Skip to content

CHANGELOG.md

Latest commit

 

History

History
231 lines (202 loc) · 13.9 KB

CHANGELOG.md

File metadata and controls

231 lines (202 loc) · 13.9 KB
Raw

Changelog

This changelog aggregates the releases of every dashu crate — dashu, dashu-base, dashu-int, dashu-float, dashu-ratio, dashu-cmplx, dashu-macros, and the dashu-python binding. Each entry is tagged with the crate and minor version it belongs to; see the per-crate CHANGELOG.md in each subdirectory (base/, integer/, …) for full detail.

Unreleased

Change

  • dashu-macros: (breaking) cbig! / static_cbig! coefficients are now decimal by default (was binary); use the 0x / 0b / 0o prefixes for other bases.

0.6.0-rc.3 — coordinated release

Add

  • dashu-base: Approximation::value_with_exact() — the value together with an is_exact flag (feeds dashu-float's exact-tracking Ball ops and hypot).
  • dashu-int: rkyv 0.7 (rkyv_v07) and rkyv 0.8 (rkyv_v08) support — UBig/IBig archive as their native word vectors (zero-copy).
  • dashu-float: tuning feature + ziv_retries()/ziv_retries_reset() — per-Ziv-loop retry counter for profiling how tight each transcendental's error radius is.
  • dashu-float: rkyv 0.7/0.8 support for Repr/Context/FBig; Repr::zero_with_sign is now public.
  • dashu-ratio: rkyv 0.7/0.8 support for RBig/Relaxed.
  • dashu-cmplx: serde and zeroize support for CBig; rkyv 0.7/0.8 support; signed-zero preservation on exact-zero sin_cos/sqr/log.
  • dashu-python: CI wheel/sdist build workflow (.github/workflows/python-wheels.yml).

Change

  • dashu-int / dashu-float / dashu-ratio / dashu-cmplx: unversioned rand/rkyv feature aliases now select rand 0.10 / rkyv 0.8 (pin rand_v08/rkyv_v07 to keep the old).
  • dashu-float: log2's error radius now derived by Ball arithmetic instead of directed intervals (precision-aware propagation; public behavior unchanged, performance at parity).
  • dashu-float: exp/exp_m1, the hyperbolic and trigonometric families, powi, and hypot migrated to the same Ball error-propagation engine.
  • dashu-ratio: (breaking) RBig::pow/Relaxed::pow take isize (was usize); negative exponents reciprocate.
  • dashu-cmplx: (breaking) complex infinity is now a terminal value — no arithmetic accepts an infinite operand anymore (rejects with FpError::InfiniteInput).
  • dashu-macros: version-aligned to rc.3 (no functional changes; skipped rc.2).

Fix

  • dashu-float: powf/powi wrongly rounded for large |y·ln x| (input-error inflation omitted the result-significand factor).
  • dashu-float: sqrt-based radius under-bound in asin/asinh/acosh/hypot.
  • dashu-float: atanh(x) for x < 0 near the pole returned wrongly-rounded values.
  • dashu-float: powf of a base < 1 no longer hangs in the Ziv loop (scale_int → exact-tracking scale_int_tracking).
  • dashu-float: cargo test --no-default-features compiles again (Ziv counter gated on std).
  • dashu-cmplx: dangling intra-doc link in the fma docs.

0.6.0-rc.2 — dashu-float only

Change

  • Unified f32/f64 range detection: into_f32_internal/into_f64_internal now return FpResult, so the range decision lives in one place.
  • Context::exp of an astronomically large negative value now returns Err(Underflow).
  • Ziv driver takes fallible closures; hoisted overflow probes removed.

