Cleanup bigint

This commit is contained in:
Adolfo Ochagavía 2014-06-06 10:56:03 +02:00
parent 3ca00ec0db
commit 0eb858b4c9

View File

@ -17,18 +17,16 @@
*/
use Integer;
use rand::Rng;
use std::cmp;
use std::{cmp, fmt};
use std::default::Default;
use std::fmt;
use std::from_str::FromStr;
use std::num::CheckedDiv;
use std::num::{Bitwise, ToPrimitive, FromPrimitive};
use std::num::{Zero, One, ToStrRadix, FromStrRadix};
use rand::Rng;
use std::string::String;
use std::uint;
use std::{i64, u64};
use std::{uint, i64, u64};
/**
A `BigDigit` is a `BigUint`'s composing element.
@ -94,7 +92,7 @@ impl Eq for BigUint {}
impl PartialOrd for BigUint {
#[inline]
fn lt(&self, other: &BigUint) -> bool {
match self.cmp(other) { Less => true, _ => false}
self.cmp(other) == Less
}
}
@ -115,7 +113,7 @@ fn cmp(&self, other: &BigUint) -> Ordering {
impl Default for BigUint {
#[inline]
fn default() -> BigUint { BigUint::new(Vec::new()) }
fn default() -> BigUint { Zero::zero() }
}
impl fmt::Show for BigUint {
@ -605,7 +603,7 @@ fn to_biguint(&self) -> Option<BigUint> {
impl ToStrRadix for BigUint {
fn to_str_radix(&self, radix: uint) -> String {
assert!(1 < radix && radix <= 16);
assert!(1 < radix && radix <= 16, "The radix must be within (1, 16]");
let (base, max_len) = get_radix_base(radix);
if base == BigDigit::base {
return fill_concat(self.data.as_slice(), radix, max_len)
@ -645,8 +643,7 @@ fn fill_concat(v: &[BigDigit], radix: uint, l: uint) -> String {
impl FromStrRadix for BigUint {
/// Creates and initializes a `BigUint`.
#[inline]
fn from_str_radix(s: &str, radix: uint)
-> Option<BigUint> {
fn from_str_radix(s: &str, radix: uint) -> Option<BigUint> {
BigUint::parse_bytes(s.as_bytes(), radix)
}
}
@ -656,14 +653,11 @@ impl BigUint {
///
/// The digits are be in base 2^32.
#[inline]
pub fn new(v: Vec<BigDigit>) -> BigUint {
pub fn new(mut digits: Vec<BigDigit>) -> BigUint {
// omit trailing zeros
let new_len = v.iter().rposition(|n| *n != 0).map_or(0, |p| p + 1);
if new_len == v.len() { return BigUint { data: v }; }
let mut v = v;
v.truncate(new_len);
return BigUint { data: v };
let new_len = digits.iter().rposition(|n| *n != 0).map_or(0, |p| p + 1);
digits.truncate(new_len);
BigUint { data: digits }
}
/// Creates and initializes a `BigUint`.
@ -671,7 +665,7 @@ pub fn new(v: Vec<BigDigit>) -> BigUint {
/// The digits are be in base 2^32.
#[inline]
pub fn from_slice(slice: &[BigDigit]) -> BigUint {
return BigUint::new(Vec::from_slice(slice));
BigUint::new(Vec::from_slice(slice))
}
/// Creates and initializes a `BigUint`.
@ -768,7 +762,6 @@ pub fn bits(&self) -> uint {
// `DoubleBigDigit` size dependent
#[inline]
fn get_radix_base(radix: uint) -> (DoubleBigDigit, uint) {
assert!(1 < radix && radix <= 16);
match radix {
2 => (4294967296, 32),
3 => (3486784401, 20),
@ -785,7 +778,7 @@ fn get_radix_base(radix: uint) -> (DoubleBigDigit, uint) {
14 => (1475789056, 8),
15 => (2562890625, 8),
16 => (4294967296, 8),
_ => fail!()
_ => fail!("The radix must be within (1, 16]")
}
}
@ -815,7 +808,7 @@ pub struct BigInt {
impl PartialEq for BigInt {
