Nim/lib/pure/algorithm.nim

#
#
#            Nim's Runtime Library
#        (c) Copyright 2015 Andreas Rumpf
#
#    See the file "copying.txt", included in this
#    distribution, for details about the copyright.
#

## This module implements some common generic algorithms on `openArray`s.
##
## Basic usage
## ===========
##

runnableExamples:
  type People = tuple
    year: int
    name: string

  var a: seq[People]

  a.add((2000, "John"))
  a.add((2005, "Marie"))
  a.add((2010, "Jane"))

  # Sorting with default system.cmp
  a.sort()
  assert a == @[(year: 2000, name: "John"), (year: 2005, name: "Marie"),
                (year: 2010, name: "Jane")]

  proc myCmp(x, y: People): int =
    cmp(x.name, y.name)

  # Sorting with custom proc
  a.sort(myCmp)
  assert a == @[(year: 2010, name: "Jane"), (year: 2000, name: "John"),
                (year: 2005, name: "Marie")]

## See also
## ========
## * `sequtils module<sequtils.html>`_ for working with the built-in seq type
## * `tables module<tables.html>`_ for sorting tables

import std/private/since

when defined(nimPreviewSlimSystem):
  import std/assertions


type
  SortOrder* = enum
    Descending, Ascending

proc `*`*(x: int, order: SortOrder): int {.inline.} =
  ## Flips the sign of `x` if `order == Descending`.
  ## If `order == Ascending` then `x` is returned.
  ##
  ## `x` is supposed to be the result of a comparator, i.e.
  ## | `< 0` for *less than*,
  ## | `== 0` for *equal*,
  ## | `> 0` for *greater than*.
  runnableExamples:
    assert -123 * Descending == 123
    assert 123 * Descending == -123
    assert -123 * Ascending == -123
    assert 123 * Ascending == 123
  var y = order.ord - 1
  result = (x xor y) - y

template fillImpl[T](a: var openArray[T], first, last: int, value: T) =
  var x = first
  while x <= last:
    a[x] = value
    inc(x)

proc fill*[T](a: var openArray[T], first, last: Natural, value: T) =
  ## Assigns `value` to all elements of the slice `a[first..last]`.
  ##
  ## If an invalid range is passed, it raises `IndexDefect`.
  runnableExamples:
    var a: array[6, int]
    a.fill(1, 3, 9)
    assert a == [0, 9, 9, 9, 0, 0]
    a.fill(3, 5, 7)
    assert a == [0, 9, 9, 7, 7, 7]
    doAssertRaises(IndexDefect, a.fill(1, 7, 9))
  fillImpl(a, first, last, value)

proc fill*[T](a: var openArray[T], value: T) =
  ## Assigns `value` to all elements of the container `a`.
  runnableExamples:
    var a: array[6, int]
    a.fill(9)
    assert a == [9, 9, 9, 9, 9, 9]
    a.fill(4)
    assert a == [4, 4, 4, 4, 4, 4]
  fillImpl(a, 0, a.high, value)


proc reverse*[T](a: var openArray[T], first, last: Natural) =
  ## Reverses the slice `a[first..last]`.
  ##
  ## If an invalid range is passed, it raises `IndexDefect`.
  ##
  ## **See also:**
  ## * `reversed proc<#reversed,openArray[T],Natural,int>`_ reverse a slice and returns a `seq[T]`
  ## * `reversed proc<#reversed,openArray[T]>`_ reverse and returns a `seq[T]`
  runnableExamples:
    var a = [1, 2, 3, 4, 5, 6]
    a.reverse(1, 3)
    assert a == [1, 4, 3, 2, 5, 6]
    a.reverse(1, 3)
    assert a == [1, 2, 3, 4, 5, 6]
    doAssertRaises(IndexDefect, a.reverse(1, 7))
  var x = first
  var y = last
  while x < y:
    swap(a[x], a[y])
    dec(y)
    inc(x)

proc reverse*[T](a: var openArray[T]) =
  ## Reverses the contents of the container `a`.
  ##
  ## **See also:**
  ## * `reversed proc<#reversed,openArray[T],Natural,int>`_ reverse a slice and returns a `seq[T]`
  ## * `reversed proc<#reversed,openArray[T]>`_ reverse and returns a `seq[T]`
  runnableExamples:
    var a = [1, 2, 3, 4, 5, 6]
    a.reverse()
    assert a == [6, 5, 4, 3, 2, 1]
    a.reverse()
    assert a == [1, 2, 3, 4, 5, 6]
  # the max is needed, since a.high is -1 if a is empty
  reverse(a, 0, max(0, a.high))

