NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
crates.io
crates.io, #4336 most downloaded on crates.io
Integer sets as fast, sorted integer ranges; Maps with integer-range keys; Full set operations
Latest stable
0.8.0
Last release 1 months ago
09 Oct 2026
Release timing varies
gaps range from 8 days to 13 months
Some releases are documented
notes for 15 of 26 stable releases
Nothing withdrawn
no release was ever pulled
4 years old
35 releases · first in 2023
Drag across the ruler to compare, or tap a release to open it.
…SymDiffKMergeMap, IntersectionKMap) remain as deprecated aliases. Code that named the old internal types directly (for example UnionIter<...> for a |…
This is a breaking release (see Changed and Removed). Code that only uses
RangeSetBlaze, RangeMapBlaze, and their operators and methods is unaffected
apart from the new minimum Rust version. It also includes the from_slice
fix described under 0.7.1 in the changelog,
which was never published to crates.io.
SortedDisjointMap::inner_join, left_join, and full_join
pair up two map streams by key, yielding (range, (left, right)) with
Option for a side that may be absent. RangeMapBlaze::inner_join,
left_join, and full_join take a function of the two values (for
example, a.inner_join(&b, |l, r| ...)) and return a map of its results.
left_join stops as soon as the left input ends, without reading the rest
of the right input.range_set_blaze::joins) explains the joins,
transform_values, and sweep, and when to use each.MultiwaySortedDisjointMap::inner_join and full_join
(and the same on collections of &RangeMapBlaze) call a closure once per
stretch of keys with every input's value there (&[V], or
&[Option<V>] for the full join) and collect its results.transform_values on map streams and on RangeMapBlaze: same keys,
new values, merging touching ranges whose new values are equal.MultiwaySortedDisjointMap::sweep, a low-level stream of every input
range's start and end in key order (SweepEvent), for computations the
joins do not cover. The multiway joins are built on it.Owned<V>, a ValueCarrier that holds its value directly, so stream
operations can produce new values without allocating. In the prelude.RangeMapBlaze::universe_with(&value) and RangeSetBlaze::universe().ValueCarrier for pairs of carriers, so joined streams compose with other
map operations.InnerJoinMap, LeftJoinMap, FullJoinMap,
TransformValuesMap, MultiwayInnerJoinMap, MultiwayFullJoinMap,
MultiwaySweep, and SweepEvent.| and
^ operators are about 1.6-2.9x faster; multiway map union and
symmetric_difference are about 1.4-4x faster. Multiway set
symmetric_difference is also faster in most cases (up to about 1.23x
slower with 2-4 sparse inputs). Results are unchanged.btree_set::Union): Union,
Intersection, Difference, SymmetricDifference, MultiwayUnion,
MultiwayIntersection, MultiwaySymmetricDifference, UnionMap,
IntersectionMap, DifferenceMap, SymmetricDifferenceMap, NotMap,
MultiwayUnionMap, MultiwayIntersectionMap, and
MultiwaySymmetricDifferenceMap. They were hidden type aliases exposing
internal types, so any change of implementation changed them; now their
fields are private. Code that only iterates or chains these results is
unaffected. The old names (UnionMerge, IntersectionMerge,
DifferenceMerge, SymDiffMerge, UnionKMerge, SymDiffKMerge,
UnionMergeMap, SymDiffMergeMap, UnionKMergeMap, SymDiffKMergeMap,
IntersectionKMap) remain as deprecated aliases. Code that named the old
internal types directly (for example UnionIter<...> for a | b) must use
the new names.SortedDisjointMap::complement now returns NotMap (was the
equivalent NotIter<..., RangeValuesToRangesIter<...>>), complement_with
returns RangeToRangeValueIter<..., NotMap<...>>, and
MultiwaySortedDisjointMap::intersection no longer has an unused lifetime
parameter.RangeMapBlaze docs now state that values should be cheap to clone,
because operations clone a value whenever they split its range; wrap large
values in Rc or Arc.just check-all now also runs the MSRV check, matching CI.KMergeMap, MergeMap, and SymDiffIterMap. They had no
public constructors and are no longer used by the crate.Your coding agent can read these notes before it upgrades. Set up the MCP server →
Qualified test-only min_value()/max_value() calls on concrete primitive integer types to the crate's Integer trait, removing deprecated warnings that…
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.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.docs/release_checklist.md, the canonical release procedure, and a
"Policy on AI-assisted development and contributions" section in the README.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).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.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.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.rog_experimental and float_experimental Cargo features. Remove
them from dependency feature lists. float_nightly_experimental remains for
nightly-only f16 and f128 support.Rog, RogsIter, and the old rogs_* methods. Use range_at,
range_or_gap_at, or fill_gaps instead.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.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.Added experimental floating-point range support through the float_experimental feature for f32/f64, and the float_nightly_experimental feature for f16
float_experimental feature
for f32/f64, and the float_nightly_experimental feature for f16/f128.
