swapping adjacent nodes of a LinkedList - algorithm

I have to swap two adjacent node(not their data) in a linked list.
e.g.
1) Input a->b->c->d->e->f, Output : b->a->d->c->f->e
2) Input a->b->c->d->e, Output : b->a->d->c->e
I have writen the following code is there any more efficient way (maybe with two temporary pointers) or simple logic?
node* swap(node* head) {
node *first = head;
node *second,*third,*result;
if(head == NULL || head->next == NULL) {
return head;
}
result = second = first->next;
third = second->next;
while(second != NULL) {
second->next=first;
first->next=(third->next==NULL ? third : third->next);
first=third;
second=(third->next==NULL ? third : third->next);
third=(second==NULL ? second : second->next);
}
return result;
}

Looks good. I added one correctness check (third==NULL) and removed one redundant expression. You are going through the whole list only once, which you have to do. So I think we can be pretty certain that this is the fastest way to do it.
node* swap(node* head) {
node *first = head;
node *second,*third,*result;
if(head == NULL || head->next == NULL) {
return head;
}
result = second = first->next;
third = second->next;
while(second != NULL) {
second->next=first;
second = first->next=((third==NULL || third->next==NULL) ? third : third->next);
first=third;
third=(second==NULL ? second : second->next);
}
return result;
}

You can do this fairly simply with a recursion:
// Swaps node b and c.
void swapTwo(node* a, node* b, node* c) {
if (a != NULL)
a->next = c;
b->next = c->next;
c->next = b;
}
void swapEveryTwo(node* prev, node* node) {
if (node != null && node->next != null) {
swapTwo(prev, node, node->next);
swapEveryTwo(node->next, node->next->next);
}
}
Every call of swapEveryTwo swaps pairs of nodes, and then sets up the recursion for the next pair. Also, because this function is tail recursive, the compiler will undoubtedly optimize it to a while loop, ensuring no extra stack frames are allocated, and thus will be optimal. If you need further explanation, feel free to ask.
Edited to add swap function as in original post:
node* swap(node *head) {
if (head != NULL && head->next != NULL) {
node *newHead = head->next;
swapEveryTwo(NULL, head);
return newHead;
} else {
return head;
}
}

Your algorithm is about the best possible. Often we can get a bit of speed through simplicity. Instead of drawing pictures and reasoning about pointers, think of popping elements off the head of the input list and using a queue add-to-tail operation to build up the result. In pseudocode, we have
set_empty(rtn);
while (head) {
fst = pop(head);
if (head) {
snd = pop(head);
add_at_tail(rtn, snd);
}
add_at_tail(rtn, fst);
}
The if is needed only to protect against the case where the input list has odd length. If we're sure the list is even in length, we can skip it.
Now pop is very easy to implement. The add-to-tail operation is easiest if we use a dummy head node. So in C, we have:
node *swap(node *head)
{
node dummy[1]; // set_empty(rtn);
node *rtn = dummy;
while (head) {
node *fst = head; // fst = pop(head);
head = head->next;
if (head) {
node *snd = head; // snd = pop(head);
head = head->next;
rtn->next = snd; // add_to_tail(rtn, snd);
rtn = rtn->next;
}
rtn->next = fst; // add_to_tail(rtn, fst);
rtn = rtn->next;
}
rtn->next = NULL; // terminate tail
return dummy->next;
}
Now I have not tested this code, but I'm pretty sure it will run fine modulo maybe a typo or two. There are fewer tests than yours (just one per element). Tests are comparatively expensive because they can interfere with pipelining, so mine ought to run just a tad faster. Almost certainly this difference is irrelevant.
However, I think my code rather simpler to understand. Of course that's just one biased opinion, but readability does count during maintenance.
NB Now I have done a quick test and it worked on the first try! On the other hand when I tried your code I got a segv at
first->next=(third->next==NULL ? third : third->next);
Below is the test frame. Do you see anything wrong?
typedef struct node_s {
struct node_s *next;
int val;
} node;
// swap goes here
int main(void)
{
node dummy[1];
node *p = dummy;
for (int i = 0; i < 16; i++) {
p->next = malloc(sizeof(node));
p = p->next;
p->next = NULL;
p->val = 'a' + i;
}
p = swap(dummy->next);
while (p) {
printf("%c ", p->val);
p = p->next;
}
printf("\n");
return 0;
}