proc reversed*[T](a: openArray[T]): seq[T] {.inline.} =
  ## Returns the elements of `a` in reverse order.
  ##
  ## **See also:**
  ## * `reverse proc<#reverse,openArray[T]>`_
  runnableExamples:
    assert [10, 11, 12].reversed == @[12, 11, 10]
    assert seq[string].default.reversed == @[]
  result = @[]
  let n = a.len
  result.setLen(n)
  for i in 0..<n: result[i] = a[n - (i + 1)]

proc reversed*[T](a: openArray[T], first: Natural, last: int): seq[T]
  {.inline, deprecated: "use: `reversed(toOpenArray(a, first, last))`".} =
  reversed(toOpenArray(a, first, last))

when defined(nimHasEffectsOf):
  {.experimental: "strictEffects".}
else:
  {.pragma: effectsOf.}

proc binarySearch*[T, K](a: openArray[T], key: K,
                         cmp: proc (x: T, y: K): int {.closure.}): int {.effectsOf: cmp.} =
  ## Binary search for `key` in `a`. Return the index of `key` or -1 if not found.
  ## Assumes that `a` is sorted according to `cmp`.
  ##
  ## `cmp` is the comparator function to use, the expected return values are
  ## the same as those of system.cmp.
  runnableExamples:
    assert binarySearch(["a", "b", "c", "d"], "d", system.cmp[string]) == 3
    assert binarySearch(["a", "b", "c", "d"], "c", system.cmp[string]) == 2
  let len = a.len

  if len == 0:
    return -1

  if len == 1:
    if cmp(a[0], key) == 0:
      return 0
    else:
      return -1

  result = 0
  if (len and (len - 1)) == 0:
    # when `len` is a power of 2, a faster shr can be used.
    var step = len shr 1
    var cmpRes: int
    while step > 0:
      let i = result or step
      cmpRes = cmp(a[i], key)
      if cmpRes == 0:
        return i

      if cmpRes < 0:
        result = i
      step = step shr 1
    if cmp(a[result], key) != 0: result = -1
  else:
    var b = len
    var cmpRes: int
    while result < b:
      var mid = (result + b) shr 1
      cmpRes = cmp(a[mid], key)
      if cmpRes == 0:
        return mid

      if cmpRes < 0:
        result = mid + 1
      else:
        b = mid
    if result >= len or cmp(a[result], key) != 0: result = -1

proc binarySearch*[T](a: openArray[T], key: T): int =
  ## Binary search for `key` in `a`. Return the index of `key` or -1 if not found.
  ## Assumes that `a` is sorted.
  runnableExamples:
    assert binarySearch([0, 1, 2, 3, 4], 4) == 4
    assert binarySearch([0, 1, 2, 3, 4], 2) == 2
  binarySearch(a, key, cmp[T])

const
  onlySafeCode = true

proc lowerBound*[T, K](a: openArray[T], key: K,
                       cmp: proc(x: T, k: K): int {.closure.}): int {.effectsOf: cmp.} =
  ## Returns the index of the first element in `a` that is not less than
  ## (i.e. greater or equal to) `key`, or last if no such element is found.
  ## In other words if you have a sorted sequence and you call
  ## `insert(thing, elm, lowerBound(thing, elm))`
  ## the sequence will still be sorted.
  ## Assumes that `a` is sorted according to `cmp`.
  ##
  ## If an invalid range is passed, it raises `IndexDefect`.
  ##
  ## This version uses `cmp` to compare the elements.
  ## The expected return values are the same as those of `system.cmp`.
  ##
  ## **See also:**
  ## * `upperBound proc<#upperBound,openArray[T],K,proc(T,K)>`_ sorted by `cmp` in the specified order
  ## * `upperBound proc<#upperBound,openArray[T],T>`_
  runnableExamples:
    var arr = @[1, 2, 3, 5, 6, 7, 8, 9]
    assert arr.lowerBound(3, system.cmp[int]) == 2
    assert arr.lowerBound(4, system.cmp[int]) == 3
    assert arr.lowerBound(5, system.cmp[int]) == 3
    arr.insert(4, arr.lowerBound(4, system.cmp[int]))
    assert arr == [1, 2, 3, 4, 5, 6, 7, 8, 9]
  result = a.low
  var count = a.high - a.low + 1
  var step, pos: int
  while count != 0:
    step = count shr 1
    pos = result + step
    if cmp(a[pos], key) < 0:
      result = pos + 1
      count -= step + 1
    else:
      count = step