Fix

  • Directed overflow/underflow of FBig results now saturate to the mode-aware endpoint (Context::unwrap_fp), so Up ≥ Down holds on both ends.
  • Directed overflow/underflow of FBig → f32/f64 (range detection delegated to encode).
  • TryFrom<FBig> for f32/f64 error variant is now mode-independent.
  • Directed ln/log2 of x ∈ [1, B) and of x just above 1 (radius under-estimates).
  • exp/exp_m1 extreme-negative endpoint carried precision 0; ulp()/ulp_lb() panic on an extreme exponent (saturating arithmetic); exp_m1 of large negative input short-circuits.
  • exp overflow probe could disagree with the computation; exp range reduction for large |x|.
  • FBig → f32/f64 subnormal mis-round (convert at width + 24 bits); underflow of a catastrophically tiny source short-circuits before conversion.
  • powi memory exhaustion / hang on extreme exponents (margined guard + exp(y·ln x) fallback); panic at the exact exponent ceiling; Repr::cmp overflow near the exponent ceiling.

0.6.0-rc.1 — coordinated release

Add

  • dashu-base: ParseError::InvalidSyntax variant for structurally malformed input (breaking for exhaustive ParseError matches).
  • dashu-float: exact Add/Sub/Mul operators for Repr (lossless intermediates).
  • dashu-float: guaranteed-correct rounding for exp/exp_m1/ln/ln_1p, the trig family, and the hyperbolic family via a Ziv retry loop.
  • dashu-float: hypot, powf, and powi now correctly rounded via Ziv (with exactness tracking for hypot); Repr::is_int is now const.
  • dashu-cmplx: correct rounding for the complex transcendentals (exp, ln, powf, powi, trig family, abs, arg, sqrt) via a complex Ziv driver.

Change

  • dashu-float: (breaking, bound) Ziv-backed transcendentals now require R: ErrorBounds instead of R: Round.
  • dashu-float: FBig::log2 now correctly rounded via Ziv (was near-correct).
  • dashu-cmplx: (breaking, bound) complex transcendentals now require R: ErrorBounds; powi correctly rounded via its own Ziv loop.
  • dashu-cmplx: FromStr returns ParseError::InvalidSyntax for malformed input.
  • dashu-ratio: from_str_radix/from_str_with_radix_prefix reject multiple / separators with ParseError::InvalidSyntax.
  • dashu-macros: literal macros surface the new ParseError::InvalidSyntax with a dedicated panic message.

Fix

  • dashu-int: NTT squaring/multiplication of an all-zero operand no longer panics (the crash behind dashu-float's hypot(3,4) under directed rounding).
  • dashu-float: directed-rounding hang on exactly-representable results (acos(±1), asin(0), exact hypot); tan pole check removed (~2× cost); no_std build of the Ziv retry counter.
  • dashu-cmplx: Sum for CBig now correctly rounded; unlimited-precision arithmetic correct; no_std build of the Ziv retry counter.

0.5.x

crates: dashu-base, dashu-int, dashu-float, dashu-ratio, dashu-cmplx, dashu-macros, dashu-python

