Fix validSim() passing the resource spectrum as the other components

[gustavdelius]

What changed

validSim.MizerSim() diagnoses a simulation that went non-finite by recomputing the rates at the last good time step. The getRates() call passed sim@n_pp[max_t_idx, ] for the n_other argument — a copy-paste slip that handed the resource number density to the rate functions where they expect the list of other components.

Any rate function reading n_other$<component> therefore failed with:

Error in `n_other$gonads`: $ operator is invalid for atomic vectors

Rather than just swapping the slot, the call now uses the existing internal helper get_sim_rate_slice() (R/sim_rates.R), which already builds exactly this slice. That matters because a plain sim@n_other[max_t_idx, ] is not enough on its own: on a list-matrix with a single component, [ drops to an unnamed length-1 list, so n_other$<component> would come back NULL for a one-component model. The helper restores the component names from dimnames(sim@n_other)$component. It also supplies the dimnames on n that the hand-rolled dim(n) <- dim(sim@n)[2:3] left off.

Why it was invisible to the test suite

The shared fixtures have no other components, so nothing in the suite ever read n_other on this path. It surfaced when the suite was run with NS_params_small replaced by a mizerSeasonal object, which adds a gonads component — test-MizerSim-class.R ("validSim works") then errored as above.

Notes for the reviewer

• There is only one occurrence of the slip. validSim.MizerSim() has a single if (!all(is.finite(sim@n))) block; is.finite() is already false for NaN as well as Inf/-Inf, and the existing test's second half exercises the NaN case through that same branch. There is no separate NaN or negative-value branch in the file — grep for is.finite|is.nan|negative|< 0 in R/MizerSim-class.R returns only the three lines of this block.
• The two other places that slice a row out of @n_other (get_sim_rate_slice() itself and finalNOther()) already restore the names correctly, so nothing else needed fixing.
• No upgrade-mizer-code entry: this turns a crash into working diagnostics and moves no results for code that was already working.

Tests

A regression test goes in tests/testthat/test-MizerSim-class.R, the file named after R/MizerSim-class.R which defines validSim (per .claude/skills/test-organisation.md). It builds on the shared NS_params_small fixture plus a one-component setComponent() model whose mort_fun reads n_other$gonads and records what it was handed, then asserts the truncation happened, that names(n_other) is "gonads", and that the value matches the one stored at that time step — so it catches both the wrong-slot slip and a names-dropping regression.

Verified the test reproduces the reported error against the unfixed code, and passes with the fix.

• devtools::test(filter = "MizerSim-class") — 61 pass, 0 fail
• devtools::test(filter = "extension|sim_rates|project|summary_methods") — 515 pass, 0 fail

NEWS.md gets an entry under Extensions.

🤖 Generated with Claude Code