proc lowerBound*[T](a: openArray[T], key: T): int = lowerBound(a, key, cmp[T])
  ## Returns the index of the first element in `a` that is not less than
  ## (i.e. greater or equal to) `key`, or last if no such element is found.
  ## In other words if you have a sorted sequence and you call
  ## `insert(thing, elm, lowerBound(thing, elm))`
  ## the sequence will still be sorted.
  ## Assumes that `a` is sorted.
  ##
  ## This version uses the default comparison function `cmp`.
  ##
  ## **See also:**
  ## * `upperBound proc<#upperBound,openArray[T],K,proc(T,K)>`_ sorted by `cmp` in the specified order
  ## * `upperBound proc<#upperBound,openArray[T],T>`_

proc upperBound*[T, K](a: openArray[T], key: K,
                       cmp: proc(x: T, k: K): int {.closure.}): int {.effectsOf: cmp.} =
  ## Returns the index of the first element in `a` that is greater than
  ## `key`, or last if no such element is found.
  ## In other words if you have a sorted sequence and you call
  ## `insert(thing, elm, upperBound(thing, elm))`
  ## the sequence will still be sorted.
  ## Assumes that `a` is sorted according to `cmp`.
  ##
  ## If an invalid range is passed, it raises `IndexDefect`.
  ##
  ## This version uses `cmp` to compare the elements. The expected
  ## return values are the same as those of `system.cmp`.
  ##
  ## **See also:**
  ## * `lowerBound proc<#lowerBound,openArray[T],K,proc(T,K)>`_ sorted by `cmp` in the specified order
  ## * `lowerBound proc<#lowerBound,openArray[T],T>`_
  runnableExamples:
    var arr = @[1, 2, 3, 5, 6, 7, 8, 9]
    assert arr.upperBound(2, system.cmp[int]) == 2
    assert arr.upperBound(3, system.cmp[int]) == 3
    assert arr.upperBound(4, system.cmp[int]) == 3
    arr.insert(4, arr.upperBound(3, system.cmp[int]))
    assert arr == [1, 2, 3, 4, 5, 6, 7, 8, 9]
  result = a.low
  var count = a.high - a.low + 1
  var step, pos: int
  while count != 0:
    step = count shr 1
    pos = result + step
    if cmp(a[pos], key) <= 0:
      result = pos + 1
      count -= step + 1
    else:
      count = step

proc upperBound*[T](a: openArray[T], key: T): int = upperBound(a, key, cmp[T])
  ## Returns the index of the first element in `a` that is greater than
  ## `key`, or last if no such element is found.
  ## In other words if you have a sorted sequence and you call
  ## `insert(thing, elm, upperBound(thing, elm))`
  ## the sequence will still be sorted.
  ## Assumes that `a` is sorted.
  ##
  ## This version uses the default comparison function `cmp`.
  ##
  ## **See also:**
  ## * `lowerBound proc<#lowerBound,openArray[T],K,proc(T,K)>`_ sorted by `cmp` in the specified order
  ## * `lowerBound proc<#lowerBound,openArray[T],T>`_

template `<-`(a, b) =
  when defined(gcDestructors):
    a = move b
  elif onlySafeCode:
    shallowCopy(a, b)
  else:
    copyMem(addr(a), addr(b), sizeof(T))

proc mergeAlt[T](a, b: var openArray[T], lo, m, hi: int,
              cmp: proc (x, y: T): int {.closure.}, order: SortOrder) {.effectsOf: cmp.} =
  # Optimization: If max(left) <= min(right) there is nothing to do!
  # 1 2 3 4 ## 5 6 7 8
  # -> O(n) for sorted arrays.
  # On random data this saves up to 40% of mergeAlt calls.
  if cmp(a[m], a[m+1]) * order <= 0: return
  var j = lo
  # copy a[j..m] into b:
  assert j <= m
  when onlySafeCode:
    var bb = 0
    while j <= m:
      b[bb] <- a[j]
      inc(bb)
      inc(j)
  else:
    copyMem(addr(b[0]), addr(a[j]), sizeof(T)*(m-j+1))
    j = m+1
  var i = 0
  var k = lo
  # copy proper element back:
  while k < j and j <= hi:
    if cmp(b[i], a[j]) * order <= 0:
      a[k] <- b[i]
      inc(i)
    else:
      a[k] <- a[j]
      inc(j)
    inc(k)
  # copy rest of b:
  when onlySafeCode:
    while k < j:
      a[k] <- b[i]
      inc(k)
      inc(i)
  else:
    if k < j: copyMem(addr(a[k]), addr(b[i]), sizeof(T)*(j-k))

