Release v3.0.0

NEWS.md

@@ -0,0 +1,33 @@
+# MortalityTables.jl release notes
+
+## v3.0.0
+
+### Breaking changes
+
+- Julia 1.10 or newer is required.
+- Indexing a parametric model (`m[x]`) is removed. Use `hazard(m, x)` or `m(x)`.
+- `UltimateTable` no longer forwards indexing or `lastindex`; index the rates directly with `table.ultimate[age]`. Tables are opaque containers, consistent with `survival`, `decrement`, and `life_expectancy` already rejecting a whole table.
+- `survival` and `decrement` on rate vectors now require an `AbstractArray` first argument, and `decrement` is a single method that forwards to `survival`. This removes the method ambiguities that previously turned some misuse into an ambiguity error instead of the intended `ArgumentError`.
+- Parametric laws have concrete, promoted field types (`Makeham{Float64}`). Constructors accept any mix of `Real` keywords (`Makeham(a=1, b=0.5, c=0)` is a `Makeham{Float64}`), and every default law is `isbits`. `survival` is roughly 20x faster for laws without a closed-form cumulative hazard.
+- `MakehamGompertz` was removed from the export list (it was never defined).
+- Loading a table from a CSV file is a package extension activated by `using CSV`; the call `MortalityTable(CSV.File(...))` is unchanged.
+
+### Bug fixes
+
+- Kannisto's `cumhazard`, and therefore its `survival`, returned wrong values in 2.x (`survival(Kannisto(), 0)` was 1.76). The closed form had its `a` and `b` parameters swapped.
+- Two-age `survival` on a parametric model is now the difference of cumulative hazards rather than a ratio of survival probabilities, so it is well defined where survival underflows to zero (previously `NaN`).
+- XTbML rates are placed by their age and duration labels instead of by position. Eight mort.SOA.org tables label grouped ages; for example, t352 ("1946-49 Basic Table, ANB") has issue ages 12, 17, 22, …. For these tables most ages returned another age's rates: `select[17]` gave issue age 37's. Each rate now sits at its own age, with `missing` between the groups. An empty cell inside a select row no longer shifts the later rates to earlier ages, and a repeated label throws.
+- `Makeham`, `Gompertz`, and `Kannisto` with `b = 0` return the constant-hazard limit instead of `NaN`. At an infinite age a zero coefficient no longer turns the result into `NaN`: `survival(Gompertz(), Inf)` and a constant hazard's survival there are `0`, and a law with no hazard keeps survival `1`.
+- 34 additional mort.SOA.org tables with an empty metadata element now load (for example `table(217)`).
+- The `get_SOA_table` "not found" error message now includes the requested table name.
+- Showing a custom table no longer prints a broken mort.SOA.org link.
+
+### Other changes
+
+- `cumhazard` is defined for every parametric law. It integrates `hazard` numerically unless the law provides a closed form, and `survival` follows from it.
+- Parametric models accept and ignore a `DeathDistribution` in `survival`, `decrement`, and `life_expectancy`, and `omega` of a parametric model is `Inf`, so a model works wherever a table's rate vector does.
+- `life_expectancy` on a rate vector is linear rather than quadratic in the number of remaining ages and allocation free. An age past omega returns `0.0`.
+- `UltimateMortality` accepts any `AbstractVector` (ranges, views, vectors containing `missing`, or a vector already indexed by age) and wraps it without copying; `start_age` sets the first age whatever the input's own axes. `mortality_vector` is an alias for it.
+- XTbML parsing uses [XML.jl](https://github.com/JuliaData/XML.jl): loading a table is about 2 to 3 times faster and `using MortalityTables` no longer loads libxml2.
+- `Pkg`, `UnPack`, `Requires`, `Memoize`, `Parsers`, and `XMLDict` are no longer dependencies.
+- The Dukes-MacDonald functions are documented on the Tables page.

Project.toml

@@ -1,7 +1,7 @@
 name = "MortalityTables"
 uuid = "4780e19d-04b9-53dc-86c2-9e9aa59b5a12"
 authors = ["Alec Loudenback <alecloudenback@gmail.com>"]
-version = "2.6.0"
+version = "3.0.0"
 
 [deps]
 Artifacts = "56f22d72-fd6d-98f1-02f0-08ddc0907c33"