mirror of
https://github.com/vlang/v.git
synced 2023-08-10 21:13:21 +03:00
math.fraction: improve documentation, remove unnecessary mut modifiers
This commit is contained in:
parent
14bba54ddc
commit
e52d35bf16
@ -7,9 +7,13 @@ import math
|
||||
import math.bits
|
||||
|
||||
// Fraction Struct
|
||||
// ---------------
|
||||
// A Fraction has a numerator (n) and a denominator (d). If the user uses
|
||||
// the helper functions in this module, then the following are guaranteed:
|
||||
// 1.
|
||||
// 1. If the user provides n and d with gcd(n, d) > 1, the fraction will
|
||||
// not be reduced automatically.
|
||||
// 2. d cannot be set to zero. The factory function will panic.
|
||||
// 3. If provided d is negative, it will be made positive. n will change as well.
|
||||
struct Fraction {
|
||||
n i64
|
||||
d i64
|
||||
@ -22,20 +26,18 @@ pub:
|
||||
// the negative denominator to positive and adjusts the numerator.
|
||||
// NOTE: Fractions created are not reduced by default.
|
||||
pub fn fraction(n, d i64) Fraction {
|
||||
if d != 0 {
|
||||
// The denominator is always guaranteed to be positive (and non-zero).
|
||||
if d < 0 {
|
||||
return fraction(-n, -d)
|
||||
} else {
|
||||
return Fraction{
|
||||
n: n
|
||||
d: d
|
||||
is_reduced: math.gcd(n, d) == 1
|
||||
}
|
||||
}
|
||||
} else {
|
||||
if d == 0 {
|
||||
panic('Denominator cannot be zero')
|
||||
}
|
||||
// The denominator is always guaranteed to be positive (and non-zero).
|
||||
if d < 0 {
|
||||
return fraction(-n, -d)
|
||||
}
|
||||
return Fraction{
|
||||
n: n
|
||||
d: d
|
||||
is_reduced: math.gcd(n, d) == 1
|
||||
}
|
||||
}
|
||||
|
||||
// To String method
|
||||
@ -51,18 +53,14 @@ pub fn (f Fraction) str() string {
|
||||
// These are implemented from Knuth, TAOCP Vol 2. Section 4.5
|
||||
//
|
||||
// Returns a correctly reduced result for both addition and subtraction
|
||||
// NOTE: requires reduced inputs
|
||||
fn general_addition_result(f1, f2 Fraction, addition bool) Fraction {
|
||||
d1 := math.gcd(f1.d, f2.d)
|
||||
// d1 happends to be 1 around 600/(pi)^2 or 61 percent of the time (Theorem 4.5.2D)
|
||||
if d1 == 1 {
|
||||
mut n := i64(0)
|
||||
num1n2d := f1.n * f2.d
|
||||
num1d2n := f1.d * f2.n
|
||||
if addition {
|
||||
n = num1n2d + num1d2n
|
||||
} else {
|
||||
n = num1n2d - num1d2n
|
||||
}
|
||||
n := if addition { num1n2d + num1d2n } else { num1n2d - num1d2n }
|
||||
return Fraction{
|
||||
n: n
|
||||
d: f1.d * f2.d
|
||||
@ -70,20 +68,15 @@ fn general_addition_result(f1, f2 Fraction, addition bool) Fraction {
|
||||
}
|
||||
}
|
||||
// Here d1 > 1.
|
||||
// Without the i64(...), t is declared as an int
|
||||
// and it does not have enough precision
|
||||
mut t := i64(0)
|
||||
term1 := f1.n * (f2.d / d1)
|
||||
term2 := f2.n * (f1.d / d1)
|
||||
if addition {
|
||||
t = term1 + term2
|
||||
} else {
|
||||
t = term1 - term2
|
||||
}
|
||||
f1den := f1.d / d1
|
||||
f2den := f2.d / d1
|
||||
term1 := f1.n * f2den
|
||||
term2 := f2.n * f1den
|
||||
t := if addition { term1 + term2 } else { term1 - term2 }
|
||||
d2 := math.gcd(t, d1)
|
||||
return Fraction{
|
||||
n: t / d2
|
||||
d: (f1.d / d1) * (f2.d / d2)
|
||||
d: f1den * (f2.d / d2)
|
||||
is_reduced: true
|
||||
}
|
||||
}
|
||||
@ -99,32 +92,32 @@ pub fn (f1 Fraction) -(f2 Fraction) Fraction {
|
||||
}
|
||||
|
||||
// Returns a correctly reduced result for both multiplication and division
|
||||
// NOTE: requires reduced inputs
|
||||
fn general_multiplication_result(f1, f2 Fraction, multiplication bool) Fraction {
|
||||
// Theorem: If f1 and f2 are reduced i.e. gcd(f1.n, f1.d) == 1 and gcd(f2.n, f2.d) == 1,
|
||||
// * Theorem: If f1 and f2 are reduced i.e. gcd(f1.n, f1.d) == 1 and gcd(f2.n, f2.d) == 1,
|
||||
// then gcd(f1.n * f2.n, f1.d * f2.d) == gcd(f1.n, f2.d) * gcd(f1.d, f2.n)
|
||||
// Knuth poses this an exercise for 4.5.1. - Exercise 2
|
||||
mut d1 := i64(0)
|
||||
mut d2 := i64(0)
|
||||
mut n := i64(0)
|
||||
mut d := i64(0)
|
||||
// * Knuth poses this an exercise for 4.5.1. - Exercise 2
|
||||
// * Also, note that:
|
||||
// The terms are flipped for multiplication and division, so the gcds must be calculated carefully
|
||||
// We do multiple divisions in order to prevent any possible overflows. Also, note that:
|
||||
// We do multiple divisions in order to prevent any possible overflows.
|
||||
// * One more thing:
|
||||
// if d = gcd(a, b) for example, then d divides both a and b
|
||||
if multiplication {
|
||||
d1 = math.gcd(f1.n, f2.d)
|
||||
d2 = math.gcd(f1.d, f2.n)
|
||||
n = (f1.n / d1) * (f2.n / d2)
|
||||
d = (f2.d / d1) * (f1.d / d2)
|
||||
d1 := math.gcd(f1.n, f2.d)
|
||||
d2 := math.gcd(f1.d, f2.n)
|
||||
return Fraction{
|
||||
n: (f1.n / d1) * (f2.n / d2)
|
||||
d: (f2.d / d1) * (f1.d / d2)
|
||||
is_reduced: true
|
||||
}
|
||||
} else {
|
||||
d1 = math.gcd(f1.n, f2.n)
|
||||
d2 = math.gcd(f1.d, f2.d)
|
||||
n = (f1.n / d1) * (f2.d / d2)
|
||||
d = (f2.n / d1) * (f1.d / d2)
|
||||
}
|
||||
return Fraction{
|
||||
n: n
|
||||
d: d
|
||||
is_reduced: true
|
||||
d1 := math.gcd(f1.n, f2.n)
|
||||
d2 := math.gcd(f1.d, f2.d)
|
||||
return Fraction{
|
||||
n: (f1.n / d1) * (f2.d / d2)
|
||||
d: (f2.n / d1) * (f1.d / d2)
|
||||
is_reduced: true
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
Loading…
Reference in New Issue
Block a user