func sort*[T](a: var openArray[T],
              cmp: proc (x, y: T): int {.closure.},
              order = SortOrder.Ascending) {.effectsOf: cmp.} =
  ## Default Nim sort (an implementation of merge sort). The sorting
  ## is guaranteed to be stable (that is, equal elements stay in the same order)
  ## and the worst case is guaranteed to be O(n log n).
  ## Sorts by `cmp` in the specified `order`.
  ##
  ## The current implementation uses an iterative
  ## mergesort to achieve this. It uses a temporary sequence of
  ## length `a.len div 2`. If you do not wish to provide your own
  ## `cmp`, you may use `system.cmp` or instead call the overloaded
  ## version of `sort`, which uses `system.cmp`.
  ##
  ##   ```nim
  ##   sort(myIntArray, system.cmp[int])
  ##   # do not use cmp[string] here as we want to use the specialized
  ##   # overload:
  ##   sort(myStrArray, system.cmp)
  ##   ```
  ##
  ## You can inline adhoc comparison procs with the `do notation
  ## <manual.html#procedures-do-notation>`_. Example:
  ##
  ##   ```nim
  ##   people.sort do (x, y: Person) -> int:
  ##     result = cmp(x.surname, y.surname)
  ##     if result == 0:
  ##       result = cmp(x.name, y.name)
  ##   ```
  ##
  ## **See also:**
  ## * `sort proc<#sort,openArray[T]>`_
  ## * `sorted proc<#sorted,openArray[T],proc(T,T)>`_ sorted by `cmp` in the specified order
  ## * `sorted proc<#sorted,openArray[T]>`_
  ## * `sortedByIt template<#sortedByIt.t,untyped,untyped>`_
  runnableExamples:
    var d = ["boo", "fo", "barr", "qux"]
    proc myCmp(x, y: string): int =
      if x.len() > y.len() or x.len() == y.len(): 1
      else: -1
    sort(d, myCmp)
    assert d == ["fo", "qux", "boo", "barr"]
  var n = a.len
  var b = newSeq[T](n div 2)
  var s = 1
  while s < n:
    var m = n-1-s
    while m >= 0:
      mergeAlt(a, b, max(m-s+1, 0), m, m+s, cmp, order)
      dec(m, s*2)
    s = s*2

proc sort*[T](a: var openArray[T], order = SortOrder.Ascending) = sort[T](a,
    system.cmp[T], order)
  ## Shortcut version of `sort` that uses `system.cmp[T]` as the comparison function.
  ##
  ## **See also:**
  ## * `sort func<#sort,openArray[T],proc(T,T)>`_
  ## * `sorted proc<#sorted,openArray[T],proc(T,T)>`_ sorted by `cmp` in the specified order
  ## * `sorted proc<#sorted,openArray[T]>`_
  ## * `sortedByIt template<#sortedByIt.t,untyped,untyped>`_

proc sorted*[T](a: openArray[T], cmp: proc(x, y: T): int {.closure.},
                order = SortOrder.Ascending): seq[T] {.effectsOf: cmp.} =
  ## Returns `a` sorted by `cmp` in the specified `order`.
  ##
  ## **See also:**
  ## * `sort func<#sort,openArray[T],proc(T,T)>`_
  ## * `sort proc<#sort,openArray[T]>`_
  ## * `sortedByIt template<#sortedByIt.t,untyped,untyped>`_
  runnableExamples:
    let
      a = [2, 3, 1, 5, 4]
      b = sorted(a, system.cmp[int])
      c = sorted(a, system.cmp[int], Descending)
      d = sorted(["adam", "dande", "brian", "cat"], system.cmp[string])
    assert b == @[1, 2, 3, 4, 5]
    assert c == @[5, 4, 3, 2, 1]
    assert d == @["adam", "brian", "cat", "dande"]
  result = newSeq[T](a.len)
  for i in 0 .. a.high:
    result[i] = a[i]
  sort(result, cmp, order)

