#![allow(clippy::unreadable_literal)]
#![allow(clippy::needless_return)]
#![allow(clippy::suspicious_arithmetic_impl)]
#![allow(clippy::suspicious_op_assign_impl)]
#![allow(clippy::redundant_field_names)]
pub extern crate num_bigint;
pub extern crate num_traits;
extern crate num_integer;
#[cfg(feature = "serde")]
extern crate serde;
use std::cmp::Ordering;
use std::convert::TryFrom;
use std::default::Default;
use std::error::Error;
use std::fmt;
use std::hash::{Hash, Hasher};
use std::num::{ParseFloatError, ParseIntError};
use std::ops::{Add, AddAssign, Div, Mul, MulAssign, Neg, Rem, Sub, SubAssign};
use std::iter::Sum;
use std::str::{self, FromStr};
use num_bigint::{BigInt, ParseBigIntError, Sign, ToBigInt};
use num_integer::Integer as IntegerTrait;
pub use num_traits::{FromPrimitive, Num, One, Signed, ToPrimitive, Zero};
const LOG2_10: f64 = 3.321928094887362_f64;
#[macro_use]
mod macros;
#[inline(always)]
fn ten_to_the(pow: u64) -> BigInt {
if pow < 20 {
BigInt::from(10u64.pow(pow as u32))
} else {
let (half, rem) = pow.div_rem(&16);
let mut x = ten_to_the(half);
for _ in 0..4 {
x = &x * &x;
}
if rem == 0 {
x
} else {
x * ten_to_the(rem)
}
}
}
#[inline(always)]
fn count_decimal_digits(int: &BigInt) -> u64 {
if int.is_zero() {
return 1;
}
let mut digits = (int.bits() as f64 / LOG2_10) as u64;
let mut num = ten_to_the(digits);
while *int >= num {
num *= 10u8;
digits += 1;
}
digits
}
#[inline(always)]
fn get_rounding_term(num: &BigInt) -> u8 {
if num.is_zero() {
return 0;
}
let digits = (num.bits() as f64 / LOG2_10) as u64;
let mut n = ten_to_the(digits);
loop {
if *num < n {
return 1;
}
n *= 5;
if *num < n {
return 0;
}
n *= 2;
}
}
#[derive(Clone, Eq)]
pub struct BigDecimal {
int_val: BigInt,
scale: i64,
}
impl BigDecimal {
#[inline]
pub fn new(digits: BigInt, scale: i64) -> BigDecimal {
BigDecimal {
int_val: digits,
scale: scale,
}
}
#[inline]
pub fn parse_bytes(buf: &[u8], radix: u32) -> Option<BigDecimal> {
str::from_utf8(buf)
.ok()
.and_then(|s| BigDecimal::from_str_radix(s, radix).ok())
}
#[inline]
pub fn with_scale(&self, new_scale: i64) -> BigDecimal {
if self.int_val.is_zero() {
return BigDecimal::new(BigInt::zero(), new_scale);
}
if new_scale > self.scale {
let scale_diff = new_scale - self.scale;
let int_val = &self.int_val * ten_to_the(scale_diff as u64);
BigDecimal::new(int_val, new_scale)
} else if new_scale < self.scale {
let scale_diff = self.scale - new_scale;
let int_val = &self.int_val / ten_to_the(scale_diff as u64);
BigDecimal::new(int_val, new_scale)
} else {
self.clone()
}
}
#[inline(always)]
fn take_and_scale(mut self, new_scale: i64) -> BigDecimal {
if self.int_val.is_zero() {
return BigDecimal::new(BigInt::zero(), new_scale);
}
if new_scale > self.scale {
self.int_val *= ten_to_the((new_scale - self.scale) as u64);
BigDecimal::new(self.int_val, new_scale)
} else if new_scale < self.scale {
self.int_val /= ten_to_the((self.scale - new_scale) as u64);
BigDecimal::new(self.int_val, new_scale)
} else {
self
}
}
#[inline]
pub fn with_prec(&self, prec: u64) -> BigDecimal {
let digits = self.digits();
if digits > prec {
let diff = digits - prec;
let p = ten_to_the(diff);
let (mut q, r) = self.int_val.div_rem(&p);
if p < 10 * &r {
q += get_rounding_term(&r);
}
BigDecimal {
int_val: q,
scale: self.scale - diff as i64,
}
} else if digits < prec {
let diff = prec - digits;
BigDecimal {
int_val: &self.int_val * ten_to_the(diff),
scale: self.scale + diff as i64,
}
} else {
self.clone()
}
}
#[inline]
pub fn sign(&self) -> num_bigint::Sign {
self.int_val.sign()
}
#[inline]
pub fn as_bigint_and_exponent(&self) -> (BigInt, i64) {
(self.int_val.clone(), self.scale)
}
#[inline]
pub fn into_bigint_and_exponent(self) -> (BigInt, i64) {
(self.int_val, self.scale)
}
#[inline]
pub fn digits(&self) -> u64 {
count_decimal_digits(&self.int_val)
}
#[inline]
pub fn abs(&self) -> BigDecimal {
BigDecimal {
int_val: self.int_val.abs(),
scale: self.scale,
}
}
#[inline]
pub fn double(&self) -> BigDecimal {
if self.is_zero() {
self.clone()
} else {
BigDecimal {
int_val: self.int_val.clone() * 2,
scale: self.scale,
}
}
}
#[inline]
pub fn half(&self) -> BigDecimal {
if self.is_zero() {
self.clone()
} else if self.int_val.is_even() {
BigDecimal {
int_val: self.int_val.clone().div(2u8),
scale: self.scale,
}
} else {
BigDecimal {
int_val: self.int_val.clone().mul(5u8),
scale: self.scale + 1,
}
}
}
#[inline]
pub fn square(&self) -> BigDecimal {
if self.is_zero() || self.is_one() {
self.clone()
} else {
BigDecimal {
int_val: self.int_val.clone() * &self.int_val,
scale: self.scale * 2,
}
}
}
#[inline]
pub fn cube(&self) -> BigDecimal {
if self.is_zero() || self.is_one() {
self.clone()
} else {
BigDecimal {
int_val: self.int_val.clone() * &self.int_val * &self.int_val,
scale: self.scale * 3,
}
}
}
#[inline]
pub fn sqrt(&self) -> Option<BigDecimal> {
if self.is_zero() || self.is_one() {
return Some(self.clone());
}
if self.is_negative() {
return None;
}
let guess = {
let magic_guess_scale = 1.1951678538495576_f64;
let initial_guess = (self.int_val.bits() as f64 - self.scale as f64 * LOG2_10) / 2.0;
let res = magic_guess_scale * initial_guess.exp2();
if res.is_normal() {
BigDecimal::try_from(res).unwrap()
} else {
let scale = (self.int_val.bits() as f64 / -LOG2_10 + self.scale as f64).round() as i64;
BigDecimal::new(BigInt::from(1), scale / 2)
}
};
let max_precision = 100;
let next_iteration = move |r: BigDecimal| {
let tmp = impl_division(
self.int_val.clone(),
&r.int_val,
self.scale - r.scale,
max_precision + 1,
);
(tmp + r).half()
};
let mut running_result = next_iteration(guess);
let mut prev_result = BigDecimal::one();
let mut result = BigDecimal::zero();
while prev_result != result {
prev_result = result;
running_result = next_iteration(running_result);
result = if running_result.digits() > max_precision {
running_result.with_prec(max_precision)
} else {
running_result.clone()
};
}
return Some(result);
}
#[inline]
pub fn cbrt(&self) -> BigDecimal {
if self.is_zero() || self.is_one() {
return self.clone();
}
if self.is_negative() {
return -self.abs().cbrt();
}
let guess = {
let magic_guess_scale = 1.124960491619939_f64;
let initial_guess = (self.int_val.bits() as f64 - self.scale as f64 * LOG2_10) / 3.0;
let res = magic_guess_scale * initial_guess.exp2();
if res.is_normal() {
BigDecimal::try_from(res).unwrap()
} else {
let scale = (self.int_val.bits() as f64 / LOG2_10 - self.scale as f64).round() as i64;
BigDecimal::new(BigInt::from(1), -scale / 3)
}
};
let max_precision = 100;
let three = BigDecimal::from(3);
let next_iteration = move |r: BigDecimal| {
let sqrd = r.square();
let tmp = impl_division(
self.int_val.clone(),
&sqrd.int_val,
self.scale - sqrd.scale,
max_precision + 1,
);
let tmp = tmp + r.double();
impl_division(tmp.int_val, &three.int_val, tmp.scale - three.scale, max_precision + 1)
};
let mut running_result = next_iteration(guess);
let mut prev_result = BigDecimal::one();
let mut result = BigDecimal::zero();
while prev_result != result {
prev_result = result;
running_result = next_iteration(running_result);
result = if running_result.digits() > max_precision {
running_result.with_prec(max_precision)
} else {
