OxiNum 0.1.2 Release

[0.1.2] - 2026-06-10

Fixed

• oxinum-complex parity_cross_validation: prop_asin_cbig_native_agree and
  prop_atanh_cbig_native_agree proptest cases timed out (>120 s) when run with
  HEAVY_CASES = 16 at full precision (40 significant digits). The two tests are
  now split into a separate proptest! block with VERY_HEAVY_CASES = 6 and
  PREC_LIGHT = 20, bringing each test well under the 120-second per-test budget
  while still exercising cross-family agreement to 1e-9 tolerance.

Maintenance

• Version bump to 0.1.2; all workspace crates aligned to this version.
• Dependency version pins updated across workspace [workspace.dependencies].

Full Changelog: v0.1.1...v0.1.2

OxiNum 0.1.1 Release

[0.1.1] - 2026-06-04

Added

• oxinum-complex (new crate): Arbitrary-precision complex numbers for the OxiNum
  workspace. CBig pairs two DBig components; native::BigComplex pairs two
  BigFloat components for binary-base complex arithmetic with explicit rounding
  control. Both types provide construction, arithmetic operators (all ownership
  variants), conjugate, norm-squared, transcendental functions (exp, ln, sqrt,
  pow), trigonometric functions (sin, cos, tan, sinh, cosh, tanh),
  inverse-trig functions (asin, acos, atan, asinh, acosh, atanh),
  Display/Debug formatting, and optional serde and num-traits features.
  CBig is re-exported from the oxinum facade as oxinum::CBig / oxinum::Complex.
• oxinum-complex num-complex feature: Two-way conversions between CBig and
  num_complex::Complex<f64> / Complex<i64> via From impls. Integer conversions
  store parts at unlimited DBig precision to prevent silent precision collapse.
• oxinum-complex TryFrom<&CBig> for (f64, f64): Fallible projection to f64
  pairs; returns OxiNumError::Overflow when either component exceeds f64::MAX.
• oxinum-complex cross-validation test suite: parity_cross_validation tests
  verify that CBig and native::BigComplex agree on known-value results; includes
  SciRS2 ArbitraryComplex compatibility tests.
• oxinum-float special module: Pure-Rust special mathematical functions on
  DBig — gamma, ln_gamma, digamma, erf, erfc, bessel_j0, euler_gamma
  (Euler–Mascheroni constant to 200 digits), catalan (Catalan's constant to 200
  digits). Gamma uses Lanczos (g=7) for x ∈ (0, 20] and Stirling series for x > 20.
• oxinum-float MpFloat and MpComplex adapter types (mp_float module):
  rug::Float/rug::Complex-compatible wrappers over DBig for drop-in replacement
  of GMP-backed types in SciRS2's arbitrary_precision module.
• oxinum-float native::bs_transcendental module: Binary-splitting evaluation of
  exp, sin, and cos for BigFloat above a 512-bit precision threshold,
  replacing the O(N²) iterative Taylor series.
• oxinum-int native::simd_ops module: SIMD-accelerated (nightly core::simd,
  with scalar fallback on stable) inner kernels for AND, OR, XOR, and within-limb
  shift operations on BigUint limb slices. Activated by oxinum_simd cfg emitted
  from build.rs only on nightly + simd feature.
• oxinum-int BitAndAssign, BitOrAssign, BitXorAssign for native::BigUint
  (both owned and borrowed right-hand-side variants).
• SciRS2 compatibility integration tests across all sub-crates (scirs2_int_compat,
  scirs2_float_compat, scirs2_rational_compat, scirs2_facade_compat,
  scirs2_trait_hierarchy_compat, scirs2_arbitrary_complex_compat).
• Allocation-profiling Criterion benchmarks for oxinum-int, oxinum-float, and
  oxinum-rational; bitwise/shift operation benchmarks for oxinum-int.

Fixed

• oxinum-float atan and atan2 precision collapse: intermediate arithmetic was
  carried at the (narrow) input precision rather than the requested guard precision,
  capping output accuracy at approximately 3 significant digits regardless of the
  precision argument. All internal literals, reductions, and halving loops now use
  dbig_at_precision / extend_precision at precision + 20 guard digits, giving
  accurate results to full requested precision.

Full Changelog: v0.1.0...v0.1.1

OxiNum 0.1.0 Release

Full Changelog: https://github.com/cool-japan/oxinum/commits/v0.1.0