use num_bigint::Sign::Plus; use num_bigint::{BigInt, ToBigInt}; use num_bigint::{BigUint, ToBigUint}; use num_integer::Integer; use std::cmp::Ordering::{Equal, Greater, Less}; use std::collections::hash_map::RandomState; use std::hash::{BuildHasher, Hash, Hasher}; use std::i64; use std::iter::repeat; use std::str::FromStr; use std::{f32, f64}; use std::{i128, u128}; use std::{u16, u32, u64, u8, usize}; use num_traits::{ pow, CheckedAdd, CheckedDiv, CheckedMul, CheckedSub, FromPrimitive, Num, One, Pow, ToPrimitive, Zero, }; mod consts; use crate::consts::*; #[macro_use] mod macros; #[test] fn test_from_bytes_be() { fn check(s: &str, result: &str) { assert_eq!( BigUint::from_bytes_be(s.as_bytes()), BigUint::parse_bytes(result.as_bytes(), 10).unwrap() ); } check("A", "65"); check("AA", "16705"); check("AB", "16706"); check("Hello world!", "22405534230753963835153736737"); assert_eq!(BigUint::from_bytes_be(&[]), BigUint::zero()); } #[test] fn test_to_bytes_be() { fn check(s: &str, result: &str) { let b = BigUint::parse_bytes(result.as_bytes(), 10).unwrap(); assert_eq!(b.to_bytes_be(), s.as_bytes()); } check("A", "65"); check("AA", "16705"); check("AB", "16706"); check("Hello world!", "22405534230753963835153736737"); let b: BigUint = Zero::zero(); assert_eq!(b.to_bytes_be(), [0]); // Test with leading/trailing zero bytes and a full BigDigit of value 0 let b = BigUint::from_str_radix("00010000000000000200", 16).unwrap(); assert_eq!(b.to_bytes_be(), [1, 0, 0, 0, 0, 0, 0, 2, 0]); } #[test] fn test_from_bytes_le() { fn check(s: &str, result: &str) { assert_eq!( BigUint::from_bytes_le(s.as_bytes()), BigUint::parse_bytes(result.as_bytes(), 10).unwrap() ); } check("A", "65"); check("AA", "16705"); check("BA", "16706"); check("!dlrow olleH", "22405534230753963835153736737"); assert_eq!(BigUint::from_bytes_le(&[]), BigUint::zero()); } #[test] fn test_to_bytes_le() { fn check(s: &str, result: &str) { let b = BigUint::parse_bytes(result.as_bytes(), 10).unwrap(); assert_eq!(b.to_bytes_le(), s.as_bytes()); } check("A", "65"); check("AA", "16705"); check("BA", "16706"); check("!dlrow olleH", "22405534230753963835153736737"); let b: BigUint = Zero::zero(); assert_eq!(b.to_bytes_le(), [0]); // Test with leading/trailing zero bytes and a full BigDigit of value 0 let b = BigUint::from_str_radix("00010000000000000200", 16).unwrap(); assert_eq!(b.to_bytes_le(), [0, 2, 0, 0, 0, 0, 0, 0, 1]); } #[test] fn test_cmp() { let data: [&[_]; 7] = [&[], &[1], &[2], &[!0], &[0, 1], &[2, 1], &[1, 1, 1]]; let data: Vec = data.iter().map(|v| BigUint::from_slice(*v)).collect(); for (i, ni) in data.iter().enumerate() { for (j0, nj) in data[i..].iter().enumerate() { let j = j0 + i; if i == j { assert_eq!(ni.cmp(nj), Equal); assert_eq!(nj.cmp(ni), Equal); assert_eq!(ni, nj); assert!(!(ni != nj)); assert!(ni <= nj); assert!(ni >= nj); assert!(!(ni < nj)); assert!(!(ni > nj)); } else { assert_eq!(ni.cmp(nj), Less); assert_eq!(nj.cmp(ni), Greater); assert!(!(ni == nj)); assert!(ni != nj); assert!(ni <= nj); assert!(!(ni >= nj)); assert!(ni < nj); assert!(!(ni > nj)); assert!(!(nj <= ni)); assert!(nj >= ni); assert!(!(nj < ni)); assert!(nj > ni); } } } } fn hash(x: &T) -> u64 { let mut hasher = ::Hasher::new(); x.hash(&mut hasher); hasher.finish() } #[test] fn test_hash() { use crate::hash; let a = BigUint::new(vec![]); let b = BigUint::new(vec![0]); let c = BigUint::new(vec![1]); let d = BigUint::new(vec![1, 0, 0, 0, 0, 0]); let e = BigUint::new(vec![0, 0, 0, 0, 0, 1]); assert!(hash(&a) == hash(&b)); assert!(hash(&b) != hash(&c)); assert!(hash(&c) == hash(&d)); assert!(hash(&d) != hash(&e)); } // LEFT, RIGHT, AND, OR, XOR const BIT_TESTS: &[(&[u32], &[u32], &[u32], &[u32], &[u32])] = &[ (&[], &[], &[], &[], &[]), (&[1, 0, 1], &[1, 1], &[1], &[1, 1, 1], &[0, 1, 1]), (&[1, 0, 1], &[0, 1, 1], &[0, 0, 1], &[1, 1, 1], &[1, 1]), ( &[268, 482, 17], &[964, 54], &[260, 34], &[972, 502, 17], &[712, 468, 17], ), ]; #[test] fn test_bitand() { for elm in BIT_TESTS { let (a_vec, b_vec, c_vec, _, _) = *elm; let a = BigUint::from_slice(a_vec); let b = BigUint::from_slice(b_vec); let c = BigUint::from_slice(c_vec); assert_op!(a & b == c); assert_op!(b & a == c); assert_assign_op!(a &= b == c); assert_assign_op!(b &= a == c); } } #[test] fn test_bitor() { for elm in BIT_TESTS { let (a_vec, b_vec, _, c_vec, _) = *elm; let a = BigUint::from_slice(a_vec); let b = BigUint::from_slice(b_vec); let c = BigUint::from_slice(c_vec); assert_op!(a | b == c); assert_op!(b | a == c); assert_assign_op!(a |= b == c); assert_assign_op!(b |= a == c); } } #[test] fn test_bitxor() { for elm in BIT_TESTS { let (a_vec, b_vec, _, _, c_vec) = *elm; let a = BigUint::from_slice(a_vec); let b = BigUint::from_slice(b_vec); let c = BigUint::from_slice(c_vec); assert_op!(a ^ b == c); assert_op!(b ^ a == c); assert_op!(a ^ c == b); assert_op!(c ^ a == b); assert_op!(b ^ c == a); assert_op!(c ^ b == a); assert_assign_op!(a ^= b == c); assert_assign_op!(b ^= a == c); assert_assign_op!(a ^= c == b); assert_assign_op!(c ^= a == b); assert_assign_op!(b ^= c == a); assert_assign_op!(c ^= b == a); } } #[test] fn test_shl() { fn check(s: &str, shift: usize, ans: &str) { let opt_biguint = BigUint::from_str_radix(s, 16).ok(); let mut bu_assign = opt_biguint.unwrap(); let bu = (bu_assign.clone() << shift).to_str_radix(16); assert_eq!(bu, ans); bu_assign <<= shift; assert_eq!(bu_assign.to_str_radix(16), ans); } check("0", 3, "0"); check("1", 3, "8"); check( "1\ 0000\ 0000\ 0000\ 0001\ 0000\ 0000\ 0000\ 0001", 3, "8\ 0000\ 0000\ 0000\ 0008\ 0000\ 0000\ 0000\ 0008", ); check( "1\ 0000\ 0001\ 0000\ 0001", 2, "4\ 0000\ 0004\ 0000\ 0004", ); check( "1\ 0001\ 0001", 1, "2\ 0002\ 0002", ); check( "\ 4000\ 0000\ 0000\ 0000", 3, "2\ 0000\ 0000\ 0000\ 0000", ); check( "4000\ 0000", 2, "1\ 0000\ 0000", ); check( "4000", 2, "1\ 0000", ); check( "4000\ 0000\ 0000\ 0000", 67, "2\ 0000\ 0000\ 0000\ 0000\ 0000\ 0000\ 0000\ 0000", ); check( "4000\ 0000", 35, "2\ 0000\ 0000\ 0000\ 0000", ); check( "4000", 19, "2\ 0000\ 0000", ); check( "fedc\ ba98\ 7654\ 3210\ fedc\ ba98\ 7654\ 3210", 4, "f\ edcb\ a987\ 6543\ 210f\ edcb\ a987\ 6543\ 2100", ); check( "88887777666655554444333322221111", 16, "888877776666555544443333222211110000", ); } #[test] fn test_shr() { fn check(s: &str, shift: usize, ans: &str) { let opt_biguint = BigUint::from_str_radix(s, 16).ok(); let mut bu_assign = opt_biguint.unwrap(); let bu = (bu_assign.clone() >> shift).to_str_radix(16); assert_eq!