In JavaScript
LinkedList.prototype.swapPairs = function() {
var recurse = function(current) {
if (!current) return this;
if (current.next) {
var save = current.next.value;
current.next.value = current.value;
current.value = save;
current = current.next;
}
return recurse(current.next);
}.bind(this);
return recurse(this.head);
}

Alex DiCarlo's recursion method is simpler but needs to be corrected slightly.
void swapEveryTwo(node* prev, node* node) {
if (node != null && node->next != null) {
swapTwo(prev, node, node->next);
swapEveryTwo(node, node->next);
}
}
Please correct me if I am wrong.
Thanks!

Related

Reverse alternate elements and append to end of the list

Given a linked list as a->x->b->y->c->z , we need to reverse alternate element and append to end of list. That is , output it as a->b->c->z->y->x.
I have an O(n) solution but it takes extra memory , we take 2 lists and fill it with alternate elements respectively , so the two lists are a b c and x y z and then we will reverse the second list and append it to the tail of first so that it becomes a b c z y x .
My question is can we do it in place ? Or is there any other algorithm for the same ?
The basic idea:
Store x.
Make a point to b.
Make y point to the stored element (x).
Make b point to c.
etc.
At the end, make the last element at an odd position point to the stored element.
Pseudo-code: (simplified end-of-list check for readability)
current = startOfList
stored = NULL
while !endOfList
temp = current.next
current.next = current.next.next
temp.next = stored
stored = temp
current = current.next
current.next = stored
Complexity:
O(n) time, O(1) space.
Here is logic in recursion mode
public static Node alRev(Node head)
{
if (head == null) return head;
if (head.next != null)
{
if (head.next.next != null)
{
Node n = head.next;
head.next = head.next.next;
n.next = null;
Node temp = alRev(head.next);
if (temp != null){
temp.next = n;
return n;
}
}
else
return head.next;
}
else
return head;
return null;
}
This is a recent question from amazon interview, the Idea looks good and there seems to be no trick in it.
Java code with comments:
static void change(Node n)
{
if(n == null)
return;
Node current = n;
Node next = null, prev = null;
while(current != null && current.next != null)
{
// One of the alternate node which is to be reversed.
Node temp = current.next;
current.next = temp.next;
// Reverse the alternate node by changing its next pointer.
temp.next = next;
next = temp;
// This node will be used in the final step
// outside the loop to attach reversed nodes.
prev = current;
current = current.next;
}
// If there are odd number of nodes in the linked list.
if(current != null)
prev = current;
// Attach the reversed list to the unreversed list.
prev.next = next;
}
here the c code which don't uses any extra space for doing this..enjoy and have fun
in case of any doubt feel free to ask
#include<stdio.h>
#include<stdlib.h>
int n;
struct link
{
int val;
struct link *next;
};
void show(struct link *);
void addatbeg(struct link **p,int num)
{
struct link *temp,*help;
help=*p;
temp=(struct link *)malloc(sizeof(struct link));
temp->val=num;
temp->next=NULL;
if(help==NULL)
{
*p=temp;
}
else
{
temp->next=help;
*p=temp;
}
n++;
show(*p);
}
void revapp(struct link **p)
{
struct link *temp,*help,*q,*r;
r=NULL;
temp=*p;
help=*p;
while(temp->next!=NULL)
{
temp=temp->next;
q=r; //this portion will revrse the even position numbers
r=temp;
temp=temp->next;
//for making a connection between odd place numbers
if(help->next->next!=NULL)
{
help->next=temp;
help=help->next;
r->next=q;
}
else
{
r->next=q;
help->next=r;
show(*p);
return;
}
}
}
void show(struct link *q)
{
struct link *temp=q;
printf("\t");
while(q!=NULL )
{
printf("%d ->",q->val);
q=q->next;
if(q==temp)
{
printf("NULL\n");
return;
}
}
printf("NULL\n");
}
int main()
{
n=0;
struct link *p;
p=NULL;
// you can take user defined input but here i am solving it on predefined list
addatbeg(&p,8);
addatbeg(&p,7);
addatbeg(&p,6);
addatbeg(&p,5);
addatbeg(&p,4);
addatbeg(&p,3);
addatbeg(&p,2);
addatbeg(&p,1);
revapp(&p);
return 0;
}`