proc sorted*[T](a: openArray[T], order = SortOrder.Ascending): seq[T] =
  ## Shortcut version of `sorted` that uses `system.cmp[T]` as the comparison function.
  ##
  ## **See also:**
  ## * `sort func<#sort,openArray[T],proc(T,T)>`_
  ## * `sort proc<#sort,openArray[T]>`_
  ## * `sortedByIt template<#sortedByIt.t,untyped,untyped>`_
  runnableExamples:
    let
      a = [2, 3, 1, 5, 4]
      b = sorted(a)
      c = sorted(a, Descending)
      d = sorted(["adam", "dande", "brian", "cat"])
    assert b == @[1, 2, 3, 4, 5]
    assert c == @[5, 4, 3, 2, 1]
    assert d == @["adam", "brian", "cat", "dande"]
  sorted[T](a, system.cmp[T], order)

template sortedByIt*(seq1, op: untyped): untyped =
  ## Convenience template around the `sorted` proc to reduce typing.
  ##
  ## The template injects the `it` variable which you can use directly in an
  ## expression.
  ##
  ## Because the underlying `cmp()` is defined for tuples you can also do
  ## a nested sort.
  ##
  ## **See also:**
  ## * `sort func<#sort,openArray[T],proc(T,T)>`_
  ## * `sort proc<#sort,openArray[T]>`_
  ## * `sorted proc<#sorted,openArray[T],proc(T,T)>`_ sorted by `cmp` in the specified order
  ## * `sorted proc<#sorted,openArray[T]>`_
  runnableExamples:
    type Person = tuple[name: string, age: int]
    var
      p1: Person = (name: "p1", age: 60)
      p2: Person = (name: "p2", age: 20)
      p3: Person = (name: "p3", age: 30)
      p4: Person = (name: "p4", age: 30)
      people = @[p1, p2, p4, p3]

    assert people.sortedByIt(it.name) == @[(name: "p1", age: 60), (name: "p2",
        age: 20), (name: "p3", age: 30), (name: "p4", age: 30)]
    # Nested sort
    assert people.sortedByIt((it.age, it.name)) == @[(name: "p2", age: 20),
       (name: "p3", age: 30), (name: "p4", age: 30), (name: "p1", age: 60)]
  var result = sorted(seq1, proc(x, y: typeof(items(seq1), typeOfIter)): int =
    var it {.inject.} = x
    let a = op
    it = y
    let b = op
    result = cmp(a, b))
  result

func isSorted*[T](a: openArray[T],
                 cmp: proc(x, y: T): int {.closure.},
                 order = SortOrder.Ascending): bool {.effectsOf: cmp.} =
  ## Checks to see whether `a` is already sorted in `order`
  ## using `cmp` for the comparison. The parameters are identical
  ## to `sort`. Requires O(n) time.
  ##
  ## **See also:**
  ## * `isSorted proc<#isSorted,openArray[T]>`_
  runnableExamples:
    let
      a = [2, 3, 1, 5, 4]
      b = [1, 2, 3, 4, 5]
      c = [5, 4, 3, 2, 1]
      d = ["adam", "brian", "cat", "dande"]
      e = ["adam", "dande", "brian", "cat"]
    assert isSorted(a) == false
    assert isSorted(b) == true
    assert isSorted(c) == false
    assert isSorted(c, Descending) == true
    assert isSorted(d) == true
    assert isSorted(e) == false
  result = true
  for i in 0..<len(a)-1:
    if cmp(a[i], a[i+1]) * order > 0:
      return false

proc isSorted*[T](a: openArray[T], order = SortOrder.Ascending): bool =
  ## Shortcut version of `isSorted` that uses `system.cmp[T]` as the comparison function.
  ##
  ## **See also:**
  ## * `isSorted func<#isSorted,openArray[T],proc(T,T)>`_
  runnableExamples:
    let
      a = [2, 3, 1, 5, 4]
      b = [1, 2, 3, 4, 5]
      c = [5, 4, 3, 2, 1]
      d = ["adam", "brian", "cat", "dande"]
      e = ["adam", "dande", "brian", "cat"]
    assert isSorted(a) == false
    assert isSorted(b) == true
    assert isSorted(c) == false
    assert isSorted(c, Descending) == true
    assert isSorted(d) == true
    assert isSorted(e) == false
  isSorted(a, system.cmp[T], order)