(bu, ans); bu_assign >>= shift; assert_eq!(bu_assign.to_str_radix(16), ans); } check("0", 3, "0"); check("f", 3, "1"); check( "1\ 0000\ 0000\ 0000\ 0001\ 0000\ 0000\ 0000\ 0001", 3, "2000\ 0000\ 0000\ 0000\ 2000\ 0000\ 0000\ 0000", ); check( "1\ 0000\ 0001\ 0000\ 0001", 2, "4000\ 0000\ 4000\ 0000", ); check( "1\ 0001\ 0001", 1, "8000\ 8000", ); check( "2\ 0000\ 0000\ 0000\ 0001\ 0000\ 0000\ 0000\ 0001", 67, "4000\ 0000\ 0000\ 0000", ); check( "2\ 0000\ 0001\ 0000\ 0001", 35, "4000\ 0000", ); check( "2\ 0001\ 0001", 19, "4000", ); check( "1\ 0000\ 0000\ 0000\ 0000", 1, "8000\ 0000\ 0000\ 0000", ); check( "1\ 0000\ 0000", 1, "8000\ 0000", ); check( "1\ 0000", 1, "8000", ); check( "f\ edcb\ a987\ 6543\ 210f\ edcb\ a987\ 6543\ 2100", 4, "fedc\ ba98\ 7654\ 3210\ fedc\ ba98\ 7654\ 3210", ); check( "888877776666555544443333222211110000", 16, "88887777666655554444333322221111", ); } // `DoubleBigDigit` size dependent #[test] fn test_convert_i64() { fn check(b1: BigUint, i: i64) { let b2: BigUint = FromPrimitive::from_i64(i).unwrap(); assert_eq!(b1, b2); assert_eq!(b1.to_i64().unwrap(), i); } check(Zero::zero(), 0); check(One::one(), 1); check(i64::MAX.to_biguint().unwrap(), i64::MAX); check(BigUint::new(vec![]), 0); check(BigUint::new(vec![1]), 1); check(BigUint::new(vec![N1]), (1 << 32) - 1); check(BigUint::new(vec![0, 1]), 1 << 32); check(BigUint::new(vec![N1, N1 >> 1]), i64::MAX); assert_eq!(i64::MIN.to_biguint(), None); assert_eq!(BigUint::new(vec![N1, N1]).to_i64(), None); assert_eq!(BigUint::new(vec![0, 0, 1]).to_i64(), None); assert_eq!(BigUint::new(vec![N1, N1, N1]).to_i64(), None); } #[test] fn test_convert_i128() { fn check(b1: BigUint, i: i128) { let b2: BigUint = FromPrimitive::from_i128(i).unwrap(); assert_eq!(b1, b2); assert_eq!(b1.to_i128().unwrap(), i); } check(Zero::zero(), 0); check(One::one(), 1); check(i128::MAX.to_biguint().unwrap(), i128::MAX); check(BigUint::new(vec![]), 0); check(BigUint::new(vec![1]), 1); check(BigUint::new(vec![N1]), (1 << 32) - 1); check(BigUint::new(vec![0, 1]), 1 << 32); check(BigUint::new(vec![N1, N1, N1, N1 >> 1]), i128::MAX); assert_eq!(i128::MIN.to_biguint(), None); assert_eq!(BigUint::new(vec![N1, N1, N1, N1]).to_i128(), None); assert_eq!(BigUint::new(vec![0, 0, 0, 0, 1]).to_i128(), None); assert_eq!(BigUint::new(vec![N1, N1, N1, N1, N1]).to_i128(), None); } // `DoubleBigDigit` size dependent #[test] fn test_convert_u64() { fn check(b1: BigUint, u: u64) { let b2: BigUint = FromPrimitive::from_u64(u).unwrap(); assert_eq!(b1, b2); assert_eq!(b1.to_u64().unwrap(), u); } check(Zero::zero(), 0); check(One::one(), 1); check(u64::MIN.to_biguint().unwrap(), u64::MIN); check(u64::MAX.to_biguint().unwrap(), u64::MAX); check(BigUint::new(vec![]), 0); check(BigUint::new(vec![1]), 1); check(BigUint::new(vec![N1]), (1 << 32) - 1); check(BigUint::new(vec![0, 1]), 1 << 32); check(BigUint::new(vec![N1, N1]), u64::MAX); assert_eq!(BigUint::new(vec![0, 0, 1]).to_u64(), None); assert_eq!(BigUint::new(vec![N1, N1, N1]).to_u64(), None); } #[test] fn test_convert_u128() { fn check(b1: BigUint, u: u128) { let b2: BigUint = FromPrimitive::from_u128(u).unwrap(); assert_eq!(b1, b2); assert_eq!(b1.to_u128().unwrap(), u); } check(Zero::zero(), 0); check(One::one(), 1); check(u128::MIN.to_biguint().unwrap(), u128::MIN); check(u128::MAX.to_biguint().unwrap(), u128::MAX); check(BigUint::new(vec![]), 0); check(BigUint::new(vec![1]), 1); check(BigUint::new(vec![N1]), (1 << 32) - 1); check(BigUint::new(vec![0, 1]), 1 << 32); check(BigUint::new(vec![N1, N1, N1, N1]), u128::MAX); assert_eq!(BigUint::new(vec![0, 0, 0, 0, 1]).to_u128(), None); assert_eq!(BigUint::new(vec![N1, N1, N1, N1, N1]).to_u128(), None); } #[test] #[allow(clippy::float_cmp)] fn test_convert_f32() { fn check(b1: &BigUint, f: f32) { let b2 = BigUint::from_f32(f).unwrap(); assert_eq!(b1, &b2); assert_eq!(b1.to_f32().unwrap(), f); } check(&BigUint::zero(), 0.0); check(&BigUint::one(), 1.0); check(&BigUint::from(u16::MAX), pow(2.0_f32, 16) - 1.0); check(&BigUint::from(1u64 << 32), pow(2.0_f32, 32)); check(&BigUint::from_slice(&[0, 0, 1]), pow(2.0_f32, 64)); check( &((BigUint::one() << 100) + (BigUint::one() << 123)), pow(2.0_f32, 100) + pow(2.0_f32, 123), ); check(&(BigUint::one() << 127), pow(2.0_f32, 127)); check(&(BigUint::from((1u64 << 24) - 1) << (128 - 24)), f32::MAX); // keeping all 24 digits with the bits at different offsets to the BigDigits let x: u32 = 0b00000000101111011111011011011101; let mut f = x as f32; let mut b = BigUint::from(x); for _ in 0..64 { check(&b, f); f *= 2.0; b <<= 1; } // this number when rounded to f64 then f32 isn't the same as when rounded straight to f32 let n: u64 = 0b0000000000111111111111111111111111011111111111111111111111111111; assert!((n as f64) as f32 != n as f32); assert_eq!(BigUint::from(n).to_f32(), Some(n as f32)); // test rounding up with the bits at different offsets to the BigDigits let mut f = ((1u64 << 25) - 1) as f32; let mut b = BigUint::from(1u64 << 25); for _ in 0..64 { assert_eq!(b.to_f32(), Some(f)); f *= 2.0; b <<= 1; } // rounding assert_eq!(BigUint::from_f32(-1.0), None); assert_eq!(BigUint::from_f32(-0.99999), Some(BigUint::zero())); assert_eq!(BigUint::from_f32(-0.5), Some(BigUint::zero())); assert_eq!(BigUint::from_f32(-0.0), Some(BigUint::zero())); assert_eq!( BigUint::from_f32(f32::MIN_POSITIVE / 2.0), Some(BigUint::zero()) ); assert_eq!(BigUint::from_f32(f32::MIN_POSITIVE), Some(BigUint::zero())); assert_eq!(BigUint::from_f32(0.5), Some(BigUint::zero())); assert_eq!(BigUint::from_f32(0.99999), Some(BigUint::zero())); assert_eq!(BigUint::from_f32(f32::consts::E), Some(BigUint::from(2u32))); assert_eq!