Fix species_params<-() diff for list, matrix and object columns

[gustavdelius]

Problem

The old-vs-new diff introduced in species_params<-() (mizer 3.2) compares every common column with:

changed <- !((old_vals == new_vals) | (is.na(old_vals) & is.na(new_vals)))

== is only reliable for plain atomic vectors:

• List columns (and columns holding S4/other objects) error with "comparison of these types is not implemented", so species_params(params) <- sp fails outright.
• Matrix columns return a matrix from == rather than one logical per species, so the changed mask and the downstream given[[col]][changed] <- ... indexing are wrong.

This means a model that carries a list column in species_params cannot round-trip through the setter at all. It surfaced in the downstream package mizerEcopath, whose ecopath_groups is a list column: every species_params(p) <- ... (including inside its Shiny tuning gadget) errored.

Fix

Take the fast == path only for plain atomic vectors (is.atomic, no dim, length no_sp), and otherwise compare per species with identical() — using row indexing for matrix columns and [[i]] for list columns. A newly-added given list column is initialised as a list rather than a scalar NA.

Behaviour for ordinary atomic columns is unchanged.

Tests

Added a regression test (species_params setter handles list and matrix columns) covering an unchanged round-trip, a detected change to a list element (recorded in given_species_params), and a matrix column round-trip. Full test-species_params.R passes (99 tests).

🤖 Generated with Claude Code