running_result.clone()
};
}
return result;
}
#[inline]
pub fn inverse(&self) -> BigDecimal {
if self.is_zero() || self.is_one() {
return self.clone();
}
if self.is_negative() {
return self.abs().inverse().neg();
}
let guess = {
let bits = self.int_val.bits() as f64;
let scale = self.scale as f64;
let magic_factor = 0.721507597259061_f64;
let initial_guess = scale * LOG2_10 - bits;
let res = magic_factor * initial_guess.exp2();
if res.is_normal() {
BigDecimal::try_from(res).unwrap()
} else {
let scale = (bits / LOG2_10 + scale).round() as i64;
BigDecimal::new(BigInt::from(1), -scale)
}
};
let max_precision = 100;
let next_iteration = move |r: BigDecimal| {
let two = BigDecimal::from(2);
let tmp = two - self * &r;
r * tmp
};
let mut running_result = next_iteration(guess);
let mut prev_result = BigDecimal::one();
let mut result = BigDecimal::zero();
while prev_result != result {
prev_result = result;
running_result = next_iteration(running_result).with_prec(max_precision);
result = if running_result.digits() > max_precision {
running_result.with_prec(max_precision)
} else {
running_result.clone()
};
}
return result;
}
pub fn round(&self, round_digits: i64) -> BigDecimal {
let (bigint, decimal_part_digits) = self.as_bigint_and_exponent();
let need_to_round_digits = decimal_part_digits - round_digits;
if round_digits >= 0 && need_to_round_digits <= 0 {
return self.clone();
}
let mut number = bigint.to_i128().unwrap();
if number < 0 {
number = -number;
}
for _ in 0..(need_to_round_digits - 1) {
number /= 10;
}
let digit = number % 10;
if digit <= 4 {
self.with_scale(round_digits)
} else if bigint.is_negative() {
self.with_scale(round_digits) - BigDecimal::new(BigInt::from(1), round_digits)
} else {
self.with_scale(round_digits) + BigDecimal::new(BigInt::from(1), round_digits)
}
}
#[inline]
pub fn is_integer(&self) -> bool {
if self.scale <= 0 {
true
} else {
(self.int_val.clone() % ten_to_the(self.scale as u64)).is_zero()
}
}
#[inline]
pub fn exp(&self) -> BigDecimal {
if self.is_zero() {
return BigDecimal::one();
}
let precision = self.digits();
let mut term = self.clone();
let mut result = self.clone() + BigDecimal::one();
let mut prev_result = result.clone();
let mut factorial = BigInt::one();
for n in 2.. {
term *= self;
factorial *= n;
result += impl_division(term.int_val.clone(), &factorial, term.scale, 117 + precision);
let trimmed_result = result.with_prec(105);
if prev_result == trimmed_result {
return trimmed_result.with_prec(100);
}
prev_result = trimmed_result;
}
return result.with_prec(100);
}
#[must_use]
pub fn normalized(&self) -> BigDecimal {
if self == &BigDecimal::zero() {
return BigDecimal::zero();
}
let (sign, mut digits) = self.int_val.to_radix_be(10);
let trailing_count = digits.iter().rev().take_while(|i| **i == 0).count();
let trunc_to = digits.len() - trailing_count as usize;
digits.truncate(trunc_to);
let int_val = BigInt::from_radix_be(sign, &digits, 10).unwrap();
let scale = self.scale - trailing_count as i64;
BigDecimal::new(int_val, scale)
}
}
#[derive(Debug, PartialEq)]
pub enum ParseBigDecimalError {
ParseDecimal(ParseFloatError),
ParseInt(ParseIntError),
ParseBigInt(ParseBigIntError),
Empty,
Other(String),
}
impl fmt::Display for ParseBigDecimalError {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
use ParseBigDecimalError::*;
match *self {
ParseDecimal(ref e) => e.fmt(f),
ParseInt(ref e) => e.fmt(f),
ParseBigInt(ref e) => e.fmt(f),
Empty => "Failed to parse empty string".fmt(f),
Other(ref reason) => reason[..].fmt(f),
}
}
}
impl Error for ParseBigDecimalError {
fn description(&self) -> &str {
"failed to parse bigint/biguint"
}
}
impl From<ParseFloatError> for ParseBigDecimalError {
fn from(err: ParseFloatError) -> ParseBigDecimalError {
ParseBigDecimalError::ParseDecimal(err)
}
}
impl From<ParseIntError> for ParseBigDecimalError {
fn from(err: ParseIntError) -> ParseBigDecimalError {
ParseBigDecimalError::ParseInt(err)
}
}
impl From<ParseBigIntError> for ParseBigDecimalError {
fn from(err: ParseBigIntError) -> ParseBigDecimalError {
ParseBigDecimalError::ParseBigInt(err)
}
}
impl FromStr for BigDecimal {
type Err = ParseBigDecimalError;
#[inline]
fn from_str(s: &str) -> Result<BigDecimal, ParseBigDecimalError> {
BigDecimal::from_str_radix(s, 10)
}
}
#[allow(deprecated)] impl Hash for BigDecimal {
fn hash<H: Hasher>(&self, state: &mut H) {
let mut dec_str = self.int_val.to_str_radix(10).to_string();
let scale = self.scale;
let zero = self.int_val.is_zero();
if scale > 0 && !zero {
let mut cnt = 0;
dec_str = dec_str
.trim_right_matches(|x| {
cnt += 1;
x == '0' && cnt <= scale
})
.to_string();
} else if scale < 0 && !zero {
dec_str.push_str(&"0".repeat(self.scale.abs() as usize));
}
dec_str.hash(state);
}
}
impl PartialOrd for BigDecimal {
#[inline]
fn partial_cmp(&self, other: &BigDecimal) -> Option<Ordering> {
Some(self.cmp(other))
}
}
impl Ord for BigDecimal {
#[inline]
fn cmp(&self, other: &BigDecimal) -> Ordering {
let scmp = self.sign().cmp(&other.sign());
if scmp != Ordering::Equal {
return scmp;
}
match self.sign() {
Sign::NoSign => Ordering::Equal,
_ => {
let tmp = self - other;
match tmp.sign() {
Sign::Plus => Ordering::Greater,
Sign::Minus => Ordering::Less,
Sign::NoSign => Ordering::Equal,
}
}
}
}
}
impl PartialEq for BigDecimal {
#[inline]
fn eq(&self, rhs: &BigDecimal) -> bool {
if self.scale > rhs.scale {
let scaled_int_val = &rhs.int_val * ten_to_the((self.scale - rhs.scale) as u64);
self.int_val == scaled_int_val
} else if self.scale < rhs.scale {
let scaled_int_val = &self.int_val * ten_to_the((rhs.scale - self.scale) as u64);
scaled_int_val == rhs.int_val
} else {
self.int_val == rhs.int_val
}
}
}
impl Default for BigDecimal {
#[inline]
fn default() -> BigDecimal {
Zero::zero()
}
}
impl Zero for BigDecimal {
#[inline]
fn zero() -> BigDecimal {
BigDecimal::new(BigInt::zero(), 0)
}
#[inline]
fn is_zero(&self) -> bool {
self.int_val.is_zero()
}
}
impl One for BigDecimal {
#[inline]
fn one() -> BigDecimal {
BigDecimal::new(BigInt::one(), 0)
}
}
impl Add<BigDecimal> for BigDecimal {
type Output = BigDecimal;
#[inline]
fn add(self, rhs: BigDecimal) -> BigDecimal {
let mut lhs = self;
match lhs.scale.cmp(&rhs.scale) {
Ordering::Equal => {
lhs.int_val += rhs.int_val;
lhs
}
Ordering::Less => lhs.take_and_scale(rhs.scale) + rhs,
Ordering::Greater => rhs.take_and_scale(lhs.scale) + lhs,
}
}
}
impl<'a> Add<&'a BigDecimal> for BigDecimal {
type Output = BigDecimal;
#[inline]
fn add(self, rhs: &'a BigDecimal) -> BigDecimal {
let mut lhs = self;
match lhs.scale.cmp(&rhs.scale) {
Ordering::Equal => {
lhs.int_val += &rhs.int_val;
lhs
}
Ordering::Less => lhs.take_and_scale(rhs.scale) + rhs,
Ordering::Greater => rhs.with_scale(lhs.scale) + lhs,
}
}
}
impl<'a> Add<BigDecimal> for &'a BigDecimal {
type Output = BigDecimal;
#[inline]
fn add(self, rhs: BigDecimal) -> BigDecimal {
rhs + self
}
}
impl<'a, 'b> Add<&'b BigDecimal> for &'a BigDecimal {
type Output = BigDecimal;
#[inline]
fn add(self, rhs: &BigDecimal) -> BigDecimal {
let lhs = self;
if self.scale < rhs.scale {
lhs.with_scale(rhs.scale) + rhs
} else if self.scale > rhs.scale {
rhs.with_scale(lhs.scale) + lhs
} else {