  • dashu-float (0.5.2): Add FBig::e (Euler's constant, exact binary splitting), FBig::fma (fused multiply–add, one rounding), FBig::ulp_lb, Context::addsub_vv/vr/rv/rr kernels; x ± 0 now rounds to the context precision; mul/sqr/cubic strictly correctly rounded.
  • dashu-base (0.5.1): Fix BitTest::bit at the sign-bit position; FloatEncoding::encode no longer returns NaN for huge exponents.
  • dashu-int (0.5.1): Fix to_f64 exactness at DoubleWord::MAX; GCD of large similarly-sized integers no longer aborts.
  • dashu-float (0.5.1): Add Repr::new_const, correctly-rounded FBig::log2; fix spurious powi overflow for base ≈ 1; single-rounding to_f64/to_f32.
  • dashu-ratio (0.5.1): Add from_str_expanded/from_str_decimal (repeating notation), num_traits::Pow<isize>; fix GCD memory abort, UBig::try_from(RBig) denominator check.
  • dashu-cmplx (0.5.1): Add CBig::from_parts_const (const cbig!).
  • dashu-macros (0.5.1): cbig! in const position for DoubleWord-fitting coefficients.
  • dashu-base (0.5.0): Remove AbsEq (use AbsOrd); full docs + #![deny(missing_docs)].
  • dashu-int (0.5.0): Add to_digits/from_digits; breaking serde format → two's-complement LE bytes, in_radix radix → u8; fix nth_root(0); full docs.
  • dashu-float (0.5.0): IEEE-754 signed zero; FpError/FpResult error model; ConstCache/CachedFBig; hyperbolic functions; hypot; sinh_cosh; breaking is_zero → is_pos_zero, correctly-rounded Sum, serde precision padding, Context returns FpResult; remove public from_str_native.
  • dashu-ratio (0.5.0): breaking From<RBig/Repr> for FBig → TryFrom (exactness in the target base); in_radix radix → u8; fix UBig::try_from(RBig).
  • dashu-cmplx (0.5.0): initial release — two-layer Context/CBig API, field arithmetic, powi/powf, transcendentals, CachedCBig, cbig!/static_cbig!, rand, num-complex interop.
  • dashu-macros (0.5.0): cbig!/static_cbig!; #![deny(missing_docs)].
  • dashu-python (0.5.0): initial release of dashu-rs on PyPI — six types, panic-free transcendentals, module math API, optional serde/rand/rkyv/zeroize.

0.4.x

crates: dashu-base, dashu-int, dashu-float, dashu-ratio, dashu-macros

  • dashu-float (0.4.5): Add quantize, cbrt/nth_root, trig functions + π (Chudnovsky, binary splitting), FpResult, rand_v09/rand_v010; fix to_f32/to_f64 rounding, add/sub rounding bugs.
  • dashu-float (0.4.4): Bump MSRV to 1.68.
  • dashu-ratio (0.4.4): Fix to_f32/to_f64 double rounding.
  • dashu-base (0.4.3): Add Sign::as_sign_str; fast integer sqrt via native f64::sqrt.
  • dashu-int (0.4.3): Add NTT multiplication, asymmetric NTT, monty module, specialized squaring, const from_u64/from_i64, rand_v09/rand_v010, version-agnostic rand module; double-word mul kernels, runtime threshold tuning (DASHU_THRESHOLD_*); assorted fixes.
  • dashu-float (0.4.3): Deprecate from_str_native; TryFrom for primitive ints/floats; formatting trait impls.
  • dashu-ratio (0.4.3): Add Binary/Octal/LowerHex/UpperHex, in_radix/in_expanded, rand_v09/rand_v010, version-agnostic rand module.
  • dashu-base (0.4.2): log2_bounds strictly enclosing; MSRV bump.
  • dashu-int (0.4.2): Add ones/as_ubig/from_chunks/to_chunks, TryFrom<f32/f64>, IBig byte conversions, mixed bit ops; fixes; MSRV bump.
  • dashu-float (0.4.2): Repr::from_static_words; FBig::from_repr_const; NumOrd/AbsOrd with ints; Debug no longer shows the rounding mode.
  • dashu-ratio (0.4.2): Div<RBig>/Div<Relaxed> for UBig/IBig; division bug fix; MSRV bump.
  • dashu-macros (0.4.2): Replace paste with pastey; MSRV bump.
  • dashu-base (0.4.1): Deprecate AbsEq; re-implement next_up/next_down.
  • dashu-int (0.4.1): AbsEq/AbsOrd; from_static_words (static macros); is_multiple_of; const trailing_zeros/trailing_ones; trailing_ones fix.
  • dashu-float (0.4.1): Fix ln/exp series termination; powf no longer panics on base 0.
  • dashu-ratio (0.4.1): AbsOrd/NumOrd with ints/floats; as_relaxed; to_float enforces precision on zero.
  • dashu-macros (0.4.1): static_ubig!/static_ibig!/static_fbig!/static_rbig!.
  • dashu-base (0.4.0): is_positive/is_negative; trait moves (SquareRoot/CubicRoot → math, AbsCmp → AbsOrd, square → sqr).
  • dashu-int (0.4.0): ConstDivisor; as_ibig; NumOrd; serde format → LE byte sequence; IntoRing refactor; Modulo → Reduced; remove PartialOrd/PartialEq between UBig/IBig.
  • dashu-float (0.4.0): NumOrd/NumHash; ErrorBounds; feature defaults (num-order on); Repr::BASE → UBig; sqrt via the SquareRoot trait.
  • dashu-ratio (0.4.0): is_int; NumOrd/NumHash; simplest_from_float; feature defaults; remove PartialOrd between RBig/Relaxed.
  • dashu-macros (0.4.0): Remove the embedded feature.

