A library set of arbitrary precision numbers (aka. big numbers) for mathematics and numerics, implemented in Rust. It's a Rust native alternative to GNU GMP + MPFR + MPC. It features:
- Pure rust, full
no_stdsupport. - Focus on ergonomics & readability, and then efficiency.
- Optimized speed and memory usage.
- Current MSRV is 1.68. The MSRV covers the default build (no optional features); optional
features may require a newer Rust version (e.g.
rkyv_v08needs Rust ≥ 1.81).
dashu-base: Common trait definitionsdashu-int: Arbitrary precision integersdashu-float: Arbitrary precision floating point numbersdashu-ratio: Arbitrary precision rational numbersdashu-cmplx: Arbitrary precision complex numbersdashu-macros: Macros for creating big numbers
dashu is a meta crate that re-exports all the types from these sub-crates. Please see the README.md in each subdirectory for crate-specific introduction.
Construct numbers with the compile-time literal macros (no precision loss, any size) and the meta-crate type aliases — readable names for every number domain:
use dashu::{ubig, ibig, fbig, dbig, rbig, cbig};
use dashu::{Natural, Integer, Real, Decimal, Rational, Complex};
// Compile-time literals — zero precision loss, any size
let n: Natural = ubig!(0x5a4653ca_67376856_5b41f775_d6947d55_cf3813d1);
let e: Real = fbig!(0x1.ffffp1023);
let pi: Decimal = dbig!(3.1415926535897932384626);
let r: Rational = rbig!(22 / 7);
let z: Complex = cbig!(3 + 4i); // complex, decimal by default
// Meta-crate type aliases cover every number domain
let _neg: Integer = ibig!(-0x10ff);
let _prod = &n * &_neg; // Natural × Integer → Integer
// Explicit radix and base prefixes work too
let _hex = ubig!(dead_beef base 16);
let _bin = ibig!(-0b1111);
// Associated constants are available on every type
let _one: Natural = Natural::ONE;
let _unit = Complex::I; // the imaginary unitParse from strings in any base, then format back with the full std::fmt mini-language —
hexadecimal, scientific, positional expansion, and more:
use dashu::{ubig, Integer, Decimal, Rational};
use core::str::FromStr;
// Parse from strings — any base, scientific notation, rational form
let a = Integer::from_str_radix("1a2b3c", 16).unwrap();
let b: Decimal = "3.1415926535897932384626".parse().unwrap();
let c: Rational = "22/7".parse().unwrap();
// Full std::fmt mini-language for every type
assert_eq!(format!("{:#x}", a), "0x1a2b3c");
assert_eq!(format!("{:e}", b), "3.1415926535897932384626e0");
assert_eq!(format!("{:#}", c.in_expanded(10)), "3.(142857)");
// in_radix formats in any base; {:.N} rounds the fractional digits
assert_eq!(format!("{}", a.in_radix(32)), "1kaps");
assert_eq!(format!("{:.3}", c.in_expanded(10)), "3.143");
// Debug prints a compact head‥tail form for large values
assert_eq!(
format!("{:?}", ubig!(1) << 1000),
"1071508607186267320..4386837205668069376"
);Enable the serde feature for compact binary encoding (e.g. postcard) and
human-readable JSON round-trips with precision preserved:
// requires: features = ["serde"]
use dashu::{Integer, Rational, Real};
// Binary format: compact little-endian byte encoding
let a = Integer::from(12345u32);
let bytes = postcard::to_stdvec(&a).unwrap();
let b: Integer = postcard::from_bytes(&bytes).unwrap();
assert_eq!(a, b);
// Human-readable: precision-preserving string format
let r = Rational::from_parts(22u8.into(), 7u8.into());
let json = serde_json::to_string(&r).unwrap();
assert_eq!(json, r#""22/7""#);
// JSON round-trips preserve the value exactly
let rt: Rational = serde_json::from_str(&json).unwrap();
assert_eq!(rt, r);
// Floats serialize losslessly too, with precision preserved
let x: Real = "0x1.8p0".parse().unwrap(); // 1.5
let json2 = serde_json::to_string(&x).unwrap();
let _rt: Real = serde_json::from_str(&json2).unwrap();Enable the rand feature for uniform sampling — integers of a given bit-length,
floats in [0, 1) at a chosen precision, and more:
// requires: features = ["rand"]
use dashu::{Natural, Integer, Real};
use dashu::base::BitTest;
use dashu::integer::rand::UniformBits;
use dashu::float::rand::Uniform01;
use rand::RngExt;
let mut rng = rand::rng();
// Uniform random integers with a bit-length limit
let a: Natural = rng.sample(UniformBits::new(256));
let b: Integer = rng.sample(UniformBits::new(64));
assert!(a.bit_len() <= 256 && b.bit_len() <= 64);
// Uniform floats in [0, 1) at chosen precisions
let x: Real = rng.sample(Uniform01::new(53)); // binary64
let _y: Real = rng.sample(Uniform01::new(200)); // higher precision
assert!(x >= Real::ZERO && x < Real::ONE);
// sample_iter streams an unbounded sequence
let _stream: Vec<Natural> = rng.sample_iter(UniformBits::new(8)).take(3).collect();Convert freely between number domains — binary to decimal floats, floats to rationals (recovering the human-intended fraction), floats to integers with a rounding direction:
use dashu::float::round::mode::Zero;
use dashu::{Real, Decimal, Rational, Integer};
// Base conversion: binary (Real) ↔ decimal (Decimal)
let x: Decimal = "3.141592653589793".parse().unwrap();
let y = x.to_binary().value(); // Decimal → binary float
let _back = y.to_decimal().value(); // and back
// Float → rational: recover the fraction the programmer meant
let r = Rational::simplest_from_f64(0.1).unwrap();
assert_eq!(r, Rational::from_parts(1u8.into(), 10u8.into()));
// Float → integer with a rounding direction
let floor: Integer = y.to_int().value(); // truncates toward zero
assert_eq!(floor, Integer::from(3u8));
// Exact rational → float at a chosen precision
let third = Rational::from_parts(1u8.into(), 3u8.into());
let _f = third.to_float::<Zero, 2>(50).value(); // 1/3 in base 2dashu-python is a user-friendly test field for the core dashu
functionalities: through the dashu-rs
package on PyPI, users can try dashu's arbitrary-precision integers, rationals,
floats, and complex numbers from Python to get an idea of what dashu is capable
of — no Rust toolchain needed. Beyond exploring dashu, it also stands on its own
as a standalone arbitrary-precision number package for the Python ecosystem.
Licensed under either of
- Apache License, Version 2.0 (LICENSE-APACHE or https://www.apache.org/licenses/LICENSE-2.0)
- MIT license (LICENSE-MIT or https://opensource.org/licenses/MIT)
at your option.
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.