge, and that the default budget reproduces every existing result.

What: low_pass and its three siblings refuse a frequency that reaches half the sampling rate, but magnitude_at, phase_at and delay_at accept any frequency at all and return the aliased answer for anything past it — a plot swept to the sampling rate mirrors back on itself with no indication, and a stability margin read off it at the wrong frequency is worse than no reading. The same three accessors on BiquadCascade inherit it. Return FrequencyOutOfRange the way the design constructors already do, and reject non-finite frequencies with it. Where: crates/multicalc/src/signal_processing/biquad.rs, cascade.rs. How to test: cargo test -p multicalc; assert each accessor refuses a frequency at and above Nyquist and agrees with today's values below it.