BigDecimal::new(lhs.int_val.clone() + &rhs.int_val, lhs.scale)
}
}
}
impl Add<BigInt> for BigDecimal {
type Output = BigDecimal;
#[inline]
fn add(self, rhs: BigInt) -> BigDecimal {
let mut lhs = self;
match lhs.scale.cmp(&0) {
Ordering::Equal => {
lhs.int_val += rhs;
lhs
}
Ordering::Greater => {
lhs.int_val += rhs * ten_to_the(lhs.scale as u64);
lhs
}
Ordering::Less => lhs.take_and_scale(0) + rhs,
}
}
}
impl<'a> Add<&'a BigInt> for BigDecimal {
type Output = BigDecimal;
#[inline]
fn add(self, rhs: &BigInt) -> BigDecimal {
let mut lhs = self;
match lhs.scale.cmp(&0) {
Ordering::Equal => {
lhs.int_val += rhs;
lhs
}
Ordering::Greater => {
lhs.int_val += rhs * ten_to_the(lhs.scale as u64);
lhs
}
Ordering::Less => lhs.take_and_scale(0) + rhs,
}
}
}
impl<'a> Add<BigInt> for &'a BigDecimal {
type Output = BigDecimal;
#[inline]
fn add(self, rhs: BigInt) -> BigDecimal {
BigDecimal::new(rhs, 0) + self
}
}
impl<'a, 'b> Add<&'a BigInt> for &'b BigDecimal {
type Output = BigDecimal;
#[inline]
fn add(self, rhs: &BigInt) -> BigDecimal {
self.with_scale(0) + rhs
}
}
forward_val_assignop!(impl AddAssign for BigDecimal, add_assign);
impl<'a> AddAssign<&'a BigDecimal> for BigDecimal {
#[inline]
fn add_assign(&mut self, rhs: &BigDecimal) {
if self.scale < rhs.scale {
let scaled = self.with_scale(rhs.scale);
self.int_val = scaled.int_val + &rhs.int_val;
self.scale = rhs.scale;
} else if self.scale > rhs.scale {
let scaled = rhs.with_scale(self.scale);
self.int_val += scaled.int_val;
} else {
self.int_val += &rhs.int_val;
}
}
}
impl<'a> AddAssign<BigInt> for BigDecimal {
#[inline]
fn add_assign(&mut self, rhs: BigInt) {
*self += BigDecimal::new(rhs, 0)
}
}
impl<'a> AddAssign<&'a BigInt> for BigDecimal {
#[inline]
fn add_assign(&mut self, rhs: &BigInt) {
match self.scale.cmp(&0) {
Ordering::Equal => self.int_val += rhs,
Ordering::Greater => self.int_val += rhs * ten_to_the(self.scale as u64),
Ordering::Less => {
self.int_val *= ten_to_the((-self.scale) as u64);
self.int_val += rhs;
self.scale = 0;
}
}
}
}
impl Sub<BigDecimal> for BigDecimal {
type Output = BigDecimal;
#[inline]
fn sub(self, rhs: BigDecimal) -> BigDecimal {
let mut lhs = self;
let scale = std::cmp::max(lhs.scale, rhs.scale);
match lhs.scale.cmp(&rhs.scale) {
Ordering::Equal => {
lhs.int_val -= rhs.int_val;
lhs
}
Ordering::Less => lhs.take_and_scale(scale) - rhs,
Ordering::Greater => lhs - rhs.take_and_scale(scale),
}
}
}
impl<'a> Sub<&'a BigDecimal> for BigDecimal {
type Output = BigDecimal;
#[inline]
fn sub(self, rhs: &BigDecimal) -> BigDecimal {
let mut lhs = self;
let scale = std::cmp::max(lhs.scale, rhs.scale);
match lhs.scale.cmp(&rhs.scale) {
Ordering::Equal => {
lhs.int_val -= &rhs.int_val;
lhs
}
Ordering::Less => lhs.take_and_scale(rhs.scale) - rhs,
Ordering::Greater => lhs - rhs.with_scale(scale),
}
}
}
impl<'a> Sub<BigDecimal> for &'a BigDecimal {
type Output = BigDecimal;
#[inline]
fn sub(self, rhs: BigDecimal) -> BigDecimal {
-(rhs - self)
}
}
impl<'a, 'b> Sub<&'b BigDecimal> for &'a BigDecimal {
type Output = BigDecimal;
#[inline]
fn sub(self, rhs: &BigDecimal) -> BigDecimal {
if self.scale < rhs.scale {
self.with_scale(rhs.scale) - rhs
} else if self.scale > rhs.scale {
let rhs = rhs.with_scale(self.scale);
self - rhs
} else {
BigDecimal::new(&self.int_val - &rhs.int_val, self.scale)
}
}
}
impl Sub<BigInt> for BigDecimal {
type Output = BigDecimal;
#[inline]
fn sub(self, rhs: BigInt) -> BigDecimal {
let mut lhs = self;
match lhs.scale.cmp(&0) {
Ordering::Equal => {
lhs.int_val -= rhs;
lhs
}
Ordering::Greater => {
lhs.int_val -= rhs * ten_to_the(lhs.scale as u64);
lhs
}
Ordering::Less => lhs.take_and_scale(0) - rhs,
}
}
}
impl<'a> Sub<&'a BigInt> for BigDecimal {
type Output = BigDecimal;
#[inline]
fn sub(self, rhs: &BigInt) -> BigDecimal {
let mut lhs = self;
match lhs.scale.cmp(&0) {
Ordering::Equal => {
lhs.int_val -= rhs;
lhs
}
Ordering::Greater => {
lhs.int_val -= rhs * ten_to_the(lhs.scale as u64);
lhs
}
Ordering::Less => lhs.take_and_scale(0) - rhs,
}
}
}
impl<'a> Sub<BigInt> for &'a BigDecimal {
type Output = BigDecimal;
#[inline]
fn sub(self, rhs: BigInt) -> BigDecimal {
BigDecimal::new(rhs, 0) - self
}
}
impl<'a, 'b> Sub<&'a BigInt> for &'b BigDecimal {
type Output = BigDecimal;
#[inline]
fn sub(self, rhs: &BigInt) -> BigDecimal {
self.with_scale(0) - rhs
}
}
forward_val_assignop!(impl SubAssign for BigDecimal, sub_assign);
impl<'a> SubAssign<&'a BigDecimal> for BigDecimal {
#[inline]
fn sub_assign(&mut self, rhs: &BigDecimal) {
if self.scale < rhs.scale {
let lhs = self.with_scale(rhs.scale);
self.int_val = lhs.int_val - &rhs.int_val;
self.scale = rhs.scale;
} else if self.scale > rhs.scale {
self.int_val -= rhs.with_scale(self.scale).int_val;
} else {
self.int_val = &self.int_val - &rhs.int_val;
}
}
}
impl<'a> SubAssign<BigInt> for BigDecimal {
#[inline(always)]
fn sub_assign(&mut self, rhs: BigInt) {
*self -= BigDecimal::new(rhs, 0)
}
}
impl<'a> SubAssign<&'a BigInt> for BigDecimal {
#[inline(always)]
fn sub_assign(&mut self, rhs: &BigInt) {
match self.scale.cmp(&0) {
Ordering::Equal => SubAssign::sub_assign(&mut self.int_val, rhs),
Ordering::Greater => SubAssign::sub_assign(&mut self.int_val, rhs * ten_to_the(self.scale as u64)),
Ordering::Less => {
self.int_val *= ten_to_the((-self.scale) as u64);
SubAssign::sub_assign(&mut self.int_val, rhs);
self.scale = 0;
}
}
}
}
impl Mul<BigDecimal> for BigDecimal {
type Output = BigDecimal;
#[inline]
fn mul(mut self, rhs: BigDecimal) -> BigDecimal {
self.scale += rhs.scale;
self.int_val *= rhs.int_val;
self
}
}
impl<'a> Mul<&'a BigDecimal> for BigDecimal {
type Output = BigDecimal;
#[inline]
fn mul(mut self, rhs: &'a BigDecimal) -> BigDecimal {
self.scale += rhs.scale;
MulAssign::mul_assign(&mut self.int_val, &rhs.int_val);
self
}
}
impl<'a> Mul<BigDecimal> for &'a BigDecimal {
type Output = BigDecimal;
#[inline]
fn mul(self, rhs: BigDecimal) -> BigDecimal {
rhs * self
}
}
impl<'a, 'b> Mul<&'b BigDecimal> for &'a BigDecimal {
type Output = BigDecimal;
#[inline]
fn mul(self, rhs: &BigDecimal) -> BigDecimal {
let scale = self.scale + rhs.scale;
BigDecimal::new(&self.int_val * &rhs.int_val, scale)
}
}
impl Mul<BigInt> for BigDecimal {
type Output = BigDecimal;
#[inline]
fn mul(mut self, rhs: BigInt) -> BigDecimal {
self.int_val *= rhs;
self
}
}
impl<'a> Mul<&'a BigInt> for BigDecimal {
type Output = BigDecimal;
#[inline]
fn mul(mut self, rhs: &BigInt) -> BigDecimal {
self.int_val *= rhs;
self
}
}
impl<'a> Mul<BigInt> for &'a BigDecimal {
type Output = BigDecimal;
#[inline]
fn mul(self, mut rhs: BigInt) -> BigDecimal {
rhs *= &self.int_val;
BigDecimal::new(rhs, self.scale)
}
}
impl<'a, 'b> Mul<&'a BigInt> for &'b BigDecimal {
type Output = BigDecimal;
#[inline]
fn mul(self, rhs: &BigInt) -> BigDecimal {
let value = &self.int_val * rhs;
BigDecimal::new(value, self.scale)
}
}
forward_val_assignop!(impl MulAssign for BigDecimal, mul_assign);
impl<'a> MulAssign<&'a BigDecimal> for BigDecimal {
#[inline]
fn mul_assign(&mut self, rhs: &BigDecimal) {
self.scale += rhs.scale;
self.int_val = &self.int_val * &rhs.int_val;
}
}
impl_div_for_primitives!();
#[inline(always)]
