Files
swift-mirror/validation-test/stdlib/Algorithm.swift
Nate Cook e5c1567957 [stdlib] Switch to a stable sort algorithm (#19717)
This switches the standard library's sort algorithm from an in-place
introsort to use a modified timsort, a stable, adaptive sort that
merges runs using a temporary buffer. This implementation performs
straight merges instead of adopting timsort's galloping strategy.

In addition to maintaining the relative order of equal/non-comparable
elements, this algorithm outperforms the introsort on data with any
intrinsic structure, such as runs of ascending or descending elements
or a significant number of equality collisions.
2018-11-07 00:05:04 -06:00

227 lines
6.2 KiB
Swift

// -*- swift -*-
// RUN: %target-run-simple-swift
// REQUIRES: executable_test
import StdlibUnittest
import StdlibCollectionUnittest
import SwiftPrivate
var Algorithm = TestSuite("Algorithm")
// FIXME(prext): remove this conformance.
extension String.UnicodeScalarView : Equatable {}
// FIXME(prext): remove this function.
public func == (
lhs: String.UnicodeScalarView, rhs: String.UnicodeScalarView) -> Bool {
return Array(lhs) == Array(rhs)
}
// FIXME(prext): move this struct to the point of use.
Algorithm.test("min,max") {
// Identities are unique in this set.
let a1 = MinimalComparableValue(0, identity: 1)
let a2 = MinimalComparableValue(0, identity: 2)
let a3 = MinimalComparableValue(0, identity: 3)
let b1 = MinimalComparableValue(1, identity: 4)
let b2 = MinimalComparableValue(1, identity: 5)
_ = MinimalComparableValue(1, identity: 6)
let c1 = MinimalComparableValue(2, identity: 7)
let c2 = MinimalComparableValue(2, identity: 8)
let c3 = MinimalComparableValue(2, identity: 9)
// 2-arg min()
expectEqual(a1.identity, min(a1, b1).identity)
expectEqual(a1.identity, min(b1, a1).identity)
expectEqual(a1.identity, min(a1, a2).identity)
// 2-arg max()
expectEqual(c1.identity, max(c1, b1).identity)
expectEqual(c1.identity, max(b1, c1).identity)
expectEqual(c1.identity, max(c2, c1).identity)
// 3-arg min()
expectEqual(a1.identity, min(a1, b1, c1).identity)
expectEqual(a1.identity, min(b1, a1, c1).identity)
expectEqual(a1.identity, min(c1, b1, a1).identity)
expectEqual(a1.identity, min(c1, a1, b1).identity)
expectEqual(a1.identity, min(a1, a2, a3).identity)
expectEqual(a1.identity, min(a1, a2, b1).identity)
expectEqual(a1.identity, min(a1, b1, a2).identity)
expectEqual(a1.identity, min(b1, a1, a2).identity)
// 3-arg max()
expectEqual(c1.identity, max(c1, b1, a1).identity)
expectEqual(c1.identity, max(a1, c1, b1).identity)
expectEqual(c1.identity, max(b1, a1, c1).identity)
expectEqual(c1.identity, max(b1, c1, a1).identity)
expectEqual(c1.identity, max(c3, c2, c1).identity)
expectEqual(c1.identity, max(c2, c1, b1).identity)
expectEqual(c1.identity, max(c2, b1, c1).identity)
expectEqual(c1.identity, max(b1, c2, c1).identity)
// 4-arg min()
expectEqual(a1.identity, min(a1, b1, a2, b2).identity)
expectEqual(a1.identity, min(b1, a1, a2, b2).identity)
expectEqual(a1.identity, min(c1, b1, b2, a1).identity)
expectEqual(a1.identity, min(c1, b1, a1, a2).identity)
// 4-arg max()
expectEqual(c1.identity, max(c2, b1, c1, b2).identity)
expectEqual(c1.identity, max(b1, c2, c1, b2).identity)
expectEqual(c1.identity, max(a1, b1, b2, c1).identity)