( BigUint::from_f32(f32::consts::PI), Some(BigUint::from(3u32)) ); // special float values assert_eq!(BigUint::from_f32(f32::NAN), None); assert_eq!(BigUint::from_f32(f32::INFINITY), None); assert_eq!(BigUint::from_f32(f32::NEG_INFINITY), None); assert_eq!(BigUint::from_f32(f32::MIN), None); // largest BigUint that will round to a finite f32 value let big_num = (BigUint::one() << 128u8) - 1u8 - (BigUint::one() << (128u8 - 25)); assert_eq!(big_num.to_f32(), Some(f32::MAX)); assert_eq!((big_num + 1u8).to_f32(), Some(f32::INFINITY)); assert_eq!( ((BigUint::one() << 128u8) - 1u8).to_f32(), Some(f32::INFINITY) ); assert_eq!((BigUint::one() << 128u8).to_f32(), Some(f32::INFINITY)); } #[test] #[allow(clippy::float_cmp)] fn test_convert_f64() { fn check(b1: &BigUint, f: f64) { let b2 = BigUint::from_f64(f).unwrap(); assert_eq!(b1, &b2); assert_eq!(b1.to_f64().unwrap(), f); } check(&BigUint::zero(), 0.0); check(&BigUint::one(), 1.0); check(&BigUint::from(u32::MAX), pow(2.0_f64, 32) - 1.0); check(&BigUint::from(1u64 << 32), pow(2.0_f64, 32)); check(&BigUint::from_slice(&[0, 0, 1]), pow(2.0_f64, 64)); check( &((BigUint::one() << 100) + (BigUint::one() << 152)), pow(2.0_f64, 100) + pow(2.0_f64, 152), ); check(&(BigUint::one() << 1023), pow(2.0_f64, 1023)); check(&(BigUint::from((1u64 << 53) - 1) << (1024 - 53)), f64::MAX); // keeping all 53 digits with the bits at different offsets to the BigDigits let x: u64 = 0b0000000000011110111110110111111101110111101111011111011011011101; let mut f = x as f64; let mut b = BigUint::from(x); for _ in 0..128 { check(&b, f); f *= 2.0; b <<= 1; } // test rounding up with the bits at different offsets to the BigDigits let mut f = ((1u64 << 54) - 1) as f64; let mut b = BigUint::from(1u64 << 54); for _ in 0..128 { assert_eq!(b.to_f64(), Some(f)); f *= 2.0; b <<= 1; } // rounding assert_eq!(BigUint::from_f64(-1.0), None); assert_eq!(BigUint::from_f64(-0.99999), Some(BigUint::zero())); assert_eq!(BigUint::from_f64(-0.5), Some(BigUint::zero())); assert_eq!(BigUint::from_f64(-0.0), Some(BigUint::zero())); assert_eq!( BigUint::from_f64(f64::MIN_POSITIVE / 2.0), Some(BigUint::zero()) ); assert_eq!(BigUint::from_f64(f64::MIN_POSITIVE), Some(BigUint::zero())); assert_eq!(BigUint::from_f64(0.5), Some(BigUint::zero())); assert_eq!(BigUint::from_f64(0.99999), Some(BigUint::zero())); assert_eq!(BigUint::from_f64(f64::consts::E), Some(BigUint::from(2u32))); assert_eq!( BigUint::from_f64(f64::consts::PI), Some(BigUint::from(3u32)) ); // special float values assert_eq!(BigUint::from_f64(f64::NAN), None); assert_eq!(BigUint::from_f64(f64::INFINITY), None); assert_eq!(BigUint::from_f64(f64::NEG_INFINITY), None); assert_eq!(BigUint::from_f64(f64::MIN), None); // largest BigUint that will round to a finite f64 value let big_num = (BigUint::one() << 1024u16) - 1u8 - (BigUint::one() << (1024u16 - 54)); assert_eq!(big_num.to_f64(), Some(f64::MAX)); assert_eq!((big_num + 1u8).to_f64(), Some(f64::INFINITY)); assert_eq!( ((BigUint::one() << 1024u16) - 1u8).to_f64(), Some(f64::INFINITY) ); assert_eq!((BigUint::one() << 1024u16).to_f64(), Some(f64::INFINITY)); } #[test] fn test_convert_to_bigint() { fn check(n: BigUint, ans: BigInt) { assert_eq!(n.to_bigint().unwrap(), ans); assert_eq!(n.to_bigint().unwrap().to_biguint().unwrap(), n); } check(Zero::zero(), Zero::zero()); check( BigUint::new(vec![1, 2, 3]), BigInt::from_biguint(Plus, BigUint::new(vec![1, 2, 3])), ); } #[test] fn test_convert_from_uint() { macro_rules! check { ($ty:ident, $max:expr) => { assert_eq!(BigUint::from($ty::zero()), BigUint::zero()); assert_eq!(BigUint::from($ty::one()), BigUint::one()); assert_eq!(BigUint::from($ty::MAX - $ty::one()), $max - BigUint::one()); assert_eq!(BigUint::from($ty::MAX), $max); }; } check!(u8, BigUint::from_slice(&[u8::MAX as u32])); check!(u16, BigUint::from_slice(&[u16::MAX as u32])); check!(u32, BigUint::from_slice(&[u32::MAX])); check!(u64, BigUint::from_slice(&[u32::MAX, u32::MAX])); check!( u128, BigUint::from_slice(&[u32::MAX, u32::MAX, u32::MAX, u32::MAX]) ); check!(usize, BigUint::from(usize::MAX as u64)); } #[test] fn test_add() { for elm in SUM_TRIPLES.iter() { let (a_vec, b_vec, c_vec) = *elm; let a = BigUint::from_slice(a_vec); let b = BigUint::from_slice(b_vec); let c = BigUint::from_slice(c_vec); assert_op!(a + b == c); assert_op!(b + a == c); assert_assign_op!(a += b == c); assert_assign_op!(b += a == c); } } #[test] fn test_sub() { for elm in SUM_TRIPLES.iter() { let (a_vec, b_vec, c_vec) = *elm; let a = BigUint::from_slice(a_vec); let b = BigUint::from_slice(b_vec); let c = BigUint::from_slice(c_vec); assert_op!(c - a == b); assert_op!(c - b == a); assert_assign_op!(c -= a == b); assert_assign_op!(c -= b == a); } } #[test] #[should_panic] fn test_sub_fail_on_underflow() { let (a, b): (BigUint, BigUint) = (Zero::zero(), One::one()); let _ = a - b; } #[test] fn test_mul() { for elm in MUL_TRIPLES.iter() { let (a_vec, b_vec, c_vec) = *elm; let a = BigUint::from_slice(a_vec); let b = BigUint::from_slice(b_vec); let c = BigUint::from_slice(c_vec); assert_op!(a * b == c); assert_op!(b * a == c); assert_assign_op!(a *= b == c); assert_assign_op!(b *= a == c); } for elm in DIV_REM_QUADRUPLES.iter() { let (a_vec, b_vec, c_vec, d_vec) = *elm; let a = BigUint::from_slice(a_vec); let b = BigUint::from_slice(b_vec); let c = BigUint::from_slice(c_vec); let d = BigUint::from_slice(d_vec); assert!(a == &b * &c + &d); assert!(a == &c * &b + &d); } } #[test] fn test_div_rem() { for elm in MUL_TRIPLES.iter() { let (a_vec, b_vec, c_vec) = *elm; let a = BigUint::from_slice(a_vec); let b = BigUint::from_slice(b_vec); let c = BigUint::from_slice(c_vec); if !a.is_zero() { assert_op!(c / a == b); assert_op!(c % a == BigUint::zero()); assert_assign_op!(c /= a == b); assert_assign_op!(c %= a == BigUint::zero()); assert_eq!(c.div_rem(&a), (b.clone(), BigUint::zero())); } if !b.is_zero() { assert_op!(c / b == a); assert_op!(c % b == BigUint::zero()); assert_assign_op!(c /= b == a); assert_assign_op!(c %= b == BigUint::zero()); assert_eq!