I wrote a recursive mergeSort
function:
func mergeSort<T: Comparable>(inout array: [T]) {
if array.count <= 1 {
return
}
var leftSlice = [T](array[0..<array.count / 2])
var rightSlice = [T](array[array.count / 2...array.endIndex - 1])
mergeSort(&leftSlice)
mergeSort(&rightSlice)
array = merge(leftSlice, rightSlice)
}
func merge<T: Comparable>(var left: [T], var right: [T]) -> [T] {
var mergedValues = [T]()
while !left.isEmpty && !right.isEmpty {
mergedValues.append(left.first! < right.first! ? left.removeAtIndex(0) : right.removeAtIndex(0))
}
if !left.isEmpty {
mergedValues += left
} else if !right.isEmpty {
mergedValues += right
}
return mergedValues
}
Now, since merge()
is only supposed to be used by mergeSort()
I placed it inside of mergeSort()
, therefore making merge()
a nested function:
func mergeSort<T: Comparable>(inout array: [T]) {
func merge<T: Comparable>(var left: [T], var right: [T]) -> [T] {
var mergedValues = [T]()
while !left.isEmpty && !right.isEmpty {
mergedValues.append(left.first! < right.first! ? left.removeAtIndex(0) : right.removeAtIndex(0))
}
if !left.isEmpty {
mergedValues += left
} else if !right.isEmpty {
mergedValues += right
}
return mergedValues
}
if array.count <= 1 {
return
}
var leftSlice = [T](array[0..<array.count / 2])
var rightSlice = [T](array[array.count / 2...array.endIndex - 1])
mergeSort(&leftSlice)
mergeSort(&rightSlice)
array = merge(leftSlice, rightSlice)
}
Now the first version works fine, but the second one doesn't.
How can that be?
Looks like you've found a bug in the compiler related to nested generic functions. Here's a reduction that also crashes the 1.2 compiler:
func f<T>(t: T) {
func g<U>(u: U) { }
}
But in this case, you don't actually need a generic version of merge
. Its generic parameter is the same as the outer function, therefore just use that:
func mergeSort<T: Comparable>(inout array: [T]) {
// no generic placeholder needed, T is the T for mergeSort
func merge(var left: [T], var right: [T]) -> [T] {
// etc.
}
}
This appears to work fine.
However, it's also worth pointing out that in your merge
function, you're calling removeAtIndex
in a loop, which is a O(n) function. This means your merge sort is not going to have the hoped-for complexity.
Here's an alternative version to consider:
func mergeSort<T: Comparable>(inout array: [T], range: Range<Int>? = nil) {
func merge(left: Range<Int>, right: Range<Int>) -> [T] {
var tmp: [T] = []
tmp.reserveCapacity(count(left) + count(right))
var l = left.startIndex, r = right.startIndex
while l != left.endIndex && r != right.endIndex {
if array[l] < array[r] {
tmp.append(array[l++])
}
else {
tmp.append(array[r++])
}
}
// where left or right may be empty, this is a no-op
tmp += source[l..<left.endIndex]
tmp += source[r..<right.endIndex]
return tmp
}
// this allows the original caller to omit the range,
// the default being the full array
let r = range ?? indices(array)
if count(r) > 1 {
let mid = r.startIndex.advancedBy(r.startIndex.distanceTo(r.endIndex)/2)
let left = r.startIndex..<mid
let right = mid..<r.endIndex
mergeSort(&array, range: left)
mergeSort(&array, range: right)
let merged = merge(left, right)
array.replaceRange(r, with: merged)
}
}
I'd also say that since merge
is probably a generically useful function in its own right, you may as well make it stand-alone rather than nesting it (similarly, partition
when implementing a quick sort). The nesting doesn't buy you anything (outside of the trick I used above of referencing the outer parameter from within it, which is probably a bad practice anyway, I mostly did it to show you can :)