Implementing the Dutch National Flag Program with Linked Lists

I wanted to sort a linked list containing 0s, 1s or 2s. Now, this is clearly a variant of the Dutch National Flag Problem.
http://en.wikipedia.org/wiki/Dutch_national_flag_problem
The algorithm for the same as given in the link is:
"Have the top group grow down from the top of the array, the bottom group grow up from the bottom, and keep the middle group just above the bottom. The algorithm stores the locations just below the top group, just above the bottom, and just above the middle in three indexes. At each step, examine the element just above the middle. If it belongs to the top group, swap it with the element just below the top. If it belongs in the bottom, swap it with the element just above the bottom. If it is in the middle, leave it. Update the appropriate index. Complexity is Θ(n) moves and examinations."
And a C++ implementation given for the same is:
void threeWayPartition(int data[], int size, int low, int high) {
int p = -1;
int q = size;
for (int i = 0; i < q;) {
if (data[i] == low) {
swap(data[i], data[++p]);
++i;
} else if (data[i] >= high) {
swap(data[i], data[--q]);
} else {
++i;
}
}
}
My only question is how do we traverse back in a linked list like we are doing here in an array?
A standard singly-linked list doesn't allow you to move backwards given a linked list cell. However, you could use a doubly-linked list, where each cell stores a next and a previous pointer. That would let you navigate the list forwards and backwards.
However, for the particular problem you're trying to solve, I don't think this is necessary. One major difference between algorithms on arrays and on linked lists is that when working with linked lists, you can rearrange the cells in the list to reorder the elements in the list. Consequently, the algorithm you've detailed above - which works by changing the contents of the array - might not actually be the most elegant algorithm on linked lists.
If you are indeed working with linked lists, one possible way to solve this problem would be the following:
Create lists holding all values that are 0, 1, or 2.
Remove all cells from the linked list and distribute them into the list of elements that are equal to 0, 1, or 2.
Concatenate these three lists together.
This does no memory allocation and purely works by rearranging the linked list cells. It still runs in time Θ(n), which is another plus. Additionally, you can do this without ever having to walk backwards (i.e. this works on a singly-linked list).
I'll leave the complete implementation to you, but as an example, here's simple C++ code to distribute the linked list cells into the zero, one, and two lists:
struct Cell {
int value;
Cell* next;
}
/* Pointers to the heads of the three lists. */
Cell* lists[3] = { NULL, NULL, NULL };
/* Distribute the cells across the lists. */
while (list != NULL) {
/* Cache a pointer to the next cell in the list, since we will be
* rewiring this linked list.
*/
Cell* next = list->next;
/* Prepend this cell to the list it belongs to. */
list->next = lists[list->value];
lists[list->value] = list;
/* Advance to the next cell in the list. */
list = next;
}
Hope this helps!
As others have said, there is no way to "back up" in a linked list without reverse links. Though it's not exactly an answer to your question, the sort can be easily accomplished with three queues implementing a bucket sort with three buckets.
The advantage of queues (vice pushing on stacks) is that the sort is stable. That is, if there are data in the list nodes (other than the 0,1,2-valued keys), these will remain in the same order for each key.
This is only one of many cases where the canonical algorithm for arrays is not the best for lists.
There is a very slick, simple way to implement the queues: circularly linked lists where the first node, say p, is the tail of the queue and consequently p->next is is the head. With this, the code is concise.
#include <stdio.h>
#include <stdlib.h>
typedef struct node_s {
struct node_s *next;
int val;
int data;
} NODE;
// Add node to tail of queue q and return the new queue.
NODE *enqueue(NODE *q, NODE *node)
{
if (q) {
node->next = q->next;
q->next = node;
}
else node->next = node;
return node;
}
// Concatenate qa and qb and return the result.