(c.div_rem(&b), (a.clone(), BigUint::zero())); } } for elm in DIV_REM_QUADRUPLES.iter() { let (a_vec, b_vec, c_vec, d_vec) = *elm; let a = BigUint::from_slice(a_vec); let b = BigUint::from_slice(b_vec); let c = BigUint::from_slice(c_vec); let d = BigUint::from_slice(d_vec); if !b.is_zero() { assert_op!(a / b == c); assert_op!(a % b == d); assert_assign_op!(a /= b == c); assert_assign_op!(a %= b == d); assert!(a.div_rem(&b) == (c, d)); } } } #[test] fn test_div_rem_big_multiple() { let a = BigUint::from(3u32).pow(100u32); let a2 = &a * &a; let (div, rem) = a2.div_rem(&a); assert_eq!(div, a); assert!(rem.is_zero()); let (div, rem) = (&a2 - 1u32).div_rem(&a); assert_eq!(div, &a - 1u32); assert_eq!(rem, &a - 1u32); } #[test] fn test_div_ceil() { fn check(a: &BigUint, b: &BigUint, d: &BigUint, m: &BigUint) { if m.is_zero() { assert_eq!(a.div_ceil(b), *d); } else { assert_eq!(a.div_ceil(b), d + 1u32); } } for elm in MUL_TRIPLES.iter() { let (a_vec, b_vec, c_vec) = *elm; let a = BigUint::from_slice(a_vec); let b = BigUint::from_slice(b_vec); let c = BigUint::from_slice(c_vec); if !a.is_zero() { check(&c, &a, &b, &Zero::zero()); } if !b.is_zero() { check(&c, &b, &a, &Zero::zero()); } } for elm in DIV_REM_QUADRUPLES.iter() { let (a_vec, b_vec, c_vec, d_vec) = *elm; let a = BigUint::from_slice(a_vec); let b = BigUint::from_slice(b_vec); let c = BigUint::from_slice(c_vec); let d = BigUint::from_slice(d_vec); if !b.is_zero() { check(&a, &b, &c, &d); } } } #[test] fn test_checked_add() { for elm in SUM_TRIPLES.iter() { let (a_vec, b_vec, c_vec) = *elm; let a = BigUint::from_slice(a_vec); let b = BigUint::from_slice(b_vec); let c = BigUint::from_slice(c_vec); assert!(a.checked_add(&b).unwrap() == c); assert!(b.checked_add(&a).unwrap() == c); } } #[test] fn test_checked_sub() { for elm in SUM_TRIPLES.iter() { let (a_vec, b_vec, c_vec) = *elm; let a = BigUint::from_slice(a_vec); let b = BigUint::from_slice(b_vec); let c = BigUint::from_slice(c_vec); assert!(c.checked_sub(&a).unwrap() == b); assert!(c.checked_sub(&b).unwrap() == a); if a > c { assert!(a.checked_sub(&c).is_none()); } if b > c { assert!(b.checked_sub(&c).is_none()); } } } #[test] fn test_checked_mul() { for elm in MUL_TRIPLES.iter() { let (a_vec, b_vec, c_vec) = *elm; let a = BigUint::from_slice(a_vec); let b = BigUint::from_slice(b_vec); let c = BigUint::from_slice(c_vec); assert!(a.checked_mul(&b).unwrap() == c); assert!(b.checked_mul(&a).unwrap() == c); } for elm in DIV_REM_QUADRUPLES.iter() { let (a_vec, b_vec, c_vec, d_vec) = *elm; let a = BigUint::from_slice(a_vec); let b = BigUint::from_slice(b_vec); let c = BigUint::from_slice(c_vec); let d = BigUint::from_slice(d_vec); assert!(a == b.checked_mul(&c).unwrap() + &d); assert!(a == c.checked_mul(&b).unwrap() + &d); } } #[test] fn test_mul_overflow() { // Test for issue #187 - overflow due to mac3 incorrectly sizing temporary let s = "5311379928167670986895882065524686273295931177270319231994441382\ 0040355986085224273916250223263671004753755210595137000079652876\ 0829212940754539968588340162273730474622005920097370111"; let a: BigUint = s.parse().unwrap(); let b = a.clone(); let _ = a.checked_mul(&b); } #[test] fn test_mul_overflow_2() { // Try a bunch of sizes that are right on the edge of multiplication length // overflow, where (x * x).data.len() == 2 * x.data.len() + 1. for i in 1u8..20 { let bits = 1u32 << i; let x = (BigUint::one() << bits) - 1u32; let x2 = (BigUint::one() << (2 * bits)) - &x - &x - 1u32; assert_eq!(&x * &x, x2); } } #[test] fn test_checked_div() { for elm in MUL_TRIPLES.iter() { let (a_vec, b_vec, c_vec) = *elm; let a = BigUint::from_slice(a_vec); let b = BigUint::from_slice(b_vec); let c = BigUint::from_slice(c_vec); if !a.is_zero() { assert!(c.checked_div(&a).unwrap() == b); } if !b.is_zero() { assert!(c.checked_div(&b).unwrap() == a); } assert!(c.checked_div(&Zero::zero()).is_none()); } } #[test] fn test_gcd() { fn check(a: usize, b: usize, c: usize) { let big_a: BigUint = FromPrimitive::from_usize(a).unwrap(); let big_b: BigUint = FromPrimitive::from_usize(b).unwrap(); let big_c: BigUint = FromPrimitive::from_usize(c).unwrap(); assert_eq!(big_a.gcd(&big_b), big_c); assert_eq!(big_a.gcd_lcm(&big_b).0, big_c); } check(10, 2, 2); check(10, 3, 1); check(0, 3, 3); check(3, 3, 3); check(56, 42, 14); } #[test] fn test_lcm() { fn check(a: usize, b: usize, c: usize) { let big_a: BigUint = FromPrimitive::from_usize(a).unwrap(); let big_b: BigUint = FromPrimitive::from_usize(b).unwrap(); let big_c: BigUint = FromPrimitive::from_usize(c).unwrap(); assert_eq!(big_a.lcm(&big_b), big_c); assert_eq!(big_a.gcd_lcm(&big_b).1, big_c); } check(0, 0, 0); check(1, 0, 0); check(0, 1, 0); check(1, 1, 1); check(8, 9, 72); check(11, 5, 55); check(99, 17, 1683); } #[test] fn test_next_multiple_of() { assert_eq!( BigUint::from(16u32).next_multiple_of(&BigUint::from(8u32)), BigUint::from(16u32) ); assert_eq!( BigUint::from(23u32).next_multiple_of(&BigUint::from(8u32)), BigUint::from(24u32) ); } #[test] fn test_prev_multiple_of() { assert_eq!( BigUint::from(16u32).prev_multiple_of(&BigUint::from(8u32)), BigUint::from(16u32) ); assert_eq!( BigUint::from(23u32).prev_multiple_of(&BigUint::from(8u32)), BigUint::from(16u32) ); } #[test] fn test_is_even() { let one: BigUint = FromStr::from_str("1").unwrap(); let two: BigUint = FromStr::from_str("2").unwrap(); let thousand: BigUint = FromStr::from_str("1000").unwrap(); let big: BigUint = FromStr::from_str("1000000000000000000000").unwrap(); let bigger: BigUint = FromStr::from_str("1000000000000000000001").unwrap(); assert!(one.is_odd()); assert!(two.is_even()); assert!(thousand.is_even()); assert!(big.is_even()); assert!(bigger.is_odd()); assert!((&one << 64u8).is_even()); assert!(((&one << 64u8) + one).is_odd()); } fn to_str_pairs() -> Vec<(BigUint, Vec<(u32, String)>)> { let bits = 32; vec![ ( Zero::zero(), vec![(2, "0".to_string()), (3, "0".to_string())], ), ( BigUint::from_slice(&[0xff]), vec![ (2, "11111111".to_string()), (3, "100110".to_string()), (4, "3333".to_string()), (5, "2010".to_string()), (6, "1103".to_string()), (7, "513".to_string()), (8, "377".to_string()), (9, "313".to_string()), (10, "255".to_string()), (11, "212".to_string()), (12, "193".to_string()), (13, "168".to_string()), (14, "143".to_string()), (15, "120".to_string()), (16, "ff".to_string()), ], ), ( BigUint::from_slice(&[0xfff]), vec![ (2, "111111111111".to_string()), (4, "333333".to_string()), (16, "fff".to_string()), ], ), ( BigUint::from_slice(&[1, 2]), vec![ ( 2, format!