expectEqual(c1.identity, max(a1, b1, c2, c1).identity)
}
Algorithm.test("sorted/strings") {
expectEqual(
["Banana", "apple", "cherry"],
["apple", "Banana", "cherry"].sorted())
let s = ["apple", "Banana", "cherry"].sorted() {
$0.count > $1.count
}
expectEqual(["Banana", "cherry", "apple"], s)
}
// A wrapper around Array<T> that disables any type-specific algorithm
// optimizations and forces bounds checking on.
struct A<T> : MutableCollection, RandomAccessCollection {
typealias Indices = CountableRange<Int>
init(_ a: Array<T>) {
impl = a
}
var startIndex: Int {
return 0
}
var endIndex: Int {
return impl.count
}
func makeIterator() -> Array<T>.Iterator {
return impl.makeIterator()
}
subscript(i: Int) -> T {
get {
expectTrue(i >= 0 && i < impl.count)
return impl[i]
}
set (x) {
expectTrue(i >= 0 && i < impl.count)
impl[i] = x
}
}
subscript(r: Range<Int>) -> Array<T>.SubSequence {
get {
expectTrue(r.lowerBound >= 0 && r.lowerBound <= impl.count)
expectTrue(r.upperBound >= 0 && r.upperBound <= impl.count)
return impl[r]
}
set (x) {
expectTrue(r.lowerBound >= 0 && r.lowerBound <= impl.count)
expectTrue(r.upperBound >= 0 && r.upperBound <= impl.count)
impl[r] = x
}
}
var impl: Array<T>
}
func randomArray() -> A<Int> {
let count = Int.random(in: 0 ..< 50)
let array = (0 ..< count).map { _ in Int.random(in: .min ... .max) }
return A(array)
}
Algorithm.test("invalidOrderings") {
withInvalidOrderings {
let a = randomArray()
_blackHole(a.sorted(by: $0))
}
withInvalidOrderings {
var a: A<Int>
a = randomArray()
let lt = $0
let first = a.first
_ = a.partition(by: { !lt($0, first!) })
}
/*
// FIXME: Disabled due to <rdar://problem/17734737> Unimplemented:
// abstraction difference in l-value
withInvalidOrderings {
var a = randomArray()
var pred = $0
_insertionSort(&a, a.indices, &pred)
}
*/
}
// The routine is based on http://www.cs.dartmouth.edu/~doug/mdmspe.pdf
func makeQSortKiller(_ len: Int) -> [Int] {
var candidate: Int = 0
var keys = [Int: Int]()
func Compare(_ x: Int, y : Int) -> Bool {
if keys[x] == nil && keys[y] == nil {
if (x == candidate) {
keys[x] = keys.count
} else {
keys[y] = keys.count
}
}
if keys[x] == nil {
candidate = x
return true
}
if keys[y] == nil {
candidate = y
return false
}
return keys[x]! > keys[y]!
}
var ary = [Int](repeating: 0, count: len)
var ret = [Int](repeating: 0, count: len)
for i in 0..<len { ary[i] = i }
ary = ary.sorted(by: Compare)
for i in 0..<len {
ret[ary[i]] = i
}
return ret
}
Algorithm.test("sorted/complexity") {
var ary: [Int] = []
// Check performance of sorting an array of repeating values.
var comparisons_100 = 0
ary = [Int](repeating: 0, count: 100)
ary.sort { comparisons_100 += 1; return $0 < $1 }
var comparisons_1000 = 0
ary = [Int](repeating: 0, count: 1000)
ary.sort { comparisons_1000 += 1; return $0 < $1 }
expectTrue(comparisons_1000/comparisons_100 < 20)
// Try to construct 'bad' case for quicksort, on which the algorithm
// goes quadratic.
comparisons_100 = 0
ary = makeQSortKiller(100)
ary.sort { comparisons_100 += 1; return $0 < $1 }
comparisons_1000 = 0
ary = makeQSortKiller(1000)
ary.sort { comparisons_1000 += 1; return $0 < $1 }
expectTrue(comparisons_1000/comparisons_100 < 20)
}
Algorithm.test("sorted/return type") {
let _: Array = ([5, 4, 3, 2, 1] as ArraySlice).sorted()
}
runAllTests()