NODE *cat(NODE *qa, NODE *qb)
{
NODE *head = qa->next;
qa->next = qb->next;
qb->next = head;
return qb;
}
// Sort a list where all values are 0, 1, or 2.
NODE *sort012(NODE *list)
{
NODE *next = NULL, *q[3] = { NULL, NULL, NULL};
for (NODE *p = list; p; p = next) {
next = p->next;
q[p->val] = enqueue(q[p->val], p);
}
NODE *result = cat(q[0], cat(q[1], q[2]));
// Now transform the circular queue to a simple linked list.
NODE *head = result->next;
result->next = NULL;
return head;
}
int main(void)
{
NODE *list = NULL;
int N = 100;
// Build a list of nodes for testing
for (int i = 0; i < N; ++i) {
NODE *p = malloc(sizeof(NODE));
p->val = rand() % 3;
p->data = N - i; // List ends up with data 1,2,3,..,N
p->next = list;
list = p;
}
list = sort012(list);
for (NODE *p = list; p; p = p->next)
printf("key val=%d, data=%d\n", p->val, p->data);
return 0;
}
This is now a complete simple test and it runs just fine.
This is untested. (I will try to test it if I get time.) But it ought to be at least very close to a solution.
Using a doubly linked list. If you have already implemented a linked list object and the related link list node object, and are able to traverse it in the forward direction it isn't a whole bunch more work to traverse in the reverse direction.
Assuming you have a Node object somewhat like:
public class Node
{
public Node Next;
public Object Value;
}
Then all you really need to do is change you Node class and you Insert method(s) up a little bit to keep track of of the Node that came previously:
public class Node
{
public Node Next;
public Node Previous;
public Object Value;
}
public void Insert(Node currentNode, Node insertedNode)
{
Node siblingNode = currentNode.Next;
insertedNode.Previous = currentNode;
insertedNode.Next = siblingNode;
if(siblingNode!= null)
siblingNode.previous = insertedNode;
currentNode.next = insertedNode;
}
PS Sorry, I didn't notice the edit that included the C++ stuff so it's more C#
Works for all cases by CHANGING NODES rather than NODE DATA.. Hoping its never too late!
METHOD(To throw some light on handling corner cases):
1. Keep three dummy nodes each for 0,1,2;
2. Iterate throught the list and add nodes to respective list.
3. Make the next of zero,one,two pointers as NULL.
4. Backup this last nodes of each list.
5. Now handle 8 different possible cases to join these list and Determine the HEAD.
zero one two
0 0 0
0 0 1
0 1 0
0 1 1
1 0 0
1 0 1
1 1 0
1 1 1
An implementation of this in C++.
Node* sortList(Node *head)
{
struct Node dummyzero,dummyone,dummytwo;
dummyzero.next = dummyone.next = dummytwo.next = NULL;
struct Node *zero =&dummyzero,*one = &dummyone,*two=&dummytwo;
Node *curr = head,*next=NULL;
while(curr)
{
next = curr->next;
if(curr->data==0)
{
zero->next = curr;
zero = zero->next;
}
else if(curr->data==1)
{
one->next = curr;
one = one->next;
}
else
{
two->next = curr;
two = two->next;
}
curr = next;
}
zero->next = one->next = two->next =NULL; //Since this dummynode, No segmentation fault here.
Node *zerolast = zero,*onelast = one,*twolast = two;
zero = dummyzero.next;
one = dummyone.next;
two = dummytwo.next;
if(zero==NULL)
{
if(one==NULL)
head = two;
else
{
head = one;
onelast->next = two;
}
}
else
{
head = zero;
if(one==NULL)
zerolast->next = two;
else
{
zerolast->next = one;
onelast->next = two;
}
}
return head;
}
The idea is to use dutch flag sorting algorithm, with a slight modification:
sort 0's and 1's as per dutch flag method,
But for 2's instead of adding them at the end of list, keep them in a separate linked list.
And finally append the 2's list to the sorted list of 0's and 1's.
Node * sort012_linked_list(Node * head) {
if (!head || !head->next)
return head;
Node * head_of_2s = NULL;
Node * prev = NULL;
Node * curr = head;
while (curr) {
if (curr->data == 0) {
if (prev == NULL || prev->data == 0) {
prev = curr;
curr = curr->next;
}
else {
prev->next = curr->next;
curr->next = head;
head = curr;
curr = prev->next;
}
}
else if (curr->data == 1) {
prev = curr;
curr = curr->next;
}
else { // curr->data == 2
if (prev == NULL) {
head = curr->next;
curr->next = head_of_2s;
head_of_2s = curr;
curr = head;
}
else {
prev->next = curr->next;
curr->next = head_of_2s;
head_of_2s = curr;
curr = prev->next;
}
}
}
if (prev)
prev->next = head_of_2s;
return head;
}