proc merge*[T](
  result: var seq[T],
  x, y: openArray[T], cmp: proc(x, y: T): int {.closure.}
) {.since: (1, 5, 1), effectsOf: cmp.} =
  ## Merges two sorted `openArray`. `x` and `y` are assumed to be sorted.
  ## If you do not wish to provide your own `cmp`,
  ## you may use `system.cmp` or instead call the overloaded
  ## version of `merge`, which uses `system.cmp`.
  ##
  ## .. note:: The original data of `result` is not cleared,
  ##    new data is appended to `result`.
  ##
  ## **See also:**
  ## * `merge proc<#merge,seq[T],openArray[T],openArray[T]>`_
  runnableExamples:
    let x = @[1, 3, 6]
    let y = @[2, 3, 4]

    block:
      var merged = @[7] # new data is appended to merged sequence
      merged.merge(x, y, system.cmp[int])
      assert merged == @[7, 1, 2, 3, 3, 4, 6]

    block:
      var merged = @[7] # if you only want new data, clear merged sequence first
      merged.setLen(0)
      merged.merge(x, y, system.cmp[int])
      assert merged.isSorted
      assert merged == @[1, 2, 3, 3, 4, 6]

    import std/sugar

    var res: seq[(int, int)]
    res.merge([(1, 1)], [(1, 2)], (a, b) => a[0] - b[0])
    assert res == @[(1, 1), (1, 2)]

    assert seq[int].default.dup(merge([1, 3], [2, 4])) == @[1, 2, 3, 4]

  let
    sizeX = x.len
    sizeY = y.len
    oldLen = result.len

  result.setLen(oldLen + sizeX + sizeY)

  var
    ix = 0
    iy = 0
    i = oldLen

  while true:
    if ix == sizeX:
      while iy < sizeY:
        result[i] = y[iy]
        inc i
        inc iy
      return

    if iy == sizeY:
      while ix < sizeX:
        result[i] = x[ix]
        inc i
        inc ix
      return

    let itemX = x[ix]
    let itemY = y[iy]

    if cmp(itemX, itemY) > 0: # to have a stable sort
      result[i] = itemY
      inc iy
    else:
      result[i] = itemX
      inc ix

    inc i

proc merge*[T](result: var seq[T], x, y: openArray[T]) {.inline, since: (1, 5, 1).} =
  ## Shortcut version of `merge` that uses `system.cmp[T]` as the comparison function.
  ##
  ## **See also:**
  ## * `merge proc<#merge,seq[T],openArray[T],openArray[T],proc(T,T)>`_
  runnableExamples:
    let x = [5, 10, 15, 20, 25]
    let y = [50, 40, 30, 20, 10].sorted

    var merged: seq[int]
    merged.merge(x, y)
    assert merged.isSorted
    assert merged == @[5, 10, 10, 15, 20, 20, 25, 30, 40, 50]
  merge(result, x, y, system.cmp)

proc product*[T](x: openArray[seq[T]]): seq[seq[T]] =
  ## Produces the Cartesian product of the array.
  ## Every element of the result is a combination of one element from each seq in `x`,
  ## with the ith element coming from `x[i]`.
  ##
  ## .. warning:: complexity may explode.
  runnableExamples:
    assert product(@[@[1], @[2]]) == @[@[1, 2]]
    assert product(@[@["A", "K"], @["Q"]]) == @[@["K", "Q"], @["A", "Q"]]
  let xLen = x.len
  result = newSeq[seq[T]]()
  if xLen == 0:
    return
  if xLen == 1:
    result = @x
    return
  var
    indices = newSeq[int](xLen)
    initial = newSeq[int](xLen)
    index = 0
  var next = newSeq[T](xLen)
  for i in 0 ..< xLen:
    if len(x[i]) == 0: return
    initial[i] = len(x[i]) - 1
  indices = initial
  while true:
    while indices[index] == -1:
      indices[index] = initial[index]
      index += 1
      if index == xLen: return
      indices[index] -= 1
    for ni, i in indices:
      next[ni] = x[ni][i]
    result.add(next)
    index = 0
    indices[index] -= 1

proc nextPermutation*[T](x: var openArray[T]): bool {.discardable.} =
  ## Calculates the next lexicographic permutation, directly modifying `x`.
  ## The result is whether a permutation happened, otherwise we have reached
  ## the last-ordered permutation.
  ##
  ## If you start with an unsorted array/seq, the repeated permutations
  ## will **not** give you all permutations but stop with the last.
  ##
  ## **See also:**
  ## * `prevPermutation proc<#prevPermutation,openArray[T]>`_
  runnableExamples:
    var v = @[0, 1, 2, 3]
    assert v.nextPermutation() == true
    assert v == @[0, 1, 3, 2]
    assert v.nextPermutation() == true
    assert v == @[0, 2, 1, 3]
    assert v.prevPermutation() == true
    assert v == @[0, 1, 3, 2]
    v = @[3, 2, 1, 0]
    assert v.nextPermutation() == false
    assert v == @[3, 2, 1, 0]
  if x.len < 2:
    return false