fn impl_division(mut num: BigInt, den: &BigInt, mut scale: i64, max_precision: u64) -> BigDecimal {
if num.is_zero() {
return BigDecimal::new(num, 0);
}
match (num.is_negative(), den.is_negative()) {
(true, true) => return impl_division(num.neg(), &den.neg(), scale, max_precision),
(true, false) => return -impl_division(num.neg(), den, scale, max_precision),
(false, true) => return -impl_division(num, &den.neg(), scale, max_precision),
(false, false) => (),
}
while num < *den {
scale += 1;
num *= 10;
}
let (mut quotient, mut remainder) = num.div_rem(den);
if remainder.is_zero() {
return BigDecimal {
int_val: quotient,
scale: scale,
};
}
let mut precision = count_decimal_digits("ient);
remainder *= 10;
while !remainder.is_zero() && precision < max_precision {
let (q, r) = remainder.div_rem(den);
quotient = quotient * 10 + q;
remainder = r * 10;
precision += 1;
scale += 1;
}
if !remainder.is_zero() {
quotient += get_rounding_term(&remainder.div(den));
}
let result = BigDecimal::new(quotient, scale);
return result;
}
impl Div<BigDecimal> for BigDecimal {
type Output = BigDecimal;
#[inline]
fn div(self, other: BigDecimal) -> BigDecimal {
if other.is_zero() {
panic!("Division by zero");
}
if self.is_zero() || other.is_one() {
return self;
}
let scale = self.scale - other.scale;
if self.int_val == other.int_val {
return BigDecimal {
int_val: 1.into(),
scale: scale,
};
}
let max_precision = 100;
return impl_division(self.int_val, &other.int_val, scale, max_precision);
}
}
impl<'a> Div<&'a BigDecimal> for BigDecimal {
type Output = BigDecimal;
#[inline]
fn div(self, other: &'a BigDecimal) -> BigDecimal {
if other.is_zero() {
panic!("Division by zero");
}
if self.is_zero() || other.is_one() {
return self;
}
let scale = self.scale - other.scale;
if self.int_val == other.int_val {
return BigDecimal {
int_val: 1.into(),
scale: scale,
};
}
let max_precision = 100;
return impl_division(self.int_val, &other.int_val, scale, max_precision);
}
}
forward_ref_val_binop!(impl Div for BigDecimal, div);
impl<'a, 'b> Div<&'b BigDecimal> for &'a BigDecimal {
type Output = BigDecimal;
#[inline]
fn div(self, other: &BigDecimal) -> BigDecimal {
if other.is_zero() {
panic!("Division by zero");
}
if self.is_zero() || other.is_one() {
return self.clone();
}
let scale = self.scale - other.scale;
let num_int = &self.int_val;
let den_int = &other.int_val;
if num_int == den_int {
return BigDecimal {
int_val: 1.into(),
scale: scale,
};
}
let max_precision = 100;
return impl_division(num_int.clone(), &den_int, scale, max_precision);
}
}
impl Rem<BigDecimal> for BigDecimal {
type Output = BigDecimal;
#[inline]
fn rem(self, other: BigDecimal) -> BigDecimal {
let scale = std::cmp::max(self.scale, other.scale);
let num = self.take_and_scale(scale).int_val;
let den = other.take_and_scale(scale).int_val;
BigDecimal::new(num % den, scale)
}
}
impl<'a> Rem<&'a BigDecimal> for BigDecimal {
type Output = BigDecimal;
#[inline]
fn rem(self, other: &BigDecimal) -> BigDecimal {
let scale = std::cmp::max(self.scale, other.scale);
let num = self.take_and_scale(scale).int_val;
let den = &other.int_val;
let result = if scale == other.scale {
num % den
} else {
num % (den * ten_to_the((scale - other.scale) as u64))
};
BigDecimal::new(result, scale)
}
}
impl<'a> Rem<BigDecimal> for &'a BigDecimal {
type Output = BigDecimal;
#[inline]
fn rem(self, other: BigDecimal) -> BigDecimal {
let scale = std::cmp::max(self.scale, other.scale);
let num = &self.int_val;
let den = other.take_and_scale(scale).int_val;
let result = if scale == self.scale {
num % den
} else {
let scaled_num = num * ten_to_the((scale - self.scale) as u64);
scaled_num % den
};
BigDecimal::new(result, scale)
}
}
impl<'a, 'b> Rem<&'b BigDecimal> for &'a BigDecimal {
type Output = BigDecimal;
#[inline]
fn rem(self, other: &BigDecimal) -> BigDecimal {
let scale = std::cmp::max(self.scale, other.scale);
let num = &self.int_val;
let den = &other.int_val;
let result = match self.scale.cmp(&other.scale) {
Ordering::Equal => num % den,
Ordering::Less => {
let scaled_num = num * ten_to_the((scale - self.scale) as u64);
scaled_num % den
}
Ordering::Greater => {
let scaled_den = den * ten_to_the((scale - other.scale) as u64);
num % scaled_den
}
};
BigDecimal::new(result, scale)
}
}
impl Neg for BigDecimal {
type Output = BigDecimal;
#[inline]
fn neg(mut self) -> BigDecimal {
self.int_val = -self.int_val;
self
}
}
impl<'a> Neg for &'a BigDecimal {
type Output = BigDecimal;
#[inline]
fn neg(self) -> BigDecimal {
-self.clone()
}
}
impl Signed for BigDecimal {
#[inline]
fn abs(&self) -> BigDecimal {
match self.sign() {
Sign::Plus | Sign::NoSign => self.clone(),
Sign::Minus => -self,
}
}
#[inline]
fn abs_sub(&self, other: &BigDecimal) -> BigDecimal {
if *self <= *other {
Zero::zero()
} else {
self - other
}
}
#[inline]
fn signum(&self) -> BigDecimal {
match self.sign() {
Sign::Plus => One::one(),
Sign::NoSign => Zero::zero(),
Sign::Minus => -Self::one(),
}
}
#[inline]
fn is_positive(&self) -> bool {
self.sign() == Sign::Plus
}
#[inline]
fn is_negative(&self) -> bool {
self.sign() == Sign::Minus
}
}
impl Sum for BigDecimal {
#[inline]
fn sum<I: Iterator<Item = BigDecimal>>(iter: I) -> BigDecimal {
iter.fold(Zero::zero(), |a, b| a + b)
}
}
impl<'a> Sum<&'a BigDecimal> for BigDecimal {
#[inline]
fn sum<I: Iterator<Item = &'a BigDecimal>>(iter: I) -> BigDecimal {
iter.fold(Zero::zero(), |a, b| a + b)
}
}
impl fmt::Display for BigDecimal {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
let mut abs_int = self.int_val.abs().to_str_radix(10);
let (before, after) = if self.scale >= abs_int.len() as i64 {
let scale = self.scale as usize;
let after = "0".repeat(scale - abs_int.len()) + abs_int.as_str();
("0".to_string(), after)
} else {
let location = abs_int.len() as i64 - self.scale;
if location > abs_int.len() as i64 {
let zeros = location as usize - abs_int.len();
let abs_int = abs_int + "0".repeat(zeros as usize).as_str();
(abs_int, "".to_string())
} else {
let after = abs_int.split_off(location as usize);
(abs_int, after)
}
};
let after = if let Some(precision) = f.precision() {
let len = after.len();
if len < precision {
after + "0".repeat(precision - len).as_str()
} else {
after[0..precision].to_string()
}
} else {
after
};
let complete_without_sign = if !after.is_empty() {
before + "." + after.as_str()
} else {
before
};
let non_negative = match self.int_val.sign() {
Sign::Plus | Sign::NoSign => true,
_ => false,
};
f.pad_integral(non_negative, "", &complete_without_sign)
}
}
impl fmt::Debug for BigDecimal {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
write!(f, "BigDecimal(\"{}\")", self)
}
}
impl Num for BigDecimal {
type FromStrRadixErr = ParseBigDecimalError;
#[inline]
fn from_str_radix(s: &str, radix: u32) -> Result<BigDecimal, ParseBigDecimalError> {
if radix != 10 {
return Err(ParseBigDecimalError::Other(String::from(
"The radix for decimal MUST be 10",
)));
}
let exp_separator: &[_] = &['e', 'E'];
let (base_part, exponent_value) = match s.find(exp_separator) {
None => (s, 0),
Some(loc) => {
let (base, exp) = (&s[..loc], &s[loc + 1..]);
let exp = match exp.chars().next() {
Some('+') => &exp[1..],
_ => exp,
};
(base, i64::from_str(exp)?)