Reversing a singly linked list when a block size is given

There is a singly connected linked list and a block size is given.For eg if my linked list is 1->2->3->4->5->6->7->8-NULL and my block size is 4 then reverse the first 4 elements and then the second 4 elements.The output of the problem should be 4->3->2->1->8->7->6->5-NULL
I was thinking of dividing the linked list into segments of size 4 and then reversing it.
But that way I am forced to use a lot of extra nodes which is not desired at all.
The space complexity should be kept to a minimum.
It will be highly appreciable if someone can come with a better solution where the usage of extra nodes would be kept to a minimum.
I tried this...seems to work fine...
node* reverse(node* head) // function to reverse a list
{
node* new_head = NULL;
while(head != NULL)
{
node* next = head->next;
head->next = new_head;
new_head = head;
head = next;
}
return new_head;
}
node* reverse_by_block(node* head, int block)
{
if(head == NULL)
return head;
node* tmp = head;
node* new_head = head;
node* new_tail = NULL;
int count = block;
while(tmp != NULL && count--)
{
new_tail = tmp;
tmp = tmp->next;
}
new_tail->next = NULL;
new_tail = new_head;
new_head = reverse(new_head);
new_tail->next = reverse_by_block(tmp,block);
return new_head;
}
You can advance swapping the current element with the next 3 times: 1234, 2134, 2314, 2341. Then do it twice to get 3421. Then once to get 4321. Then advance 4 steps and repeat the process with the next block.
This can be done in linear-time, with constant space.
Here is a brief description:
Split the linked list into two parts by block-size
int split(node* head, node** listA, node** listB size_t block_size)
{
node* cur = head;
while(block_size && cur)
{
cur = cur->next;
--block_size;
}
if(!cur) { /* error : invalid block size */ return -1; }
*listA = head;
*listB = cur->next;
cur->next = NULL; /* terminate list A */
return 0;
}
Reverse the two sub-parts, (use a non-recursive linear time, constant space function)
reverse(listA);
reverse(listB);
Link them to get the desired linked list.
cur = *listA;
/* goto last but one element of listA */
while(cur->next) cur = cur->next;
cur->next = *listB;

searching a binary tree

I'm writing an iterative function to search a binary tree for a certain value. This is localized to signed ints until I get into how to genericize classes.
Assume that my class is BinarySearchTree, and it has a pointer to the root node of the tree. Also assume that nodes are inserted through an insert function, and have pointers to two children. Here is a much abbreviated version of the Node struct:
struct Node
{
public:
Node *left_, *right_;
int value_
Node(int val) : value_(val), left_(0), right_(0) { }
//done in this manner to always make sure blank children are
//init to zero, or null
Node(int val, Node *left, Node *right) : value_(val), left_(0), right_(0)
{ left_ = left; right_ = right; }
}
So, you can safely assume that a node's uninit pointers will be NULL.
Here is my code:
int BinarySearchTree::search(int val)
{
Node* next = this->root();
while (next->left() != 0 || next->right () != 0)
{
if (val == next->value())
{
return next->value();
}
else if (val < next->value())
{
next = next->left();
}
else if (val > next->value())
{
next = next->right();
}
}
//not found
return 0;
}
This code is being rejected by a friend for two reasons:
1) If next has no children, both will evaluate to zero and I will prematurely exit the loop (I will never check the searched val against next's value).
2) If next has one child, but the data you are searching for should be on the empty side of the tree, next will be set to 0, and it will loop again, comparing next (which is 0) to the left and right trees like while(0->left()), resulting in undefined behavior.
I am told that the solution to both problems lies in the loop condition, but I can't see what I can do to easily remedy the situation. Can the community of Stack Overflow offer any insights?
I think you should be testing if next is not NULL in your loop like so:
int BinarySearchTree::search(int val)
{
Node* next = this->root();
while (next)
{
if (val == next->value())
{
return next->value();
}
else if (val < next->value())
{
next = next->left();
}
else if (val > next->value())
{
next = next->right();
}
}
//not found
return 0;
}
Try this:
while (next != NULL) ?
First of all, I'm not sure why you are returning an int. What if you are searching for 0 in the tree. You probably want something like this:
bool BinarySearchTree::Search(int val) {
Node* current = root();
while (current != NULL) {
// Check if it's here
if (val == current->value()) {
return true;
}
if (val < current->value()) {
current = current->left();
} else {
current = current->right();
}
}
// Not found
return false;
}
Notice that the loop invariant: at the beginning of each loop, you are at a non null node that you need to "process". First check if it's the node you want. If not, make a branch, and let the loop decide if the branch was "good" (ie - non null). Then you'll let the next loop iteration take care of testing.