  var i = x.high
  while i > 0 and x[i-1] >= x[i]:
    dec i

  if i == 0:
    return false

  var j = x.high
  while j >= i and x[j] <= x[i-1]:
    dec j

  swap x[j], x[i-1]
  x.reverse(i, x.high)

  result = true

proc prevPermutation*[T](x: var openArray[T]): bool {.discardable.} =
  ## Calculates the previous lexicographic permutation, directly modifying
  ## `x`. The result is whether a permutation happened, otherwise we have
  ## reached the first-ordered permutation.
  ##
  ## **See also:**
  ## * `nextPermutation proc<#nextPermutation,openArray[T]>`_
  runnableExamples:
    var v = @[0, 1, 2, 3]
    assert v.prevPermutation() == false
    assert v == @[0, 1, 2, 3]
    assert v.nextPermutation() == true
    assert v == @[0, 1, 3, 2]
    assert v.prevPermutation() == true
    assert v == @[0, 1, 2, 3]
  if x.len < 2:
    return false

  var i = x.high
  while i > 0 and x[i-1] <= x[i]:
    dec i

  if i == 0:
    return false

  x.reverse(i, x.high)

  var j = x.high
  while j >= i and x[j-1] < x[i-1]:
    dec j

  swap x[i-1], x[j]

  result = true

proc rotateInternal[T](arg: var openArray[T]; first, middle, last: int): int =
  ## A port of std::rotate from C++.
  ## Ported from [this reference](https://www.cplusplus.com/reference/algorithm/rotate/).
  result = first + last - middle

  if first == middle or middle == last:
    return

  assert first < middle
  assert middle < last

  # m prefix for mutable
  var
    mFirst = first
    mMiddle = middle
    next = middle

  swap(arg[mFirst], arg[next])
  mFirst += 1
  next += 1
  if mFirst == mMiddle:
    mMiddle = next

  while next != last:
    swap(arg[mFirst], arg[next])
    mFirst += 1
    next += 1
    if mFirst == mMiddle:
      mMiddle = next

  next = mMiddle
  while next != last:
    swap(arg[mFirst], arg[next])
    mFirst += 1
    next += 1
    if mFirst == mMiddle:
      mMiddle = next
    elif next == last:
      next = mMiddle

proc rotatedInternal[T](arg: openArray[T]; first, middle, last: int): seq[T] =
  let argLen = arg.len
  result = newSeq[T](argLen)
  for i in 0 ..< first:
    result[i] = arg[i]
  let n = last - middle
  let m = middle - first
  for i in 0 ..< n:
    result[first+i] = arg[middle+i]
  for i in 0 ..< m:
    result[first+n+i] = arg[first+i]
  for i in last ..< argLen:
    result[i] = arg[i]

proc rotateLeft*[T](arg: var openArray[T]; slice: HSlice[int, int];
                    dist: int): int {.discardable.} =
  ## Performs a left rotation on a range of elements. If you want to rotate
  ## right, use a negative `dist`. Specifically, `rotateLeft` rotates
  ## the elements at `slice` by `dist` positions.
  ##
  ## | The element at index `slice.a + dist` will be at index `slice.a`.
  ## | The element at index `slice.b` will be at `slice.a + dist - 1`.
  ## | The element at index `slice.a` will be at `slice.b + 1 - dist`.
  ## | The element at index `slice.a + dist - 1` will be at `slice.b`.
  ##
  ## Elements outside of `slice` will be left unchanged.
  ## The time complexity is linear to `slice.b - slice.a + 1`.
  ## If an invalid range (`HSlice`) is passed, it raises `IndexDefect`.
  ##
  ## `slice`
  ## : The indices of the element range that should be rotated.
  ##
  ## `dist`
  ## : The distance in amount of elements that the data should be rotated.
  ##   Can be negative, can be any number.
  ##
  ## **See also:**
  ## * `rotateLeft proc<#rotateLeft,openArray[T],int>`_ for a version which rotates the whole container
  ## * `rotatedLeft proc<#rotatedLeft,openArray[T],HSlice[int,int],int>`_ for a version which returns a `seq[T]`
  runnableExamples:
    var a = [0, 1, 2, 3, 4, 5]
    a.rotateLeft(1 .. 4, 3)
    assert a == [0, 4, 1, 2, 3, 5]
    a.rotateLeft(1 .. 4, 3)
    assert a == [0, 3, 4, 1, 2, 5]
    a.rotateLeft(1 .. 4, -3)
    assert a == [0, 4, 1, 2, 3, 5]
    doAssertRaises(IndexDefect, a.rotateLeft(1 .. 7, 2))
  let sliceLen = slice.b + 1 - slice.a
  let distLeft = ((dist mod sliceLen) + sliceLen) mod sliceLen
  arg.rotateInternal(slice.a, slice.a + distLeft, slice.b + 1)