("10{}1", repeat("0").take(bits - 1).collect::()), ), ( 4, format!("2{}1", repeat("0").take(bits / 2 - 1).collect::()), ), ( 10, match bits { 64 => "36893488147419103233".to_string(), 32 => "8589934593".to_string(), 16 => "131073".to_string(), _ => panic!(), }, ), ( 16, format!("2{}1", repeat("0").take(bits / 4 - 1).collect::()), ), ], ), ( BigUint::from_slice(&[1, 2, 3]), vec![ ( 2, format!( "11{}10{}1", repeat("0").take(bits - 2).collect::(), repeat("0").take(bits - 1).collect::() ), ), ( 4, format!( "3{}2{}1", repeat("0").take(bits / 2 - 1).collect::(), repeat("0").take(bits / 2 - 1).collect::() ), ), ( 8, match bits { 64 => "14000000000000000000004000000000000000000001".to_string(), 32 => "6000000000100000000001".to_string(), 16 => "140000400001".to_string(), _ => panic!(), }, ), ( 10, match bits { 64 => "1020847100762815390427017310442723737601".to_string(), 32 => "55340232229718589441".to_string(), 16 => "12885032961".to_string(), _ => panic!(), }, ), ( 16, format!( "3{}2{}1", repeat("0").take(bits / 4 - 1).collect::(), repeat("0").take(bits / 4 - 1).collect::() ), ), ], ), ] } #[test] fn test_to_str_radix() { let r = to_str_pairs(); for num_pair in r.iter() { let &(ref n, ref rs) = num_pair; for str_pair in rs.iter() { let &(ref radix, ref str) = str_pair; assert_eq!(n.to_str_radix(*radix), *str); } } } #[test] fn test_from_and_to_radix() { const GROUND_TRUTH: &[(&[u8], u32, &[u8])] = &[ (b"0", 42, &[0]), ( b"ffffeeffbb", 2, &[ 1, 1, 0, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ], ), ( b"ffffeeffbb", 3, &[ 2, 2, 1, 1, 2, 1, 1, 2, 0, 0, 0, 0, 0, 1, 2, 0, 0, 0, 0, 1, 0, 0, 2, 2, 0, 1, ], ), ( b"ffffeeffbb", 4, &[3, 2, 3, 2, 3, 3, 3, 3, 2, 3, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3], ), ( b"ffffeeffbb", 5, &[0, 4, 3, 3, 1, 4, 2, 4, 1, 4, 4, 2, 3, 0, 0, 1, 2, 1], ), ( b"ffffeeffbb", 6, &[5, 5, 4, 5, 5, 0, 0, 1, 2, 5, 3, 0, 1, 0, 2, 2], ), ( b"ffffeeffbb", 7, &[4, 2, 3, 6, 0, 1, 6, 1, 6, 2, 0, 3, 2, 4, 1], ), ( b"ffffeeffbb", 8, &[3, 7, 6, 7, 7, 5, 3, 7, 7, 7, 7, 7, 7, 1], ), (b"ffffeeffbb", 9, &[8, 4, 5, 7, 0, 0, 3, 2, 0, 3, 0, 8, 3]), (b"ffffeeffbb", 10, &[5, 9, 5, 3, 1, 5, 0, 1, 5, 9, 9, 0, 1]), (b"ffffeeffbb", 11, &[10, 7, 6, 5, 2, 0, 3, 3, 3, 4, 9, 3]), (b"ffffeeffbb", 12, &[11, 8, 5, 10, 1, 10, 3, 1, 1, 9, 5, 1]), (b"ffffeeffbb", 13, &[0, 5, 7, 4, 6, 5, 6, 11, 8, 12, 7]), (b"ffffeeffbb", 14, &[11, 4, 4, 11, 8, 4, 6, 0, 3, 11, 3]), (b"ffffeeffbb", 15, &[5, 11, 13, 2, 1, 10, 2, 0, 9, 13, 1]), (b"ffffeeffbb", 16, &[11, 11, 15, 15, 14, 14, 15, 15, 15, 15]), (b"ffffeeffbb", 17, &[0, 2, 14, 12, 2, 14, 8, 10, 4, 9]), (b"ffffeeffbb", 18, &[17, 15, 5, 13, 10, 16, 16, 13, 9, 5]), (b"ffffeeffbb", 19, &[14, 13, 2, 8, 9, 0, 1, 14, 7, 3]), (b"ffffeeffbb", 20, &[15, 19, 3, 14, 0, 17, 19, 18, 2, 2]), (b"ffffeeffbb", 21, &[11, 5, 4, 13, 5, 18, 9, 1, 8, 1]), (b"ffffeeffbb", 22, &[21, 3, 7, 21, 15, 12, 17, 0, 20]), (b"ffffeeffbb", 23, &[21, 21, 6, 9, 10, 7, 21, 0, 14]), (b"ffffeeffbb", 24, &[11, 10, 19, 14, 22, 11, 17, 23, 9]), (b"ffffeeffbb", 25, &[20, 18, 21, 22, 21, 14, 3, 5, 7]), (b"ffffeeffbb", 26, &[13, 15, 24, 11, 17, 6, 23, 6, 5]), (b"ffffeeffbb", 27, &[17, 16, 7, 0, 21, 0, 3, 24, 3]), (b"ffffeeffbb", 28, &[11, 16, 11, 15, 14, 18, 13, 25, 2]), (b"ffffeeffbb", 29, &[6, 8, 7, 19, 14, 13, 21, 5, 2]), (b"ffffeeffbb", 30, &[5, 13, 18, 11, 10, 7, 8, 20, 1]), (b"ffffeeffbb", 31, &[22, 26, 15, 19, 8, 27, 29, 8, 1]), (b"ffffeeffbb", 32, &[27, 29, 31, 29, 30, 31, 31, 31]), (b"ffffeeffbb", 33, &[32, 20, 27, 12, 1, 12, 26, 25]), (b"ffffeeffbb", 34, &[17, 9, 16, 33, 13, 25, 31, 20]), (b"ffffeeffbb", 35, &[25, 32, 2, 25, 11, 4, 3, 17]), (b"ffffeeffbb", 36, &[35, 34, 5, 6, 32, 3, 1, 14]), (b"ffffeeffbb", 37, &[16, 21, 18, 4, 33, 19, 21, 11]), (b"ffffeeffbb", 38, &[33, 25, 19, 29, 20, 6, 23, 9]), (b"ffffeeffbb", 39, &[26, 27, 29, 23, 16, 18, 0, 8]), (b"ffffeeffbb", 40, &[35, 39, 30, 11, 16, 17, 28, 6]), (b"ffffeeffbb", 41, &[36, 30, 9, 18, 12, 19, 26, 5]), (b"ffffeeffbb", 42, &[11, 34, 37, 27, 1, 13, 32, 4]), (b"ffffeeffbb", 43, &[3, 24, 11, 2, 10, 40, 1, 4]), (b"ffffeeffbb", 44, &[43, 12, 40, 32, 3, 23, 19, 3]), (b"ffffeeffbb", 45, &[35, 38, 44, 18, 22, 18, 42, 2]), (b"ffffeeffbb", 46, &[21, 45, 18, 41, 17, 2, 24, 2]), (b"ffffeeffbb", 47, &[37, 37, 11, 12, 6, 0, 8, 2]), (b"ffffeeffbb", 48, &[11, 41, 40, 43, 5, 43, 41, 1]), (b"ffffeeffbb", 49, &[18, 45, 7, 13, 20, 21, 30, 1]), (b"ffffeeffbb", 50, &[45, 21, 5, 34, 21, 18, 20, 1]), (b"ffffeeffbb", 51, &[17, 6, 26, 22, 38, 24, 11, 1]), (b"ffffeeffbb", 52, &[39, 33, 38, 30, 46, 31, 3, 1]), (b"ffffeeffbb", 53, &[31, 7, 44, 23, 9, 32, 49]), (b"ffffeeffbb", 54, &[17, 35, 8, 37, 31, 18, 44]), (b"ffffeeffbb", 55, &[10, 52, 9, 48, 36, 39, 39]), (b"ffffeeffbb", 56, &[11, 50, 51, 22, 25, 36, 35]), (b"ffffeeffbb", 57, &[14, 55, 12, 43, 20, 3, 32]), (b"ffffeeffbb", 58, &[35, 18, 45, 56, 9, 51, 28]), (b"ffffeeffbb", 59, &[51, 28, 20, 26, 55, 3, 26]), (b"ffffeeffbb", 60, &[35, 6, 27, 46, 58, 33, 23]), (b"ffffeeffbb", 61, &[58, 7, 6, 54, 49, 20, 21]), (b"ffffeeffbb", 62, &[53, 59, 3, 14, 10, 22, 19]), (b"ffffeeffbb", 63, &[53, 50, 23, 4, 56, 36, 17]), (b"ffffeeffbb", 64, &[59, 62, 47, 59, 63, 63, 15]), (b"ffffeeffbb", 65, &[0, 53, 39, 4, 40, 37, 14]), (b"ffffeeffbb", 66, &[65, 59, 39, 1, 64, 19, 13]), (b"ffffeeffbb", 67, &[35, 14, 19, 16, 25, 10, 12]), (b"ffffeeffbb", 68, &[51, 38, 63, 50, 15, 8, 11]), (b"ffffeeffbb", 69, &[44, 45, 18, 58, 68, 12, 10]), (b"ffffeeffbb", 70, &[25, 51, 0, 60, 13, 24, 9]), (b"ffffeeffbb", 71, &[54, 30, 9, 65, 28, 41, 8]), (b"ffffeeffbb", 72, &[35, 35, 55, 54, 17, 64, 7]), (b"ffffeeffbb", 73, &[34, 4, 48, 40, 27, 19, 7]), (b"ffffeeffbb", 74, &[53, 47, 4, 56, 36, 51, 6]), (b"ffffeeffbb", 75, &[20, 56, 10, 72, 24, 13, 