}
};
if base_part == "" {
return Err(ParseBigDecimalError::Empty);
}
let (digits, decimal_offset): (String, _) = match base_part.find('.') {
None => (base_part.to_string(), 0),
Some(loc) => {
let (lead, trail) = (&base_part[..loc], &base_part[loc + 1..]);
let mut digits = String::from(lead);
digits.push_str(trail);
(digits, trail.len() as i64)
}
};
let scale = decimal_offset - exponent_value;
let big_int = BigInt::from_str_radix(&digits, radix)?;
Ok(BigDecimal::new(big_int, scale))
}
}
impl ToPrimitive for BigDecimal {
fn to_i64(&self) -> Option<i64> {
match self.sign() {
Sign::Minus | Sign::Plus => self.with_scale(0).int_val.to_i64(),
Sign::NoSign => Some(0),
}
}
fn to_u64(&self) -> Option<u64> {
match self.sign() {
Sign::Plus => self.with_scale(0).int_val.to_u64(),
Sign::NoSign => Some(0),
Sign::Minus => None,
}
}
fn to_f64(&self) -> Option<f64> {
self.int_val.to_f64().map(|x| x * 10f64.powi(-self.scale as i32))
}
}
impl From<i64> for BigDecimal {
#[inline]
fn from(n: i64) -> Self {
BigDecimal {
int_val: BigInt::from(n),
scale: 0,
}
}
}
impl From<u64> for BigDecimal {
#[inline]
fn from(n: u64) -> Self {
BigDecimal {
int_val: BigInt::from(n),
scale: 0,
}
}
}
impl From<(BigInt, i64)> for BigDecimal {
#[inline]
fn from((int_val, scale): (BigInt, i64)) -> Self {
BigDecimal {
int_val: int_val,
scale: scale,
}
}
}
impl From<BigInt> for BigDecimal {
#[inline]
fn from(int_val: BigInt) -> Self {
BigDecimal {
int_val: int_val,
scale: 0,
}
}
}
macro_rules! impl_from_type {
($FromType:ty, $AsType:ty) => {
impl From<$FromType> for BigDecimal {
#[inline]
#[allow(clippy::cast_lossless)]
fn from(n: $FromType) -> Self {
BigDecimal::from(n as $AsType)
}
}
};
}
impl_from_type!(u8, u64);
impl_from_type!(u16, u64);
impl_from_type!(u32, u64);
impl_from_type!(i8, i64);
impl_from_type!(i16, i64);
impl_from_type!(i32, i64);
impl TryFrom<f32> for BigDecimal {
type Error = ParseBigDecimalError;
#[inline]
fn try_from(n: f32) -> Result<Self, Self::Error> {
BigDecimal::from_str(&format!("{:.PRECISION$e}", n, PRECISION = ::std::f32::DIGITS as usize))
}
}
impl TryFrom<f64> for BigDecimal {
type Error = ParseBigDecimalError;
#[inline]
fn try_from(n: f64) -> Result<Self, Self::Error> {
BigDecimal::from_str(&format!("{:.PRECISION$e}", n, PRECISION = ::std::f64::DIGITS as usize))
}
}
impl FromPrimitive for BigDecimal {
#[inline]
fn from_i64(n: i64) -> Option<Self> {
Some(BigDecimal::from(n))
}
#[inline]
fn from_u64(n: u64) -> Option<Self> {
Some(BigDecimal::from(n))
}
#[inline]
fn from_f32(n: f32) -> Option<Self> {
BigDecimal::try_from(n).ok()
}
#[inline]
fn from_f64(n: f64) -> Option<Self> {
BigDecimal::try_from(n).ok()
}
}
impl ToBigInt for BigDecimal {
fn to_bigint(&self) -> Option<BigInt> {
Some(self.with_scale(0).int_val)
}
}
#[cfg(feature = "serde")]
mod bigdecimal_serde {
use super::BigDecimal;
use serde::{de, ser};
use std::convert::TryFrom;
use std::fmt;
use std::str::FromStr;
#[allow(unused_imports)]
use num_traits::FromPrimitive;
impl ser::Serialize for BigDecimal {
fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
where
S: ser::Serializer,
{
serializer.collect_str(&self)
}
}
struct BigDecimalVisitor;
impl<'de> de::Visitor<'de> for BigDecimalVisitor {
type Value = BigDecimal;
fn expecting(&self, formatter: &mut fmt::Formatter) -> fmt::Result {
write!(formatter, "a number or formatted decimal string")
}
fn visit_str<E>(self, value: &str) -> Result<BigDecimal, E>
where
E: de::Error,
{
BigDecimal::from_str(value).map_err(|err| E::custom(format!("{}", err)))
}
fn visit_u64<E>(self, value: u64) -> Result<BigDecimal, E>
where
E: de::Error,
{
Ok(BigDecimal::from(value))
}
fn visit_i64<E>(self, value: i64) -> Result<BigDecimal, E>
where
E: de::Error,
{
Ok(BigDecimal::from(value))
}
fn visit_f64<E>(self, value: f64) -> Result<BigDecimal, E>
where
E: de::Error,
{
BigDecimal::try_from(value).map_err(|err| E::custom(format!("{}", err)))
}
}
#[cfg(not(feature = "string-only"))]
impl<'de> de::Deserialize<'de> for BigDecimal {
fn deserialize<D>(d: D) -> Result<Self, D::Error>
where
D: de::Deserializer<'de>,
{
d.deserialize_any(BigDecimalVisitor)
}
}
#[cfg(feature = "string-only")]
impl<'de> de::Deserialize<'de> for BigDecimal {
fn deserialize<D>(d: D) -> Result<Self, D::Error>
where
D: de::Deserializer<'de>,
{
d.deserialize_str(BigDecimalVisitor)
}
}
#[cfg(test)]
extern crate serde_json;
#[test]
fn test_serde_serialize() {
use std::str::FromStr;
let vals = vec![
("1.0", "1.0"),
("0.5", "0.5"),
("50", "50"),
("50000", "50000"),
("1e-3", "0.001"),
("1e12", "1000000000000"),
("0.25", "0.25"),
("12.34", "12.34"),
("0.15625", "0.15625"),
("0.3333333333333333", "0.3333333333333333"),
("3.141592653589793", "3.141592653589793"),
("94247.77960769380", "94247.77960769380"),
("10.99", "10.99"),
("12.0010", "12.0010"),
];
for (s, v) in vals {
let expected = format!("\"{}\"", v);
let value = serde_json::to_string(&BigDecimal::from_str(s).unwrap()).unwrap();
assert_eq!(expected, value);
}
}
#[test]
fn test_serde_deserialize_str() {
use std::str::FromStr;
let vals = vec![
("1.0", "1.0"),
("0.5", "0.5"),
("50", "50"),
("50000", "50000"),
("1e-3", "0.001"),
("1e12", "1000000000000"),
("0.25", "0.25"),
("12.34", "12.34"),
("0.15625", "0.15625"),
("0.3333333333333333", "0.3333333333333333"),
("3.141592653589793", "3.141592653589793"),
("94247.77960769380", "94247.77960769380"),
("10.99", "10.99"),
("12.0010", "12.0010"),
];
for (s, v) in vals {
let expected = BigDecimal::from_str(v).unwrap();
let value: BigDecimal = serde_json::from_str(&format!("\"{}\"", s)).unwrap();
assert_eq!(expected, value);
}
}
#[test]
#[cfg(not(feature = "string-only"))]
fn test_serde_deserialize_int() {
let vals = vec![0, 1, 81516161, -370, -8, -99999999999];
for n in vals {
let expected = BigDecimal::from_i64(n).unwrap();
let value: BigDecimal = serde_json::from_str(&serde_json::to_string(&n).unwrap()).unwrap();
assert_eq!(expected, value);
}
}
#[test]
#[cfg(not(feature = "string-only"))]
fn test_serde_deserialize_f64() {
let vals = vec![
1.0,
0.5,
0.25,
50.0,
50000.,
0.001,
12.34,
5.0 * 0.03125,
::std::f64::consts::PI,
::std::f64::consts::PI * 10000.0,
::std::f64::consts::PI * 30000.0,
];
for n in vals {
let expected = BigDecimal::from_f64(n).unwrap();
let value: BigDecimal = serde_json::from_str(&serde_json::to_string(&n).unwrap()).unwrap();
assert_eq!(expected, value);
}
}
}
#[rustfmt::skip]
#[cfg(test)]
mod bigdecimal_tests {
use BigDecimal;
use num_traits::{ToPrimitive, FromPrimitive, Signed, Zero, One};
use std::convert::TryFrom;
use std::str::FromStr;
use num_bigint;
#[test]
fn test_sum() {
let vals = vec![
BigDecimal::from_f32(2.5).unwrap(),
BigDecimal::from_f32(0.3).unwrap(),
BigDecimal::from_f32(0.001).unwrap(),
];
let expected_sum = BigDecimal::from_f32(2.801).unwrap();
let sum = vals.iter().sum::<BigDecimal>();
assert_eq!(expected_sum, sum);
}
#[test]
fn test_sum1() {
let vals = vec![
BigDecimal::from_f32(0.1).unwrap(),
BigDecimal::from_f32(0.2).unwrap(),
];
let expected_sum = BigDecimal::from_f32(0.3).unwrap();
let sum = vals.iter().sum::<BigDecimal>();
assert_eq!(expected_sum, sum);
}
#[test]
fn test_to_i64() {
let vals = vec![
("12.34", 12),
("3.14", 3),
("50", 50),
("50000", 50000),
("0.001", 0),
];
for (s, ans) in vals {
let calculated = BigDecimal::from_str(s).unwrap().to_i64().unwrap();
assert_eq!(ans, calculated);
}
}
#[test]
fn test_to_f64() {
let vals = vec![
("12.34", 12.34),
("3.14", 3.14),
("50", 50.),
("50000", 50000.),
("0.001", 0.001),
];
for (s, ans) in vals {
let diff = BigDecimal::from_str(s).unwrap().to_f64().unwrap() - ans;
let diff = diff.abs();
assert!(diff < 1e-10);
}
}
#[test]
fn test_from_i8() {
let vals = vec![
("0", 0),
