range-set-blaze

Releases

1. ▸▾0.7.015 Sept 2026Deprecated3 sources agree

   …types to the crate's Integer trait, removing deprecated warnings that arose because Rust resolves those names to std's deprecated inherent methods bef…

   Added

   • Added the experimental cursor_nightly_experimental feature, which uses Rust's
     nightly-only B-tree cursor API to speed up insert/ranges_insert on both
     RangeSetBlaze (~2x) and RangeMapBlaze (~1.7x), with the same public API.
   • Floating-point range support for f32 and f64, plus range_at,
     range_or_gap_at, and fill_gaps, are now unconditional public APIs.
   • RangeSetBlaze::fill_gaps returns a RangeMapBlaze<T, bool> and
     RangeMapBlaze::fill_gaps returns a RangeMapBlaze<T, Option<V>>, following
     the crate's usual two-layer design: collection methods return gathered
     collections, while SortedDisjoint::fill_gaps and
     SortedDisjointMap::fill_gaps return the lazy streams
     (FillGapsIter and FillGapsIterMap) for advanced composition. Both cover
     the complete integer domain from T::min_value() through T::max_value().
   • FillGapsIterMap and FillGapsIter now implement SortedDisjointMap and can be
     collected directly into RangeMapBlaze<T, Option<V>> and
     RangeMapBlaze<T, bool>, respectively. bool now implements ValueCarrier as
     the canonical by-value carrier for totalized sets.
   • Added docs/release_checklist.md, the canonical release procedure, and a
     "Policy on AI-assisted development and contributions" section in the README.

   Fixed

   • Fixed doc links on FillGapsIter/FillGapsIterMap that resolved to the wrong
     path because the linked-to trait docs used a bare relative .html path instead
     of an intra-doc crate:: path; this broke once a type implementing
     SortedDisjoint/SortedDisjointMap first existed in a nested module (gaps).

   Changed

   • CI's nightly toolchain (used for the cursor_nightly_experimental/
     float_nightly_experimental checks, Clippy, and docs) is now pinned to a fixed
     dated nightly instead of floating nightly, mirroring the already-pinned
     stable toolchain in rust-toolchain.toml. A floating nightly previously broke
     CI when upstream Clippy added the single_range_in_vec_init lint against this
     crate's idiomatic from_iter([a..=b]) single-range construction; that lint is
     now explicitly allowed for the nightly Clippy run. Bump the pinned nightly
     deliberately, the same way the stable toolchain is bumped.
   • The 32-bit Ubuntu CI job now retries its 32-bit apt-get package install
     with a fresh package list, since awalsh128/cache-apt-pkgs-action reports
     success even when its internal install fails (e.g. a transient mirror 404),
     which previously surfaced later as a confusing linker error instead of a
     clear package-install failure.
   • Qualified test-only min_value()/max_value() calls on concrete primitive
     integer types to the crate's Integer trait, removing deprecated warnings
     that arose because Rust resolves those names to std's deprecated inherent
     methods before the trait method.

   Breaking

   • Removed the rog_experimental and float_experimental Cargo features. Remove
     them from dependency feature lists. float_nightly_experimental remains for
     nightly-only f16 and f128 support.
   • Removed Rog, RogsIter, and the old rogs_* methods. Use range_at,
     range_or_gap_at, or fill_gaps instead.
   • Renamed ValueRef to ValueCarrier and redesigned it with a Value
     associated type and value_eq method; generic code now uses explicit
     borrowing only where borrowed access to the logical value is required.
   • Renamed Finite/FiniteF32/FiniteF64/FiniteF16/FiniteF128 to
     NotNan/NotNanF32/NotNanF64/NotNanF16/NotNanF128 (module finite ->
     not_nan, shorthand constructors ff64/... -> nnf64/...,
     FiniteSliceExt/FiniteRangeExt -> NotNanSliceExt/NotNanRangeExt). The
     type's legal domain grew to match: it now excludes only NaN, so +infinity
     and -infinity are legal values and MIN/MAX are the two infinities
     (previously f64::MIN/f64::MAX); MAX_SIZE grew by 2 to account for them.