proc rotateLeft*[T](arg: var openArray[T]; dist: int): int {.discardable.} =
  ## Same as `rotateLeft`, but with default arguments for slice,
  ## so that this procedure operates on the entire
  ## `arg`, and not just on a part of it.
  ##
  ## **See also:**
  ## * `rotateLeft proc<#rotateLeft,openArray[T],HSlice[int,int],int>`_ for a version which rotates a range
  ## * `rotatedLeft proc<#rotatedLeft,openArray[T],int>`_ for a version which returns a `seq[T]`
  runnableExamples:
    var a = [1, 2, 3, 4, 5]
    a.rotateLeft(2)
    assert a == [3, 4, 5, 1, 2]
    a.rotateLeft(4)
    assert a == [2, 3, 4, 5, 1]
    a.rotateLeft(-6)
    assert a == [1, 2, 3, 4, 5]
  let argLen = arg.len
  let distLeft = ((dist mod argLen) + argLen) mod argLen
  arg.rotateInternal(0, distLeft, argLen)

proc rotatedLeft*[T](arg: openArray[T]; slice: HSlice[int, int],
                     dist: int): seq[T] =
  ## Same as `rotateLeft`, just with the difference that it does
  ## not modify the argument. It creates a new `seq` instead.
  ##
  ## Elements outside of `slice` will be left unchanged.
  ## If an invalid range (`HSlice`) is passed, it raises `IndexDefect`.
  ##
  ## `slice`
  ## : The indices of the element range that should be rotated.
  ##
  ## `dist`
  ## : The distance in amount of elements that the data should be rotated.
  ##   Can be negative, can be any number.
  ##
  ## **See also:**
  ## * `rotateLeft proc<#rotateLeft,openArray[T],HSlice[int,int],int>`_ for the in-place version of this proc
  ## * `rotatedLeft proc<#rotatedLeft,openArray[T],int>`_ for a version which rotates the whole container
  runnableExamples:
    var a = @[1, 2, 3, 4, 5]
    a = rotatedLeft(a, 1 .. 4, 3)
    assert a == @[1, 5, 2, 3, 4]
    a = rotatedLeft(a, 1 .. 3, 2)
    assert a == @[1, 3, 5, 2, 4]
    a = rotatedLeft(a, 1 .. 3, -2)
    assert a == @[1, 5, 2, 3, 4]
  let sliceLen = slice.b + 1 - slice.a
  let distLeft = ((dist mod sliceLen) + sliceLen) mod sliceLen
  arg.rotatedInternal(slice.a, slice.a + distLeft, slice.b + 1)

proc rotatedLeft*[T](arg: openArray[T]; dist: int): seq[T] =
  ## Same as `rotateLeft`, just with the difference that it does
  ## not modify the argument. It creates a new `seq` instead.
  ##
  ## **See also:**
  ## * `rotateLeft proc<#rotateLeft,openArray[T],int>`_ for the in-place version of this proc
  ## * `rotatedLeft proc<#rotatedLeft,openArray[T],HSlice[int,int],int>`_ for a version which rotates a range
  runnableExamples:
    var a = @[1, 2, 3, 4, 5]
    a = rotatedLeft(a, 2)
    assert a == @[3, 4, 5, 1, 2]
    a = rotatedLeft(a, 4)
    assert a == @[2, 3, 4, 5, 1]
    a = rotatedLeft(a, -6)
    assert a == @[1, 2, 3, 4, 5]
  let argLen = arg.len
  let distLeft = ((dist mod argLen) + argLen) mod argLen
  arg.rotatedInternal(0, distLeft, argLen)