[mmyrte]

Stacked on #57, as asked in its review. Merge #57 first; this PR's base then moves to develop.

The argument checks spelled out length(x) == 1L && as.integer(x) == x and similar combinations in every function. R/util.R now has predicates for the conditions of a named stopifnot(), so the name stays the message the user sees:

stopifnot("id_period must be a single integer" = is_scalar_integerish(id_period))

• is_integerish(): finite whole numbers, any length including 0;
• is_scalar_integerish();
• is_scalar_numeric();
• is_scalar_character();
• is_scalar_logical(): TRUE or FALSE.

None of them accepts NA.

They replace the hand-written checks throughout R/: allocators, views, neighbors_t, pred_data_t, runs_t, trans_models_t, lp_problem, ducklake_db, coords_t and validation. devdocs/style/r.md points to them.

The only check left spelled out is distance_breaks in neighbors_t, which needs at least two elements. The trans_rates_lp penalty checks stay as they are because terminal_band may be NA.

Stricter checks

• id_periods in alloc_clumpy(), alloc_dinamica() and alloc_greedy() must be whole numbers, not just numeric.
• select_maximize may not be NA. In alloc_dinamica()'s eval_alloc_params_t() path, the old select_maximize || !select_maximize also accepted non-logical values.
• $next_id(n = ) rejects a fractional n instead of truncating it.
• create_coords_t_square() accepts an integer resolution as well as a double, and now reports through stopifnot() instead of stop().

Tests

test_util.R checks each predicate against 11 candidate values: 1L, 1, 1.5, c(1, 2), integer(0), NA_integer_, Inf, "1", "a", TRUE and NA. tinytest::build_install_test(): all ok, 656 results.

---

Generated by Claude Code