Export record_given_species_params() so callers can record changes without recalculating

[gustavdelius]

The gap

species_params<-() does two things: it records the values the caller changed in given_species_params, and it then rebuilds the species parameters and recalculates every size-dependent rate.

Code that fits a species parameter together with the rate array it determines needs the first but not the second. An optimiser that solves for ks and writes the matching metab itself, or for z_ext and the matching mu_b, gets no benefit from the recalculation — and it can be actively wrong, because the recalculation overwrites the arrays the caller just set.

The obvious workaround, writing into the species_params slot directly, avoids the recalculation but loses the recording. The value is then silently reverted the next time anything triggers a recalculation, since it was never marked as given. This failure is quiet: the model keeps running and returns plausible numbers. It bit mizerEcopath in four places — setFeedingLevels() lost its h and ks, and matchCatch() its fitted z_ext and D_ext, as soon as any species parameter was assigned afterwards.

This PR

Extracts the change detection out of species_params<-() into record_given_species_params(given, value, old_sp) and exports it, so that such code can record its changes the same way. It returns the updated given_species_params data frame and does nothing else — no rebuild, no rate recalculation.

sp_before <- species_params(params)
params@species_params$ks[sp_select] <- ks
params@metab[sp_select, ] <- new_metab
params@given_species_params <-
    record_given_species_params(given_species_params(params),
                                species_params(params), sp_before)

species_params<-() now calls the same function, so the matrix/list/S4 column handling added in #465 serves both paths and there is one implementation to maintain rather than two. Behaviour of species_params<-() is unchanged.

Only entries that actually changed are recorded, which is the important part: recording an unchanged value would turn a calculated species parameter into a given one and thereby stop it responding to the parameters it is derived from. The comparison is per species per parameter, so a parameter is protected only for the species whose value moved, and a parameter the caller left alone is never pinned.

Alternative considered

Adding a recalculate = FALSE argument to species_params<-(). That is possible — replacement functions do take extra arguments — but it would not be enough on its own: the method rebuilds the whole table from given_species_params before it calls setParams(), so suppressing only setParams() still re-derives the calculated parameters while leaving the rate arrays they feed stale. A public setter that can return an internally inconsistent object also seemed like the wrong thing to offer.

Testing

Seven new tests in test-species_params.R cover recording, the no-change case, per-species recording, NA treated as a value, a wholly new column, survival of a subsequent recalculation, and the row-count checks. Full suite: 3362 pass, 0 fail.

🤖 Generated with Claude Code