Merge Sort a Linked List

I was recently brushing up on some fundamentals and found merge sorting a linked list to be a pretty good challenge. If you have a good implementation then show it off here.
Wonder why it should be big challenge as it is stated here, here is a straightforward implementation in Java with out any "clever tricks".
//The main function
public static Node merge_sort(Node head)
{
if(head == null || head.next == null)
return head;
Node middle = getMiddle(head); //get the middle of the list
Node left_head = head;
Node right_head = middle.next;
middle.next = null; //split the list into two halfs
return merge(merge_sort(left_head), merge_sort(right_head)); //recurse on that
}
//Merge subroutine to merge two sorted lists
public static Node merge(Node a, Node b)
{
Node dummyHead = new Node();
for(Node current = dummyHead; a != null && b != null; current = current.next;)
{
if(a.data <= b.data)
{
current.next = a;
a = a.next;
}
else
{
current.next = b;
b = b.next;
}
}
dummyHead.next = (a == null) ? b : a;
return dummyHead.next;
}
//Finding the middle element of the list for splitting
public static Node getMiddle(Node head)
{
if(head == null)
return head;
Node slow = head, fast = head;
while(fast.next != null && fast.next.next != null)
{
slow = slow.next;
fast = fast.next.next;
}
return slow;
}
A simpler/clearer implementation might be the recursive implementation, from which the NLog(N) execution time is more clear.
typedef struct _aList {
struct _aList* next;
struct _aList* prev; // Optional.
// some data
} aList;
aList* merge_sort_list_recursive(aList *list,int (*compare)(aList *one,aList *two))
{
// Trivial case.
if (!list || !list->next)
return list;
aList *right = list,
*temp = list,
*last = list,
*result = 0,
*next = 0,
*tail = 0;
// Find halfway through the list (by running two pointers, one at twice the speed of the other).
while (temp && temp->next)
{
last = right;
right = right->next;
temp = temp->next->next;
}
// Break the list in two. (prev pointers are broken here, but we fix later)
last->next = 0;
// Recurse on the two smaller lists:
list = merge_sort_list_recursive(list, compare);
right = merge_sort_list_recursive(right, compare);
// Merge:
while (list || right)
{
// Take from empty lists, or compare:
if (!right) {
next = list;
list = list->next;
} else if (!list) {
next = right;
right = right->next;
} else if (compare(list, right) < 0) {
next = list;
list = list->next;
} else {
next = right;
right = right->next;
}
if (!result) {
result=next;
} else {
tail->next=next;
}
next->prev = tail; // Optional.
tail = next;
}
return result;
}
NB: This has a Log(N) storage requirement for the recursion. Performance should be roughly comparable with the other strategy I posted. There is a potential optimisation here by running the merge loop while (list && right), and simple appending the remaining list (since we don't really care about the end of the lists; knowing that they're merged suffices).
Heavily based on the EXCELLENT code from: http://www.chiark.greenend.org.uk/~sgtatham/algorithms/listsort.html
Trimmed slightly, and tidied:
typedef struct _aList {
struct _aList* next;
struct _aList* prev; // Optional.
// some data
} aList;
aList *merge_sort_list(aList *list,int (*compare)(aList *one,aList *two))
{
int listSize=1,numMerges,leftSize,rightSize;
aList *tail,*left,*right,*next;
if (!list || !list->next) return list; // Trivial case
do { // For each power of two<=list length
numMerges=0,left=list;tail=list=0; // Start at the start
while (left) { // Do this list_len/listSize times:
numMerges++,right=left,leftSize=0,rightSize=listSize;
// Cut list into two halves (but don't overrun)
while (right && leftSize<listSize) leftSize++,right=right->next;
// Run through the lists appending onto what we have so far.
while (leftSize>0 || (rightSize>0 && right)) {
// Left empty, take right OR Right empty, take left, OR compare.
if (!leftSize) {next=right;right=right->next;rightSize--;}
else if (!rightSize || !right) {next=left;left=left->next;leftSize--;}
