Add other_mort() and other_encounter() accessors (#579)

[gustavdelius]

Fixes #579.

The reported bug

The guide to creating an extension package described mizerStarvation as adding starvation mortality "via the other_mort argument in setComponent()". There is no such argument (the mortality argument is mort_fun), and mizerStarvation does not call setComponent() at all — it assigns params@other_mort[["starvation"]] <- "starvMort" directly. A reader following the guide's simplest worked example hit Error: argument "dynamics_fun" is missing, with no default.

The gap behind it

@other_mort and @other_encounter are named lists of function names that getMort() and getEncounter() iterate over and add to their results, and they had no exported accessor. other_params() deliberately reaches only params@other_params$other, and setComponent() was the only exported writer of the two slots — but it requires dynamics_fun and initial_value, which a term carrying no state of its own has nothing to supply. mizer already half-filled the table:

	fixed array	state-dependent function
mortality	ext_mort() / ext_mort<-()	@other_mort — nothing
encounter	ext_encounter() / ext_encounter<-()	@other_encounter — nothing

The route mizerStarvation actually takes is also a one-way door: getComponent() and removeComponent() are keyed on names(params@other_dynamics), so a bare @other_mort entry can neither be inspected nor removed through the API.

Changes

New accessors (R/extension.R) — other_mort() / other_encounter() and their replacement forms, as plain functions alongside other_params():

other_mort(params)[["starvation"]] <- "starvMort"

• They filter out component-owned entries, as other_params() does and for the same reason. An entry created by setComponent(mort_fun = ) stays the property of its component — getComponent() reports it, removeComponent() removes it — so assigning a whole list through the accessor can no longer wipe it out. Getter and setter have the same scope, so other_mort(params) <- other_mort(params) is an exact identity.
• Validation rejects a value that is not a function name, an unnamed entry, and a name that collides with a component (the error names setComponent()). Assigning NULL removes an entry; other_mort(params) <- NULL clears the free-standing ones only.
• other_dynamics deliberately gets no accessor: it is coupled to @initial_n_other (MizerSim validity checks the two agree), so it stays behind setComponent(). other_mort and other_encounter are independent registries with nothing to keep in sync.
• All four share one internal setRateContributions().

One behaviour change — setComponent() now refuses a component name that a free-standing contribution is already registered under, instead of silently taking it over and hiding it from the accessors. Recorded in inst/skills/upgrade-mizer-code/references/mizer-3.4.md with a row in the symptom index.

Documentation — the wrong sentence in inst/skills/create-extension-package/SKILL.md is rewritten to describe the supported route, with the two neighbouring mechanisms (ext_mort() for a fixed array, setComponent(mort_fun = ) for a component with state) spelled out so they cannot be confused again; a bullet added to the package-author checklist; a row and a paragraph added to inst/skills/extend-mizer/SKILL.md. The articles were regenerated with build_guides() — no generated vignette was edited by hand.

Verification

• Full test suite: [ FAIL 0 | WARN 0 | SKIP 49 | PASS 4947 ]
• 8 new tests in tests/testthat/test-extension.R: round trip, filtering, preservation across a whole-list assignment, the four validation errors, removal via NULL, both rates actually rising, projection of a model with a free-standing contribution and no components at all, the setComponent() collision guard, and time_modified for both setters.
• pkgdown::check_pkgdown() clean; the new page is picked up by has_concept("extension tools") with no _pkgdown.yml edit.
• lintr::lint() clean on the new lines.
• build_guides() reports "All links resolve" — which is also what checks that the new symptom-index row resolves to its heading and that its quoted error text is still a literal in R/.

inst/llms.txt / docs/llms.txt do not yet list the new exports; they are regenerated from a full pkgdown site build.

🤖 Generated with Claude Code