6]), (b"ffffeeffbb", 76, &[71, 31, 52, 60, 48, 53, 5]), (b"ffffeeffbb", 77, &[32, 73, 14, 63, 15, 21, 5]), (b"ffffeeffbb", 78, &[65, 13, 17, 32, 64, 68, 4]), (b"ffffeeffbb", 79, &[37, 56, 2, 56, 25, 41, 4]), (b"ffffeeffbb", 80, &[75, 59, 37, 41, 43, 15, 4]), (b"ffffeeffbb", 81, &[44, 68, 0, 21, 27, 72, 3]), (b"ffffeeffbb", 82, &[77, 35, 2, 74, 46, 50, 3]), (b"ffffeeffbb", 83, &[52, 51, 19, 76, 10, 30, 3]), (b"ffffeeffbb", 84, &[11, 80, 19, 19, 76, 10, 3]), (b"ffffeeffbb", 85, &[0, 82, 20, 14, 68, 77, 2]), (b"ffffeeffbb", 86, &[3, 12, 78, 37, 62, 61, 2]), (b"ffffeeffbb", 87, &[35, 12, 20, 8, 52, 46, 2]), (b"ffffeeffbb", 88, &[43, 6, 54, 42, 30, 32, 2]), (b"ffffeeffbb", 89, &[49, 52, 85, 21, 80, 18, 2]), (b"ffffeeffbb", 90, &[35, 64, 78, 24, 18, 6, 2]), (b"ffffeeffbb", 91, &[39, 17, 83, 63, 17, 85, 1]), (b"ffffeeffbb", 92, &[67, 22, 85, 79, 75, 74, 1]), (b"ffffeeffbb", 93, &[53, 60, 39, 29, 4, 65, 1]), (b"ffffeeffbb", 94, &[37, 89, 2, 72, 76, 55, 1]), (b"ffffeeffbb", 95, &[90, 74, 89, 9, 9, 47, 1]), (b"ffffeeffbb", 96, &[59, 20, 46, 35, 81, 38, 1]), (b"ffffeeffbb", 97, &[94, 87, 60, 71, 3, 31, 1]), (b"ffffeeffbb", 98, &[67, 22, 63, 50, 62, 23, 1]), (b"ffffeeffbb", 99, &[98, 6, 69, 12, 61, 16, 1]), (b"ffffeeffbb", 100, &[95, 35, 51, 10, 95, 9, 1]), (b"ffffeeffbb", 101, &[87, 27, 7, 8, 62, 3, 1]), (b"ffffeeffbb", 102, &[17, 3, 32, 79, 59, 99]), (b"ffffeeffbb", 103, &[30, 22, 90, 0, 87, 94]), (b"ffffeeffbb", 104, &[91, 68, 87, 68, 38, 90]), (b"ffffeeffbb", 105, &[95, 80, 54, 73, 15, 86]), (b"ffffeeffbb", 106, &[31, 30, 24, 16, 17, 82]), (b"ffffeeffbb", 107, &[51, 50, 10, 12, 42, 78]), (b"ffffeeffbb", 108, &[71, 71, 96, 78, 89, 74]), (b"ffffeeffbb", 109, &[33, 18, 93, 22, 50, 71]), (b"ffffeeffbb", 110, &[65, 53, 57, 88, 29, 68]), (b"ffffeeffbb", 111, &[53, 93, 67, 90, 27, 65]), (b"ffffeeffbb", 112, &[11, 109, 96, 65, 43, 62]), (b"ffffeeffbb", 113, &[27, 23, 106, 56, 76, 59]), (b"ffffeeffbb", 114, &[71, 84, 31, 112, 11, 57]), (b"ffffeeffbb", 115, &[90, 22, 1, 56, 76, 54]), (b"ffffeeffbb", 116, &[35, 38, 98, 57, 40, 52]), (b"ffffeeffbb", 117, &[26, 113, 115, 62, 17, 50]), (b"ffffeeffbb", 118, &[51, 14, 5, 18, 7, 48]), (b"ffffeeffbb", 119, &[102, 31, 110, 108, 8, 46]), (b"ffffeeffbb", 120, &[35, 93, 96, 50, 22, 44]), (b"ffffeeffbb", 121, &[87, 61, 2, 36, 47, 42]), (b"ffffeeffbb", 122, &[119, 64, 1, 22, 83, 40]), (b"ffffeeffbb", 123, &[77, 119, 32, 90, 6, 39]), (b"ffffeeffbb", 124, &[115, 122, 31, 79, 62, 37]), (b"ffffeeffbb", 125, &[95, 108, 47, 74, 3, 36]), (b"ffffeeffbb", 126, &[53, 25, 116, 39, 78, 34]), (b"ffffeeffbb", 127, &[22, 23, 125, 67, 35, 33]), (b"ffffeeffbb", 128, &[59, 127, 59, 127, 127, 31]), (b"ffffeeffbb", 129, &[89, 36, 1, 59, 100, 30]), (b"ffffeeffbb", 130, &[65, 91, 123, 89, 79, 29]), (b"ffffeeffbb", 131, &[58, 72, 39, 63, 65, 28]), (b"ffffeeffbb", 132, &[131, 62, 92, 82, 57, 27]), (b"ffffeeffbb", 133, &[109, 31, 51, 123, 55, 26]), (b"ffffeeffbb", 134, &[35, 74, 21, 27, 60, 25]), (b"ffffeeffbb", 135, &[125, 132, 49, 37, 70, 24]), (b"ffffeeffbb", 136, &[51, 121, 117, 133, 85, 23]), (b"ffffeeffbb", 137, &[113, 60, 135, 22, 107, 22]), (b"ffffeeffbb", 138, &[113, 91, 73, 93, 133, 21]), (b"ffffeeffbb", 139, &[114, 75, 102, 51, 26, 21]), (b"ffffeeffbb", 140, &[95, 25, 35, 16, 62, 20]), (b"ffffeeffbb", 141, &[131, 137, 16, 110, 102, 19]), (b"ffffeeffbb", 142, &[125, 121, 108, 34, 6, 19]), (b"ffffeeffbb", 143, &[65, 78, 138, 55, 55, 18]), (b"ffffeeffbb", 144, &[107, 125, 121, 15, 109, 17]), (b"ffffeeffbb", 145, &[35, 13, 122, 42, 22, 17]), (b"ffffeeffbb", 146, &[107, 38, 103, 123, 83, 16]), (b"ffffeeffbb", 147, &[116, 96, 71, 98, 2, 16]), (b"ffffeeffbb", 148, &[127, 23, 75, 99, 71, 15]), (b"ffffeeffbb", 149, &[136, 110, 53, 114, 144, 14]), (b"ffffeeffbb", 150, &[95, 140, 133, 130, 71, 14]), (b"ffffeeffbb", 151, &[15, 50, 29, 137, 0, 14]), (b"ffffeeffbb", 152, &[147, 15, 89, 121, 83, 13]), (b"ffffeeffbb", 153, &[17, 87, 93, 72, 17, 13]), (b"ffffeeffbb", 154, &[109, 113, 3, 133, 106, 12]), (b"ffffeeffbb", 155, &[115, 141, 120, 139, 44, 12]), (b"ffffeeffbb", 156, &[143, 45, 4, 82, 140, 11]), (b"ffffeeffbb", 157, &[149, 92, 15, 106, 82, 11]), (b"ffffeeffbb", 158, &[37, 107, 79, 46, 26, 11]), (b"ffffeeffbb", 159, &[137, 37, 146, 51, 130, 10]), (b"ffffeeffbb", 160, &[155, 69, 29, 115, 77, 10]), (b"ffffeeffbb", 161, &[67, 98, 46, 68, 26, 10]), (b"ffffeeffbb", 162, &[125, 155, 60, 63, 138, 9]), (b"ffffeeffbb", 163, &[96, 43, 118, 93, 90, 9]), (b"ffffeeffbb", 164, &[159, 99, 123, 152, 43, 9]), (b"ffffeeffbb", 165, &[65, 17, 1, 69, 163, 8]), (b"ffffeeffbb", 166, &[135, 108, 25, 165, 119, 8]), (b"ffffeeffbb", 167, &[165, 116, 164, 103, 77, 8]), (b"ffffeeffbb", 168, &[11, 166, 67, 44, 36, 8]), (b"ffffeeffbb", 169, &[65, 59, 71, 149, 164, 7]), (b"ffffeeffbb", 170, &[85, 83, 26, 76, 126, 7]), (b"ffffeeffbb", 171, &[71, 132, 140, 157, 88, 7]), (b"ffffeeffbb", 172, &[3, 6, 127, 47, 52, 7]), (b"ffffeeffbb", 173, &[122, 66, 53, 83, 16, 7]), (b"ffffeeffbb", 174, &[35, 6, 5, 88, 155, 6]), (b"ffffeeffbb", 175, &[95, 20, 84, 56, 122, 6]), (b"ffffeeffbb", 176, &[43, 91, 57, 159, 89, 6]), (b"ffffeeffbb", 177, &[110, 127, 54, 40, 58, 6]), (b"ffffeeffbb", 178, &[49, 115, 43, 47, 27, 6]), (b"ffffeeffbb", 179, &[130, 91, 4, 178, 175, 5]), (b"ffffeeffbb", 180, &[35, 122, 109, 70, 147, 5]), (b"ffffeeffbb", 181, &[94, 94, 4, 79, 119, 5]), (b"ffffeeffbb", 182, &[39, 54, 66, 19, 92, 5]), (b"ffffeeffbb", 183, &[119, 2, 143, 69, 65, 5]), (b"ffffeeffbb", 184, &[67, 57, 90, 44, 39, 5]), (b"ffffeeffbb", 185, &[90, 63, 141, 123, 13, 5]), (b"ffffeeffbb", 186, &[53, 123, 172, 119, 174, 4]), (b"ffffeeffbb", 187, &[153, 21, 68, 28, 151, 4]), (b"ffffeeffbb", 188, &[131, 138, 94, 32, 128, 4]), (b"ffffeeffbb", 189, &[179, 121, 156, 130, 105, 4]), (b"ffffeeffbb", 190, &[185, 179, 164, 131, 83, 4]), (b"ffffeeffbb", 191, &[118, 123, 37, 31, 62, 4]), (b"ffffeeffbb", 192, &[59, 106, 83, 16, 41, 4]), (b"ffffeeffbb", 193, &[57, 37, 47, 86, 20, 4]), (b"ffffeeffbb", 194, &[191, 140, 63, 45, 0, 4]), (b"ffffeeffbb", 195, &[65, 169, 83, 84, 175, 3]), (b"ffffeeffbb", 196, &[67, 158, 64, 6, 157, 3]), (b"ffffeeffbb", 