Finite supports finite values with zero canonicalized to +0.0, while Total supports
all floating-point values in total order, including NaNs, infinities, and signed zero.
Both provide constructors, range operations, and after/before navigation APIs.Fix wasm-pack test command syntax for improved compatibility
Fix wasm-pack test command syntax for improved compatibility
AssumeSortedStarts now verifies, in debug builds only, that the ranges it is given
actually have sorted starts, panicking with a clear message otherwise. Release builds are
unaffected (zero runtime cost). (PR #29)AssumeSortedStarts regains a leading type parameter: AssumeSortedStarts<I> ->
AssumeSortedStarts<T, I>. Most AssumeSortedStarts::new(iter) calls are unchanged due to
inference; explicit type aliases/annotations may need updating.Update rand package versions to 0.8.6 and 0.9.4 in Cargo.lock
Update rand package versions to 0.8.6 and 0.9.4 in Cargo.lock
SortedStarts/SortedDisjoint (and the *Map equivalents) for applicable
core::iter/std::iter adapter types whose iterators cannot break the sorted-disjoint
invariants, so these standard iterators can be used directly with the library's operators
(PR #27).Option<SortedDisjoint> into a SortedDisjoint via
FlatMap/Flatten.core::iter::StepBy is not yet supported because it does not implement FusedIterator
upstream; see the TODO in src/sorted_disjoint.rs and
https://internals.rust-lang.org/t/implement-fusediterator-for-core-stepby/24074Removed redundant struct-level trait bounds across iterator/map/set wrapper structs and kept constraints on impls/usage sites where needed.
AssumeSortedStarts and AssumePrioritySortedStartsMap now take fewer generic parameters.
AssumeSortedStarts<T, I> -> AssumeSortedStarts<I>AssumePrioritySortedStartsMap<T, VR, I> -> AssumePrioritySortedStartsMap<I>AssumeSortedStarts::new(iter) and
AssumePrioritySortedStartsMap::new(iter) are typically unchanged due to inference; explicit
type aliases/annotations may need updates.Parallelize check-all, skip benches locally, and fix SIMD/example reg…
Parallelize check-all, skip benches locally, and fix SIMD/example reg…
RangeSetBlaze::ranges_insert and RangeMapBlaze::ranges_insert now accept any RangeBounds<T>, not just RangeInclusive<T> (issue #24).Update lockfile for 0.4.3
Update lockfile for 0.4.3
cargo check-all now runs independent checks in parallelcheck-all test target now uses --lib --tests --examples to skip slow benchesfrom_slice SIMD imports/bounds for compatibility with current Rust nightly APIs.examples/nine_rules_maps assertions to match right-to-left precedence semantics.Fix SIMD compatibility for nightly: remove LaneCount/SupportedLaneCount
from_slice feature for compatibility with Rust nightly (post-2026-01-28)
LaneCount and SupportedLaneCount trait bounds from SIMD codefrom_slice featureImplemented RangeOnce , a zero-allocation adapter that yields 0 or 1 non-empty inclusive ranges.
Implemented RangeOnce<T>, a zero-allocation adapter that yields 0 or 1
non-empty inclusive ranges.
RangeOnce implements SortedStarts<T> and SortedDisjoint<T>,
providing a sound and ergonomic way to work with single ranges.
Example:
&a & RangeOnce::new(15, 35);
&a | RangeOnce::new(22, 25);
Empty ranges (start > end) now produce an empty iterator,
preserving all invariants.
Added From<RangeInclusive<T>> for RangeSetBlaze, allowing direct
conversion of single ranges:
RangeSetBlaze::from(5..=10);
RangeSetBlaze::from(5..=4); // yields empty set
Changed precedence of RangeMapBlaze to always be right-to-left.
RangeMapBlaze to always be right-to-left.Some breaking changes in the rest of the package to improve the API and performance.
RangeMapBlaze.char, IpAddV4, and IpAddV6 integer-like types.Nothing published for this version
Nothing published for this version
Added RangeSetBlaze::from_sorted_starts
RangeSetBlaze::from_sorted_startsSortedStarts and AssumeSortedStartsCheckSortedDisjoint::new to support IntoIteratorAssumeSortedStarts::new to support IntoIteratorNothing published for this version
Added DoubleEndedIterator when iterating integer elements. Thanks to enh.
Added optional from_slice cargo feature and support for a new RangeSetBlaze::from_slice constructor. The feature uses SIMD and requires Rust nightly.
from_slice cargo feature and support
for a new RangeSetBlaze::from_slice constructor.
The feature uses SIMD and requires Rust nightly.Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
One column per quarter.