#[inline]
fn eq(&self, other: &BigInt) -> bool {
match self.cmp(other) { Equal => true, _ => false }
self.cmp(other) == Equal
}
}
@ -824,7 +817,7 @@ impl Eq for BigInt {}
impl PartialOrd for BigInt {
#[inline]
fn lt(&self, other: &BigInt) -> bool {
match self.cmp(other) { Less => true, _ => false}
self.cmp(other) == Less
}
}
@ -844,7 +837,7 @@ fn cmp(&self, other: &BigInt) -> Ordering {
impl Default for BigInt {
#[inline]
fn default() -> BigInt { BigInt::new(Zero, Vec::new()) }
fn default() -> BigInt { Zero::zero() }
}
impl fmt::Show for BigInt {
@ -929,8 +922,7 @@ fn add(&self, other: &BigInt) -> BigInt {
match (self.sign, other.sign) {
(Zero, _) => other.clone(),
(_, Zero) => self.clone(),
(Plus, Plus) => BigInt::from_biguint(Plus,
self.data + other.data),
(Plus, Plus) => BigInt::from_biguint(Plus, self.data + other.data),
(Plus, Minus) => self - (-*other),
(Minus, Plus) => other - (-*self),
(Minus, Minus) => -((-self) + (-*other))
@ -975,7 +967,7 @@ impl Div<BigInt, BigInt> for BigInt {
#[inline]
fn div(&self, other: &BigInt) -> BigInt {
let (q, _) = self.div_rem(other);
return q;
q
}
}
@ -983,7 +975,7 @@ impl Rem<BigInt, BigInt> for BigInt {
#[inline]
fn rem(&self, other: &BigInt) -> BigInt {
let (_, r) = self.div_rem(other);
return r;
r
}
}
@ -1045,13 +1037,13 @@ fn div_rem(&self, other: &BigInt) -> (BigInt, BigInt) {
#[inline]
fn div_floor(&self, other: &BigInt) -> BigInt {
let (d, _) = self.div_mod_floor(other);
return d;
d
}
#[inline]
fn mod_floor(&self, other: &BigInt) -> BigInt {
let (_, m) = self.div_mod_floor(other);
return m;
m
}
fn div_mod_floor(&self, other: &BigInt) -> (BigInt, BigInt) {
@ -1265,7 +1257,7 @@ fn gen_biguint(&mut self, bit_size: uint) -> BigUint {
let final_digit: BigDigit = self.gen();
data.push(final_digit >> (BigDigit::bits - rem));
}
return BigUint::new(data);
BigUint::new(data)
}
fn gen_bigint(&mut self, bit_size: uint) -> BigInt {
@ -1287,7 +1279,7 @@ fn gen_bigint(&mut self, bit_size: uint) -> BigInt {
} else {
Minus
};
return BigInt::from_biguint(sign, biguint);
BigInt::from_biguint(sign, biguint)
}
fn gen_biguint_below(&mut self, bound: &BigUint) -> BigUint {
@ -1322,8 +1314,8 @@ impl BigInt {
///
/// The digits are be in base 2^32.
#[inline]
pub fn new(sign: Sign, v: Vec<BigDigit>) -> BigInt {
BigInt::from_biguint(sign, BigUint::new(v))
pub fn new(sign: Sign, digits: Vec<BigDigit>) -> BigInt {
BigInt::from_biguint(sign, BigUint::new(digits))
}
/// Creates and initializes a `BigInt`.
@ -1334,7 +1326,7 @@ pub fn from_biguint(sign: Sign, data: BigUint) -> BigInt {
if sign == Zero || data.is_zero() {
return BigInt { sign: Zero, data: Zero::zero() };
}
return BigInt { sign: sign, data: data };
BigInt { sign: sign, data: data }
}
/// Creates and initializes a `BigInt`.
@ -1344,8 +1336,7 @@ pub fn from_slice(sign: Sign, slice: &[BigDigit]) -> BigInt {
}
/// Creates and initializes a `BigInt`.
pub fn parse_bytes(buf: &[u8], radix: uint)
-> Option<BigInt> {
pub fn parse_bytes(buf: &[u8], radix: uint) -> Option<BigInt> {
if buf.is_empty() { return None; }
let mut sign = Plus;
let mut start = 0;