197, &[121, 26, 167, 3, 139, 3]), (b"ffffeeffbb", 198, &[197, 151, 165, 75, 121, 3]), (b"ffffeeffbb", 199, &[55, 175, 36, 22, 104, 3]), (b"ffffeeffbb", 200, &[195, 167, 162, 38, 87, 3]), (b"ffffeeffbb", 201, &[35, 27, 136, 124, 70, 3]), (b"ffffeeffbb", 202, &[87, 64, 153, 76, 54, 3]), (b"ffffeeffbb", 203, &[151, 191, 14, 94, 38, 3]), (b"ffffeeffbb", 204, &[119, 103, 135, 175, 22, 3]), (b"ffffeeffbb", 205, &[200, 79, 123, 115, 7, 3]), (b"ffffeeffbb", 206, &[133, 165, 202, 115, 198, 2]), (b"ffffeeffbb", 207, &[44, 153, 193, 175, 184, 2]), (b"ffffeeffbb", 208, &[91, 190, 125, 86, 171, 2]), (b"ffffeeffbb", 209, &[109, 151, 34, 53, 158, 2]), (b"ffffeeffbb", 210, &[95, 40, 171, 74, 145, 2]), (b"ffffeeffbb", 211, &[84, 195, 162, 150, 132, 2]), (b"ffffeeffbb", 212, &[31, 15, 59, 68, 120, 2]), (b"ffffeeffbb", 213, &[125, 57, 127, 36, 108, 2]), (b"ffffeeffbb", 214, &[51, 132, 2, 55, 96, 2]), (b"ffffeeffbb", 215, &[175, 133, 177, 122, 84, 2]), (b"ffffeeffbb", 216, &[179, 35, 78, 23, 73, 2]), (b"ffffeeffbb", 217, &[53, 101, 208, 186, 61, 2]), (b"ffffeeffbb", 218, &[33, 9, 214, 179, 50, 2]), (b"ffffeeffbb", 219, &[107, 147, 175, 217, 39, 2]), (b"ffffeeffbb", 220, &[175, 81, 179, 79, 29, 2]), (b"ffffeeffbb", 221, &[0, 76, 95, 204, 18, 2]), (b"ffffeeffbb", 222, &[53, 213, 16, 150, 8, 2]), (b"ffffeeffbb", 223, &[158, 161, 42, 136, 221, 1]), (b"ffffeeffbb", 224, &[123, 54, 52, 162, 212, 1]), (b"ffffeeffbb", 225, &[170, 43, 151, 2, 204, 1]), (b"ffffeeffbb", 226, &[27, 68, 224, 105, 195, 1]), (b"ffffeeffbb", 227, &[45, 69, 157, 20, 187, 1]), (b"ffffeeffbb", 228, &[71, 213, 64, 199, 178, 1]), (b"ffffeeffbb", 229, &[129, 203, 66, 186, 170, 1]), (b"ffffeeffbb", 230, &[205, 183, 57, 208, 162, 1]), (b"ffffeeffbb", 231, &[32, 50, 164, 33, 155, 1]), (b"ffffeeffbb", 232, &[35, 135, 53, 123, 147, 1]), (b"ffffeeffbb", 233, &[209, 47, 89, 13, 140, 1]), (b"ffffeeffbb", 234, &[143, 56, 175, 168, 132, 1]), (b"ffffeeffbb", 235, &[225, 157, 216, 121, 125, 1]), (b"ffffeeffbb", 236, &[51, 66, 119, 105, 118, 1]), (b"ffffeeffbb", 237, &[116, 150, 26, 119, 111, 1]), (b"ffffeeffbb", 238, &[221, 15, 87, 162, 104, 1]), (b"ffffeeffbb", 239, &[234, 155, 214, 234, 97, 1]), (b"ffffeeffbb", 240, &[155, 46, 84, 96, 91, 1]), (b"ffffeeffbb", 241, &[187, 48, 90, 225, 84, 1]), (b"ffffeeffbb", 242, &[87, 212, 151, 140, 78, 1]), (b"ffffeeffbb", 243, &[206, 22, 189, 81, 72, 1]), (b"ffffeeffbb", 244, &[119, 93, 122, 48, 66, 1]), (b"ffffeeffbb", 245, &[165, 224, 117, 40, 60, 1]), (b"ffffeeffbb", 246, &[77, 121, 100, 57, 54, 1]), (b"ffffeeffbb", 247, &[52, 128, 242, 98, 48, 1]), (b"ffffeeffbb", 248, &[115, 247, 224, 164, 42, 1]), (b"ffffeeffbb", 249, &[218, 127, 223, 5, 37, 1]), (b"ffffeeffbb", 250, &[95, 54, 168, 118, 31, 1]), (b"ffffeeffbb", 251, &[121, 204, 240, 3, 26, 1]), (b"ffffeeffbb", 252, &[179, 138, 123, 162, 20, 1]), (b"ffffeeffbb", 253, &[21, 50, 1, 91, 15, 1]), (b"ffffeeffbb", 254, &[149, 11, 63, 40, 10, 1]), (b"ffffeeffbb", 255, &[170, 225, 247, 9, 5, 1]), (b"ffffeeffbb", 256, &[187, 255, 238, 255, 255]), ]; for &(bigint, radix, inbaseradix_le) in GROUND_TRUTH.iter() { let bigint = BigUint::parse_bytes(bigint, 16).unwrap(); // to_radix_le assert_eq!(bigint.to_radix_le(radix), inbaseradix_le); // to_radix_be let mut inbase_be = bigint.to_radix_be(radix); inbase_be.reverse(); // now le assert_eq!(inbase_be, inbaseradix_le); // from_radix_le assert_eq!( BigUint::from_radix_le(inbaseradix_le, radix).unwrap(), bigint ); // from_radix_be let mut inbaseradix_be = Vec::from(inbaseradix_le); inbaseradix_be.reverse(); assert_eq!( BigUint::from_radix_be(&inbaseradix_be, radix).unwrap(), bigint ); } assert!(BigUint::from_radix_le(&[10, 100, 10], 50).is_none()); assert_eq!(BigUint::from_radix_le(&[], 2), Some(BigUint::zero())); assert_eq!(BigUint::from_radix_be(&[], 2), Some(BigUint::zero())); } #[test] fn test_from_str_radix() { let r = to_str_pairs(); for num_pair in r.iter() { let &(ref n, ref rs) = num_pair; for str_pair in rs.iter() { let &(ref radix, ref str) = str_pair; assert_eq!(n, &BigUint::from_str_radix(str, *radix).unwrap()); } } let zed = BigUint::from_str_radix("Z", 10).ok(); assert_eq!(zed, None); let blank = BigUint::from_str_radix("_", 2).ok(); assert_eq!(blank, None); let blank_one = BigUint::from_str_radix("_1", 2).ok(); assert_eq!(blank_one, None); let plus_one = BigUint::from_str_radix("+1", 10).ok(); assert_eq!(plus_one, Some(BigUint::from_slice(&[1]))); let plus_plus_one = BigUint::from_str_radix("++1", 10).ok(); assert_eq!(plus_plus_one, None); let minus_one = BigUint::from_str_radix("-1", 10).ok(); assert_eq!(minus_one, None); let zero_plus_two = BigUint::from_str_radix("0+2", 10).ok(); assert_eq!(zero_plus_two, None); let three = BigUint::from_str_radix("1_1", 2).ok(); assert_eq!(three, Some(BigUint::from_slice(&[3]))); let ff = BigUint::from_str_radix("1111_1111", 2).ok(); assert_eq!(ff, Some(BigUint::from_slice(&[0xff]))); } #[test] fn test_all_str_radix() { let n = BigUint::new((0..10).collect()); for radix in 2..37 { let s = n.to_str_radix(radix); let x = BigUint::from_str_radix(&s, radix); assert_eq!(x.unwrap(), n); let s = s.to_ascii_uppercase(); let x = BigUint::from_str_radix(&s, radix); assert_eq!(x.unwrap(), n); } } #[test] fn test_big_str() { for n in 2..=20_u32 { let x: BigUint = BigUint::from(n).pow(10_000_u32); let s = x.to_string(); let y: BigUint = s.parse().unwrap(); assert_eq!(x, y); } } #[test] fn test_lower_hex() { let a = BigUint::parse_bytes(b"A", 16).unwrap(); let hello = BigUint::parse_bytes(b"22405534230753963835153736737", 10).unwrap(); assert_eq!(format!("{:x}", a), "a"); assert_eq!(format!("{:x}", hello), "48656c6c6f20776f726c6421"); assert_eq!(format!("{:♥>+#8x}", a), "♥♥♥♥+0xa"); } #[test] fn test_upper_hex() { let a = BigUint::parse_bytes(b"A", 16).unwrap(); let hello = BigUint::parse_bytes(b"22405534230753963835153736737", 10).unwrap(); assert_eq!(format!("{:X}", a), "A"); assert_eq!(format!("{:X}", hello), "48656C6C6F20776F726C6421"); assert_eq!(format!("{:♥>+#8X}", a), "♥♥♥♥+0xA"); } #[test] fn test_binary() { let a = BigUint::parse_bytes(b"A", 16).unwrap(); let hello = BigUint::parse_bytes(b"224055342307539", 10).unwrap(); assert_eq!