("1", 1),
("12", 12),
("-13", -13),
("111", 111),
("-128", ::std::i8::MIN),
("127", ::std::i8::MAX),
];
for (s, n) in vals {
let expected = BigDecimal::from_str(s).unwrap();
let value = BigDecimal::from_i8(n).unwrap();
assert_eq!(expected, value);
}
}
#[test]
fn test_from_f32() {
let vals = vec![
("1.0", 1.0),
("0.5", 0.5),
("0.25", 0.25),
("50.", 50.0),
("50000", 50000.),
("0.001", 0.001),
("12.34", 12.34),
("0.15625", 5.0 * 0.03125),
("3.141593", ::std::f32::consts::PI),
("31415.93", ::std::f32::consts::PI * 10000.0),
("94247.78", ::std::f32::consts::PI * 30000.0),
];
for (s, n) in vals {
let expected = BigDecimal::from_str(s).unwrap();
let value = BigDecimal::from_f32(n).unwrap();
assert_eq!(expected, value);
}
}
#[test]
fn test_from_f64() {
let vals = vec![
("1.0", 1.0f64),
("0.5", 0.5),
("50", 50.),
("50000", 50000.),
("1e-3", 0.001),
("0.25", 0.25),
("12.34", 12.34),
("0.15625", 5.0 * 0.03125),
("0.3333333333333333", 1.0 / 3.0),
("3.141592653589793", ::std::f64::consts::PI),
("31415.92653589793", ::std::f64::consts::PI * 10000.0f64),
("94247.77960769380", ::std::f64::consts::PI * 30000.0f64),
];
for (s, n) in vals {
let expected = BigDecimal::from_str(s).unwrap();
let value = BigDecimal::from_f64(n).unwrap();
assert_eq!(expected, value);
}
}
#[test]
fn test_nan_float() {
assert!(BigDecimal::try_from(std::f32::NAN).is_err());
assert!(BigDecimal::try_from(std::f64::NAN).is_err());
}
#[test]
fn test_add() {
let vals = vec![
("12.34", "1.234", "13.574"),
("12.34", "-1.234", "11.106"),
("1234e6", "1234e-6", "1234000000.001234"),
("1234e-6", "1234e6", "1234000000.001234"),
("18446744073709551616.0", "1", "18446744073709551617"),
("184467440737e3380", "0", "184467440737e3380"),
];
for &(x, y, z) in vals.iter() {
let mut a = BigDecimal::from_str(x).unwrap();
let b = BigDecimal::from_str(y).unwrap();
let c = BigDecimal::from_str(z).unwrap();
assert_eq!(a.clone() + b.clone(), c);
assert_eq!(a.clone() + &b, c);
assert_eq!(&a + b.clone(), c);
assert_eq!(&a + &b, c);
a += b;
assert_eq!(a, c);
}
}
#[test]
fn test_sub() {
let vals = vec![
("12.34", "1.234", "11.106"),
("12.34", "-1.234", "13.574"),
("1234e6", "1234e-6", "1233999999.998766"),
];
for &(x, y, z) in vals.iter() {
let mut a = BigDecimal::from_str(x).unwrap();
let b = BigDecimal::from_str(y).unwrap();
let c = BigDecimal::from_str(z).unwrap();
assert_eq!(a.clone() - b.clone(), c);
assert_eq!(a.clone() - &b, c);
assert_eq!(&a - b.clone(), c);
assert_eq!(&a - &b, c);
a -= b;
assert_eq!(a, c);
}
}
#[test]
fn test_mul() {
let vals = vec![
("2", "1", "2"),
("12.34", "1.234", "15.22756"),
("2e1", "1", "20"),
("3", ".333333", "0.999999"),
("2389472934723", "209481029831", "500549251119075878721813"),
("1e-450", "1e500", ".1e51"),
];
for &(x, y, z) in vals.iter() {
let mut a = BigDecimal::from_str(x).unwrap();
let b = BigDecimal::from_str(y).unwrap();
let c = BigDecimal::from_str(z).unwrap();
assert_eq!(a.clone() * b.clone(), c);
assert_eq!(a.clone() * &b, c);
assert_eq!(&a * b.clone(), c);
assert_eq!(&a * &b, c);
a *= b;
assert_eq!(a, c);
}
}
#[test]
fn test_div() {
let vals = vec![
("0", "1", "0"),
("0", "10", "0"),
("2", "1", "2"),
("2e1", "1", "2e1"),
("10", "10", "1"),
("100", "10.0", "1e1"),
("20.0", "200", ".1"),
("4", "2", "2.0"),
("15", "3", "5.0"),
("1", "2", "0.5"),
("1", "2e-2", "5e1"),
("1", "0.2", "5"),
("1.0", "0.02", "50"),
("1", "0.020", "5e1"),
("5.0", "4.00", "1.25"),
("5.0", "4.000", "1.25"),
("5", "4.000", "1.25"),
("5", "4", "125e-2"),
("100", "5", "20"),
("-50", "5", "-10"),
("200", "-5", "-40."),
("1", "3", ".3333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333"),
("-2", "-3", ".6666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666667"),
("-12.34", "1.233", "-10.00811030008110300081103000811030008110300081103000811030008110300081103000811030008110300081103001"),
("125348", "352.2283", "355.8714617763535752237966114591019517738921035021887792661748076460636467881768727839301952739175132"),
];
for &(x, y, z) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap();
let b = BigDecimal::from_str(y).unwrap();
let c = BigDecimal::from_str(z).unwrap();
assert_eq!(a.clone() / b.clone(), c);
assert_eq!(a.clone() / &b, c);
assert_eq!(&a / b.clone(), c);
assert_eq!(&a / &b, c);
}
}
#[test]
#[should_panic(expected = "Division by zero")]
fn test_division_by_zero_panics() {
let x = BigDecimal::from_str("3.14").unwrap();
let _r = x / 0;
}
#[test]
#[should_panic(expected = "Division by zero")]
fn test_division_by_zero_panics_v2() {
let x = BigDecimal::from_str("3.14").unwrap();
let _r = x / BigDecimal::zero();
}
#[test]
fn test_rem() {
let vals = vec![
("100", "5", "0"),
("2e1", "1", "0"),
("2", "1", "0"),
("1", "3", "1"),
("1", "0.5", "0"),
("1.5", "1", "0.5"),
("1", "3e-2", "1e-2"),
("10", "0.003", "0.001"),
("3", "2", "1"),
("-3", "2", "-1"),
("3", "-2", "1"),
("-3", "-2", "-1"),
("12.34", "1.233", "0.01"),
];
for &(x, y, z) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap();
let b = BigDecimal::from_str(y).unwrap();
let c = BigDecimal::from_str(z).unwrap();
let rem = &a % &b;
assert_eq!(rem, c, "{} [&{} % &{}] == {}", rem, a, b, c);
let rem = a.clone() % &b;
assert_eq!(rem, c, "{} [{} % &{}] == {}", rem, a, b, c);
let rem = &a % b.clone();
assert_eq!(rem, c, "{} [&{} % {}] == {}", rem, a, b, c);
let rem = a.clone() % b.clone();
assert_eq!(rem, c, "{} [{} % {}] == {}", rem, a, b, c);
}
let vals = vec![
(("100", -2), ("50", -1), "0"),
(("100", 0), ("50", -1), "0"),
(("100", -2), ("30", 0), "10"),
(("100", 0), ("30", -1), "10"),
];
for &((x, xs), (y, ys), z) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap().with_scale(xs);
let b = BigDecimal::from_str(y).unwrap().with_scale(ys);
let c = BigDecimal::from_str(z).unwrap();
let rem = &a % &b;
assert_eq!(rem, c, "{} [{} % {}] == {}", rem, a, b, c);
}
}
#[test]
fn test_equal() {
let vals = vec![
("2", ".2e1"),
("0e1", "0.0"),
("0e0", "0.0"),
("0e-0", "0.0"),
("-0901300e-3", "-901.3"),
("-0.901300e+3", "-901.3"),
("-0e-1", "-0.0"),
("2123121e1231", "212.3121e1235"),
];
for &(x, y) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap();
let b = BigDecimal::from_str(y).unwrap();
assert_eq!(a, b);
}
}
#[test]
fn test_not_equal() {
let vals = vec![
("2", ".2e2"),
("1e45", "1e-900"),
("1e+900", "1e-900"),
];
for &(x, y) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap();
let b = BigDecimal::from_str(y).unwrap();
assert!(a != b, "{} == {}", a, b);
}
}
#[test]
fn test_hash_equal() {
use std::hash::{Hash, Hasher};
use std::collections::hash_map::DefaultHasher;
fn hash<T>(obj: &T) -> u64
where T: Hash
{
let mut hasher = DefaultHasher::new();
obj.hash(&mut hasher);
hasher.finish()
}
let vals = vec![
("1.1234", "1.1234000"),
("1.12340000", "1.1234"),
("001.1234", "1.1234000"),
("001.1234", "0001.1234"),
("1.1234000000", "1.1234000"),
("1.12340", "1.1234000000"),
("-0901300e-3", "-901.3"),
("-0.901300e+3", "-901.3"),
("100", "100.00"),
("100.00", "100"),
("0.00", "0"),
("0.00", "0.000"),
("-0.00", "0.000"),
("0.00", "-0.000"),
];
for &(x,y) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap();
let b = BigDecimal::from_str(y).unwrap();
assert_eq!(a, b);
assert_eq!(hash(&a), hash(&b), "hash({}) != hash({})", a, b);
}
}
#[test]
fn test_hash_not_equal() {
use std::hash::{Hash, Hasher};
use std::collections::hash_map::DefaultHasher;
fn hash<T>(obj: &T) -> u64
where T: Hash
{
let mut hasher = DefaultHasher::new();
obj.hash(&mut hasher);
hasher.finish()
}
let vals = vec![
("1.1234", "1.1234001"),
("10000", "10"),
("10", "10000"),
("10.0", "100"),
];
for &(x,y) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap();
let b = BigDecimal::from_str(y).unwrap();
assert!(a != b, "{} == {}", a, b);