0.3.x

crates: dashu-base, dashu-int, dashu-float, dashu-ratio, dashu-macros

  • dashu-float (0.3.2): Default precision from the actual digits of integer conversions.
  • dashu-ratio (0.3.2): Multiplication bug fix.
  • dashu-base (0.3.1): Inverse trait; AbsCmp/AbsEq for primitives.
  • dashu-int (0.3.1): UniformBits; count_ones/count_zeros; cubic; rand_v08/ num-traits_v02 feature flags.
  • dashu-float (0.3.1): num_traits/rand/serde/diesel/postgres impls; Uniform01/ UniformFBig; to_f32/to_f64; round; rand_v08/num-traits_v02.
  • dashu-ratio (0.3.1): Sum/Product/Rem/num_traits/rand/serde impls; cubic/pow/ round; rand_v08/num-traits_v02.
  • dashu-macros (0.3.1): Fix ibig/rbig macros using incorrect crate names.
  • dashu-base (0.3.0): AbsCmp/AbsEq/FloatEncoding/Signed/SquareRoot/CubicRoot/ EstimatedLog2; error types moved from int; BitTest rework.
  • dashu-int (0.3.0): gcd/gcd_ext/div_rem between UBig/IBig; BitTest for IBig; TryFrom<f32/f64>; trait-only exposure; to_le_bytes → Box<[u8]>.
  • dashu-float (0.3.0): Subnormal f32/f64 conversion; split_at_point.
  • dashu-ratio (0.3.0): initial release — basic arithmetic, numeric conversion, Diophantine approximations.
  • dashu-macros (0.3.0): Const context limited to u32-fitting literals; embedded feature.

0.2.x

crates: dashu-base, dashu-int, dashu-float, dashu-macros

  • dashu-base (0.2.1): RootRem/Root for unsigned primitives.
  • dashu-int (0.2.1): sqrt/sqrt_rem/nth_root; Sum/Product; num_traits/ num_integer/num_order/zeroize impls; serde is_human_readable.
  • dashu-float (0.2.1): Sum/Product; powf; sqrt.
  • dashu-base (0.2.0): Approximation, Sign, EstimatedLog2.
  • dashu-int (0.2.0): Expose Sign/DoubleWord/as_words; const constructors; split_bits/clear_high_bits; ilog/log2_bounds; macros moved to a separate dashu-macro crate; ZERO/ONE/NEG_ONE; pow speed-up.
  • dashu-float (0.2.0): initial release — basic arithmetic (add/sub/mul/div/ exp/ln) and base conversion.
  • dashu-macros (0.2.0): initial release — integer and float literal macros.

0.1.x

crates: dashu-base, dashu-int

  • dashu-base (0.1.1): GCD algorithm bug fix.
  • dashu-int (0.1.1): Modular inverse; gcd/extended_gcd.
  • dashu-base (0.1.0): initial release — common trait definitions.
  • dashu-int (0.1.0): initial release — ported from ibig @ 0.3.5 with modifications (see integer/NOTICE.md).