(format!("{:b}", a), "1010"); assert_eq!( format!("{:b}", hello), "110010111100011011110011000101101001100011010011" ); assert_eq!(format!("{:♥>+#8b}", a), "♥+0b1010"); } #[test] fn test_octal() { let a = BigUint::parse_bytes(b"A", 16).unwrap(); let hello = BigUint::parse_bytes(b"22405534230753963835153736737", 10).unwrap(); assert_eq!(format!("{:o}", a), "12"); assert_eq!(format!("{:o}", hello), "22062554330674403566756233062041"); assert_eq!(format!("{:♥>+#8o}", a), "♥♥♥+0o12"); } #[test] fn test_display() { let a = BigUint::parse_bytes(b"A", 16).unwrap(); let hello = BigUint::parse_bytes(b"22405534230753963835153736737", 10).unwrap(); assert_eq!(format!("{}", a), "10"); assert_eq!(format!("{}", hello), "22405534230753963835153736737"); assert_eq!(format!("{:♥>+#8}", a), "♥♥♥♥♥+10"); } #[test] fn test_factor() { fn factor(n: usize) -> BigUint { let mut f: BigUint = One::one(); for i in 2..=n { // FIXME(#5992): assignment operator overloads // f *= FromPrimitive::from_usize(i); let bu: BigUint = FromPrimitive::from_usize(i).unwrap(); f *= bu; } f } fn check(n: usize, s: &str) { let n = factor(n); let ans = BigUint::from_str_radix(s, 10).unwrap(); assert_eq!(n, ans); } check(3, "6"); check(10, "3628800"); check(20, "2432902008176640000"); check(30, "265252859812191058636308480000000"); } #[test] fn test_bits() { assert_eq!(BigUint::new(vec![0, 0, 0, 0]).bits(), 0); let n: BigUint = FromPrimitive::from_usize(0).unwrap(); assert_eq!(n.bits(), 0); let n: BigUint = FromPrimitive::from_usize(1).unwrap(); assert_eq!(n.bits(), 1); let n: BigUint = FromPrimitive::from_usize(3).unwrap(); assert_eq!(n.bits(), 2); let n: BigUint = BigUint::from_str_radix("4000000000", 16).unwrap(); assert_eq!(n.bits(), 39); let one: BigUint = One::one(); assert_eq!((one << 426u16).bits(), 427); } #[test] fn test_iter_sum() { let result: BigUint = FromPrimitive::from_isize(1234567).unwrap(); let data: Vec = vec![ FromPrimitive::from_u32(1000000).unwrap(), FromPrimitive::from_u32(200000).unwrap(), FromPrimitive::from_u32(30000).unwrap(), FromPrimitive::from_u32(4000).unwrap(), FromPrimitive::from_u32(500).unwrap(), FromPrimitive::from_u32(60).unwrap(), FromPrimitive::from_u32(7).unwrap(), ]; assert_eq!(result, data.iter().sum::()); assert_eq!(result, data.into_iter().sum::()); } #[test] fn test_iter_product() { let data: Vec = vec![ FromPrimitive::from_u32(1001).unwrap(), FromPrimitive::from_u32(1002).unwrap(), FromPrimitive::from_u32(1003).unwrap(), FromPrimitive::from_u32(1004).unwrap(), FromPrimitive::from_u32(1005).unwrap(), ]; let result = data.get(0).unwrap() * data.get(1).unwrap() * data.get(2).unwrap() * data.get(3).unwrap() * data.get(4).unwrap(); assert_eq!(result, data.iter().product::()); assert_eq!(result, data.into_iter().product::()); } #[test] fn test_iter_sum_generic() { let result: BigUint = FromPrimitive::from_isize(1234567).unwrap(); let data = vec![1000000_u32, 200000, 30000, 4000, 500, 60, 7]; assert_eq!(result, data.iter().sum::()); assert_eq!(result, data.into_iter().sum::()); } #[test] fn test_iter_product_generic() { let data = vec![1001_u32, 1002, 1003, 1004, 1005]; let result = data[0].to_biguint().unwrap() * data[1].to_biguint().unwrap() * data[2].to_biguint().unwrap() * data[3].to_biguint().unwrap() * data[4].to_biguint().unwrap(); assert_eq!(result, data.iter().product::()); assert_eq!(result, data.into_iter().product::()); } #[test] fn test_pow() { let one = BigUint::from(1u32); let two = BigUint::from(2u32); let four = BigUint::from(4u32); let eight = BigUint::from(8u32); let tentwentyfour = BigUint::from(1024u32); let twentyfourtyeight = BigUint::from(2048u32); macro_rules! check { ($t:ty) => { assert_eq!(Pow::pow(&two, 0 as $t), one); assert_eq!(Pow::pow(&two, 1 as $t), two); assert_eq!(Pow::pow(&two, 2 as $t), four); assert_eq!(Pow::pow(&two, 3 as $t), eight); assert_eq!(Pow::pow(&two, 10 as $t), tentwentyfour); assert_eq!(Pow::pow(&two, 11 as $t), twentyfourtyeight); assert_eq!(Pow::pow(&two, &(11 as $t)), twentyfourtyeight); }; } check!(u8); check!(u16); check!(u32); check!(u64); check!(u128); check!(usize); let pow_1e10000 = BigUint::from(10u32).pow(10_000_u32); let manual_1e10000 = repeat(10u32).take(10_000).product::(); assert!(manual_1e10000 == pow_1e10000); } #[test] fn test_trailing_zeros() { assert!(BigUint::from(0u8).trailing_zeros().is_none()); assert_eq!(BigUint::from(1u8).trailing_zeros().unwrap(), 0); assert_eq!(BigUint::from(2u8).trailing_zeros().unwrap(), 1); let x: BigUint = BigUint::one() << 128; assert_eq!(x.trailing_zeros().unwrap(), 128); } #[test] fn test_trailing_ones() { assert_eq!(BigUint::from(0u8).trailing_ones(), 0); assert_eq!(BigUint::from(1u8).trailing_ones(), 1); assert_eq!(BigUint::from(2u8).trailing_ones(), 0); assert_eq!(BigUint::from(3u8).trailing_ones(), 2); let x: BigUint = (BigUint::from(3u8) << 128) | BigUint::from(3u8); assert_eq!(x.trailing_ones(), 2); let x: BigUint = (BigUint::one() << 128) - BigUint::one(); assert_eq!(x.trailing_ones(), 128); } #[test] fn test_count_ones() { assert_eq!(BigUint::from(0u8).count_ones(), 0); assert_eq!(BigUint::from(1u8).count_ones(), 1); assert_eq!(BigUint::from(2u8).count_ones(), 1); assert_eq!(BigUint::from(3u8).count_ones(), 2); let x: BigUint = (BigUint::from(3u8) << 128) | BigUint::from(3u8); assert_eq!(x.count_ones(), 4); } #[test] fn test_bit() { assert!(!BigUint::from(0u8).bit(0)); assert!(!BigUint::from(0u8).bit(100)); assert!(!BigUint::from(42u8).bit(4)); assert!(BigUint::from(42u8).bit(5)); let x: BigUint = (BigUint::from(3u8) << 128) | BigUint::from(3u8); assert!(x.bit(129)); assert!(!x.bit(130)); } #[test] fn test_set_bit() { let mut x = BigUint::from(3u8); x.set_bit(128, true); x.set_bit(129, true); assert_eq!(x, (BigUint::from(3u8) << 128) | BigUint::from(3u8)); x.set_bit(0, false); x.set_bit(128, false); x.set_bit(130, false); assert_eq!(x, (BigUint::from(2u8) << 128) | BigUint::from(2u8)); x.set_bit(129, false); x.set_bit(1, false); assert_eq!(x, BigUint::zero()); }