assert!(hash(&a) != hash(&b), "hash({}) == hash({})", a, b);
}
}
#[test]
fn test_hash_equal_scale() {
use std::hash::{Hash, Hasher};
use std::collections::hash_map::DefaultHasher;
fn hash<T>(obj: &T) -> u64
where T: Hash
{
let mut hasher = DefaultHasher::new();
obj.hash(&mut hasher);
hasher.finish()
}
let vals = vec![
("1234.5678", -2, "1200", 0),
("1234.5678", -2, "1200", -2),
("1234.5678", 0, "1234.1234", 0),
("1234.5678", -3, "1200", -3),
("-1234", -2, "-1200", 0),
];
for &(x,xs,y,ys) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap().with_scale(xs);
let b = BigDecimal::from_str(y).unwrap().with_scale(ys);
assert_eq!(a, b);
assert_eq!(hash(&a), hash(&b), "hash({}) != hash({})", a, b);
}
}
#[test]
fn test_with_prec() {
let vals = vec![
("7", 1, "7"),
("7", 2, "7.0"),
("895", 2, "900"),
("8934", 2, "8900"),
("8934", 1, "9000"),
("1.0001", 5, "1.0001"),
("1.0001", 4, "1"),
("1.00009", 6, "1.00009"),
("1.00009", 5, "1.0001"),
("1.00009", 4, "1.000"),
];
for &(x, p, y) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap().with_prec(p);
assert_eq!(a, BigDecimal::from_str(y).unwrap());
}
}
#[test]
fn test_digits() {
let vals = vec![
("0", 1),
("7", 1),
("10", 2),
("8934", 4),
];
for &(x, y) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap();
assert_eq!(a.digits(), y);
}
}
#[test]
fn test_get_rounding_term() {
use num_bigint::BigInt;
use super::get_rounding_term;
let vals = vec![
("0", 0),
("4", 0),
("5", 1),
("10", 0),
("15", 0),
("49", 0),
("50", 1),
("51", 1),
("8934", 1),
("9999", 1),
("10000", 0),
("50000", 1),
("99999", 1),
("100000", 0),
("100001", 0),
("10000000000", 0),
("9999999999999999999999999999999999999999", 1),
("10000000000000000000000000000000000000000", 0),
];
for &(x, y) in vals.iter() {
let a = BigInt::from_str(x).unwrap();
assert_eq!(get_rounding_term(&a), y, "{}", x);
}
}
#[test]
fn test_abs() {
let vals = vec![
("10", "10"),
("-10", "10"),
];
for &(x, y) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap().abs();
let b = BigDecimal::from_str(y).unwrap();
assert!(a == b, "{} == {}", a, b);
}
}
#[test]
fn test_count_decimal_digits() {
use num_bigint::BigInt;
use super::count_decimal_digits;
let vals = vec![
("10", 2),
("1", 1),
("9", 1),
("999", 3),
("1000", 4),
("9900", 4),
("9999", 4),
("10000", 5),
("99999", 5),
("100000", 6),
("999999", 6),
("1000000", 7),
("9999999", 7),
("999999999999", 12),
("999999999999999999999999", 24),
("999999999999999999999999999999999999999999999999", 48),
("999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999", 96),
("199999911199999999999999999999999999999999999999999999999999999999999999999999999999999999999000", 96),
("999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999991", 192),
("199999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999", 192),
];
for &(x, y) in vals.iter() {
let a = BigInt::from_str(x).unwrap();
let b = count_decimal_digits(&a);
assert_eq!(b, y);
}
}
#[test]
fn test_half() {
let vals = vec![
("100", "50."),
("2", "1"),
(".2", ".1"),
("42", "21"),
("3", "1.5"),
("99", "49.5"),
("3.141592653", "1.5707963265"),
("3.1415926536", "1.5707963268"),
];
for &(x, y) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap().half();
let b = BigDecimal::from_str(y).unwrap();
assert_eq!(a, b);
assert_eq!(a.scale, b.scale);
}
}
#[test]
fn test_round() {
let test_cases = vec![
("1.45", 1, "1.5"),
("1.444445", 1, "1.4"),
("1.44", 1, "1.4"),
("0.444", 2, "0.44"),
("0.0045", 2, "0.00"),
("-1.555", 2, "-1.56"),
("-1.555", 99, "-1.555"),
("5.5", 0, "6"),
("-1", -1, "0"),
("5", -1, "10"),
("44", -1, "40"),
("44", -99, "0"),
("1.4499999999", 1, "1.4"),
("-1.4499999999", 1, "-1.4"),
("1.449999999", 1, "1.4"),
("-9999.444455556666", 10, "-9999.4444555567"),
("-12345678987654321.123456789", 8, "-12345678987654321.12345679"),
];
for &(x, digits, y) in test_cases.iter() {
let a = BigDecimal::from_str(x).unwrap();
let b = BigDecimal::from_str(y).unwrap();
assert_eq!(a.round(digits), b);
}
}
#[test]
fn test_is_integer() {
let true_vals = vec![
"100",
"100.00",
"1724e4",
"31.47e8",
"-31.47e8",
"-0.0",
];
let false_vals = vec![
"100.1",
"0.001",
"3147e-3",
"3147e-8",
"-0.01",
"-1e-3",
];
for s in true_vals {
let d = BigDecimal::from_str(&s).unwrap();
assert!(d.is_integer());
}
for s in false_vals {
let d = BigDecimal::from_str(&s).unwrap();
assert!(!d.is_integer());
}
}
#[test]
fn test_inverse() {
let vals = vec![
("100", "0.01"),
("2", "0.5"),
(".2", "5"),
("3.141592653", "0.3183098862435492205742690218851870990799646487459493049686604293188738877535183744268834079171116523"),
];
for &(x, y) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap();
let i = a.inverse();
let b = BigDecimal::from_str(y).unwrap();
assert_eq!(i, b);
assert_eq!(BigDecimal::from(1)/&a, b);
assert_eq!(i.inverse(), a);
}
}
#[test]
fn test_sqrt() {
let vals = vec![
("1e-232", "1e-116"),
("1.00", "1"),
("1.001", "1.000499875062460964823258287700109753027590031219780479551442971840836093890879944856933288426795152"),
("100", "10"),
("49", "7"),
(".25", ".5"),
("0.0152399025", ".12345"),
("152399025", "12345"),
(".00400", "0.06324555320336758663997787088865437067439110278650433653715009705585188877278476442688496216758600590"),
(".1", "0.3162277660168379331998893544432718533719555139325216826857504852792594438639238221344248108379300295"),
("2", "1.414213562373095048801688724209698078569671875376948073176679737990732478462107038850387534327641573"),
("125348", "354.0451948551201563108487193176101314241016013304294520812832530590100407318465590778759640828114535"),
("18446744073709551616.1099511", "4294967296.000000000012799992691725492477397918722952224079252026972356303360555051219312462698703293"),
("3.141592653589793115997963468544185161590576171875", "1.772453850905515992751519103139248439290428205003682302442979619028063165921408635567477284443197875"),
(".000000000089793115997963468544185161590576171875", "0.000009475922962855041517561783740144225422359796851494316346796373337470068631250135521161989831460407155"),
("0.7177700109762963922745342343167413624881759290454997218753321040760896053150388903350654937434826216803814031987652326749140535150336357405672040727695124057298138872112244784753994931999476811850580200000000000000000000000000000000", "0.8472130847527653667042980517799020703921106560594525833177762276594388966885185567535692987624493813"),
("0.01234567901234567901234567901234567901234567901234567901234567901234567901234567901234567901234567901", "0.1111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111"),
("0.1108890000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000444", "0.3330000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000667"),
];
for &(x, y) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap().sqrt().unwrap();
let b = BigDecimal::from_str(y).unwrap();
assert_eq!(a, b);
}
}
#[test]
fn test_big_sqrt() {
use num_bigint::BigInt;
let vals = vec![
(("2", -70), "141421356237309504880168872420969807.8569671875376948073176679737990732478462107038850387534327641573"),
(("3", -170), "17320508075688772935274463415058723669428052538103806280558069794519330169088000370811.46186757248576"),
(("9", -199), "9486832980505137995996680633298155601158665417975650480572514558377783315917714664032744325137900886"),
(("7", -200), "26457513110645905905016157536392604257102591830824501803683344592010688232302836277603928864745436110"),
(("777", -204), "2.787471972953270789531596912111625325974789615194854615319795902911796043681078997362635440358922503E+103"),
(("7", -600), "2.645751311064590590501615753639260425710259183082450180368334459201068823230283627760392886474543611E+300"),
(("2", -900), "1.414213562373095048801688724209698078569671875376948073176679737990732478462107038850387534327641573E+450"),
(("7", -999), "8.366600265340755479781720257851874893928153692986721998111915430804187725943170098308147119649515362E+499"),
(("74908163946345982392040522594123773796", -999), "2.736935584670307552030924971360722787091742391079630976117950955395149091570790266754718322365663909E+518"),
(("20", -1024), "4.472135954999579392818347337462552470881236719223051448541794490821041851275609798828828816757564550E512"),
(("3", 1025), "5.477225575051661134569697828008021339527446949979832542268944497324932771227227338008584361638706258E-513"),
];
for &((s, scale), e) in vals.iter() {
let expected = BigDecimal::from_str(e).unwrap();
let sqrt = BigDecimal::new(BigInt::from_str(s).unwrap(), scale).sqrt().unwrap();
assert_eq!(sqrt, expected);
}
}
#[test]
fn test_cbrt() {
let vals = vec![
("0.00", "0"),
("1.00", "1"),
("1.001", "1.000333222283909495175449559955220102010284758197360454054345461242739715702641939155238095670636841"),
("10", "2.154434690031883721759293566519350495259344942192108582489235506346411106648340800185441503543243276"),
("-59283293e25", "-84006090355.84281237113712383191213626687332139035750444925827809487776780721673264524620270275301685"),
("94213372931e-127", "2.112049945275324414051072540210070583697242797173805198575907094646677475250362108901530353886613160E-39"),
];
for &(x, y) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap().cbrt();
let b = BigDecimal::from_str(y).unwrap();
assert_eq!(a, b);
}
}
#[test]
fn test_double() {
let vals = vec![
("1", "2"),
("1.00", "2.00"),
("1.50", "3.00"),
("5", "10"),
("5.0", "10.0"),
("5.5", "11.0"),
("5.05", "10.10"),
];
for &(x, y) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap().double();
let b = BigDecimal::from_str(y).unwrap();
assert_eq!(a, b);
assert_eq!(a.scale, b.scale);
}
}
#[test]
fn test_square() {
let vals = vec![
("1.00", "1.00"),
("1.5", "2.25"),
("1.50", "2.2500"),
("5", "25"),
("5.0", "25.00"),
("-5.0", "25.00"),
("5.5", "30.25"),
("0.80", "0.6400"),
("0.01234", "0.0001522756"),
("3.1415926", "9.86960406437476"),
];
for &(x, y) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap().square();
let b = BigDecimal::from_str(y).unwrap();
assert_eq!(a, b);
assert_eq!(a.scale, b.scale);
}
}
#[test]
fn test_cube() {
let vals = vec![
("1.00", "1.00"),
("1.50", "3.375000"),
("5", "125"),
("5.0", "125.000"),
("5.00", "125.000000"),
("-5", "-125"),
("-5.0", "-125.000"),
("2.01", "8.120601"),
("5.5", "166.375"),
("0.01234", "0.000001879080904"),
("3.1415926", "31.006275093569669642776"),
];
for &(x, y) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap().cube();
let b = BigDecimal::from_str(y).unwrap();
assert_eq!(a, b);
assert_eq!(a.scale, b.scale);
}
}
#[test]
fn test_exp() {
let vals = vec![
("0", "1"),
("1", "2.718281828459045235360287471352662497757247093699959574966967627724076630353547594571382178525166427"),
("1.01", "2.745601015016916493989776316660387624073750819595962291667398087987297168243899027802501018008905180"),
("0.5", "1.648721270700128146848650787814163571653776100710148011575079311640661021194215608632776520056366643"),
("-1", "0.3678794411714423215955237701614608674458111310317678345078368016974614957448998033571472743459196437"),
("-0.01", "0.9900498337491680535739059771800365577720790812538374668838787452931477271687452950182155307793838110"),
("-10.04", "0.00004361977305405268676261569570537884674661515701779752139657120453194647205771372804663141467275928595"),
("-20.07", "1.921806899438469499721914055500607234723811054459447828795824348465763824284589956630853464778332349E-9"),
("10", "22026.46579480671651695790064528424436635351261855678107423542635522520281857079257519912096816452590"),
("20", "485165195.4097902779691068305415405586846389889448472543536108003159779961427097401659798506527473494"),
];
for &(x, y) in vals.iter() {
let a = BigDecimal::from_str(x).unwrap().exp();
let b = BigDecimal::from_str(y).unwrap();
assert_eq!(a, b);
}
}
#[test]
fn test_from_str() {
let vals = vec![
("1331.107", 1331107, 3),
("1.0", 10, 1),
("2e1", 2, -1),
("0.00123", 123, 5),
("-123", -123, 0),
("-1230", -1230, 0),
("12.3", 123, 1),
("123e-1", 123, 1),
("1.23e+1", 123, 1),
("1.23E+3", 123, -1),
("1.23E-8", 123, 10),
("-1.23E-10", -123, 12),
];
for &(source, val, scale) in vals.iter() {
let x = BigDecimal::from_str(source).unwrap();
assert_eq!(x.int_val.to_i32().unwrap(), val);
assert_eq!(x.scale, scale);
}
}
#[test]
fn test_fmt() {
let vals = vec![
(1, 0, ( "1", "1.0", "1.0000", " 1.0", "+01.0", "1.0 " )),
(1, 1, ( "0.1", "0.1", "0.1000", " 0.1", "+00.1", "0.1 " )),
(1, 2, ( "0.01", "0.0", "0.0100", " 0.0", "+00.0", "0.0 " )),
(1, -2, ("100", "100.0", "100.0000", "100.0", "+100.0", "100.0" )),
(-1, 0, ( "-1", "-1.0", "-1.0000", "-1.0", "-01.0", "-1.0" )),
(-1, 1, ( "-0.1", "-0.1", "-0.1000", "-0.1", "-00.1", "-0.1" )),
(-1, 2, ( "-0.01", "-0.0", "-0.0100", "-0.0", "-00.0", "-0.0" )),
];
for (i, scale, results) in vals {
let x = BigDecimal::new(num_bigint::BigInt::from(i), scale);
assert_eq!(format!("{}", x), results.0);
assert_eq!(format!("{:.1}", x), results.1);
assert_eq!(format!("{:.4}", x), results.2);
assert_eq!(format!("{:4.1}", x), results.3);
assert_eq!(format!("{:+05.1}", x), results.4);
assert_eq!(format!("{:<4.1}", x), results.5);
}
}
#[test]
fn test_debug() {
let vals = vec![
("BigDecimal(\"123.456\")", "123.456"),
("BigDecimal(\"123.400\")", "123.400"),
("BigDecimal(\"1.20\")", "01.20"),
("BigDecimal(\"1200\")", "01.2E3"),
];
for (expected, source) in vals {
let var = BigDecimal::from_str(source).unwrap();
assert_eq!(format!("{:?}", var), expected);
}
}
#[test]
fn test_signed() {
assert!(!BigDecimal::zero().is_positive());
assert!(!BigDecimal::one().is_negative());
assert!(BigDecimal::one().is_positive());
assert!((-BigDecimal::one()).is_negative());
assert!((-BigDecimal::one()).abs().is_positive());
}
#[test]
fn test_normalize() {
use num_bigint::BigInt;
let vals = vec![
(BigDecimal::new(BigInt::from(10), 2),
BigDecimal::new(BigInt::from(1), 1),
"0.1"),
(BigDecimal::new(BigInt::from(132400), -4),
BigDecimal::new(BigInt::from(1324), -6),
"1324000000"),
(BigDecimal::new(BigInt::from(1_900_000), 3),
BigDecimal::new(BigInt::from(19), -2),
"1900"),
(BigDecimal::new(BigInt::from(0), -3),
BigDecimal::zero(),
"0"),
(BigDecimal::new(BigInt::from(0), 5),
BigDecimal::zero(),
"0"),
];
for (not_normalized, normalized, string) in vals {
assert_eq!(not_normalized.normalized(), normalized);
assert_eq!(not_normalized.normalized().to_string(), string);
assert_eq!(normalized.to_string(), string);
}
}
#[test]
#[should_panic(expected = "InvalidDigit")]
fn test_bad_string_nan() {
BigDecimal::from_str("hello").unwrap();
}
#[test]
#[should_panic(expected = "Empty")]
fn test_bad_string_empty() {
BigDecimal::from_str("").unwrap();
}
#[test]
#[should_panic(expected = "InvalidDigit")]
fn test_bad_string_invalid_char() {
BigDecimal::from_str("12z3.12").unwrap();
}
#[test]
#[should_panic(expected = "InvalidDigit")]
fn test_bad_string_nan_exponent() {
BigDecimal::from_str("123.123eg").unwrap();
}
#[test]
#[should_panic(expected = "Empty")]
fn test_bad_string_empty_exponent() {
BigDecimal::from_str("123.123E").unwrap();
}
#[test]
#[should_panic(expected = "InvalidDigit")]
fn test_bad_string_multiple_decimal_points() {
BigDecimal::from_str("123.12.45").unwrap();
}
#[test]
#[should_panic(expected = "Empty")]
fn test_bad_string_only_decimal() {
BigDecimal::from_str(".").unwrap();
}
#[test]
#[should_panic(expected = "Empty")]
fn test_bad_string_only_decimal_and_exponent() {
BigDecimal::from_str(".e4").unwrap();
}
}