else if (compare(left,right)<0) {next=left;left=left->next;leftSize--;}
else {next=right;right=right->next;rightSize--;}
// Update pointers to keep track of where we are:
if (tail) tail->next=next; else list=next;
// Sort prev pointer
next->prev=tail; // Optional.
tail=next;
}
// Right is now AFTER the list we just sorted, so start the next sort there.
left=right;
}
// Terminate the list, double the list-sort size.
tail->next=0,listSize<<=1;
} while (numMerges>1); // If we only did one merge, then we just sorted the whole list.
return list;
}
NB: This is O(NLog(N)) guaranteed, and uses O(1) resources (no recursion, no stack, nothing).
One interesting way is to maintain a stack, and only merge if the list on the stack has the same number of elements, and otherwise push the list, until you run out of elements in the incoming list, and then merge up the stack.
The simplest is from
Gonnet + Baeza Yates Handbook of Algorithms. You call it with the number of sorted elements you want, which recursively gets bisected until it reaches a request for a size one list which you then just peel off the front of the original list. These all get merged up into a full sized sorted list.
[Note that the cool stack-based one in the first post is called the Online Mergesort and it gets the tiniest mention in an exercise in Knuth Vol 3]
Here's an alternative recursive version. This does not need to step along the list to split it: we supply a pointer to a head element (which is not part of the sort) and a length, and the recursive function returns a pointer to the end of the sorted list.
element* mergesort(element *head,long lengtho)
{
long count1=(lengtho/2), count2=(lengtho-count1);
element *next1,*next2,*tail1,*tail2,*tail;
if (lengtho<=1) return head->next; /* Trivial case. */
tail1 = mergesort(head,count1);
tail2 = mergesort(tail1,count2);
tail=head;
next1 = head->next;
next2 = tail1->next;
tail1->next = tail2->next; /* in case this ends up as the tail */
while (1) {
if(cmp(next1,next2)<=0) {
tail->next=next1; tail=next1;
if(--count1==0) { tail->next=next2; return tail2; }
next1=next1->next;
} else {
tail->next=next2; tail=next2;
if(--count2==0) { tail->next=next1; return tail1; }
next2=next2->next;
}
}
}
I'd been obsessing over optimizing clutter for this algorithm and below is what I've finally arrived at. Lot of other code on Internet and StackOverflow is horribly bad. There are people trying to get middle point of the list, doing recursion, having multiple loops for left over nodes, maintaining counts of ton of things - ALL of which is unnecessary. MergeSort naturally fits to linked list and algorithm can be beautiful and compact but it's not trivial to get to that state.
Below code maintains minimum number of variables and has minimum number of logical steps needed for the algorithm (i.e. without making code unmaintainable/unreadable) as far as I know. However I haven't tried to minimize LOC and kept as much white space as necessary to keep things readable. I've tested this code through fairly rigorous unit tests.
Note that this answer combines few techniques from other answer https://stackoverflow.com/a/3032462/207661. While the code is in C#, it should be trivial to convert in to C++, Java, etc.
SingleListNode<T> SortLinkedList<T>(SingleListNode<T> head) where T : IComparable<T>
{
int blockSize = 1, blockCount;
do
{
//Maintain two lists pointing to two blocks, left and right
SingleListNode<T> left = head, right = head, tail = null;
head = null; //Start a new list
blockCount = 0;
//Walk through entire list in blocks of size blockCount
while (left != null)
{
blockCount++;
//Advance right to start of next block, measure size of left list while doing so
int leftSize = 0, rightSize = blockSize;
for (;leftSize < blockSize && right != null; ++leftSize)
right = right.Next;
//Merge two list until their individual ends
bool leftEmpty = leftSize == 0, rightEmpty = rightSize == 0 || right == null;
while (!leftEmpty || !rightEmpty)
{
SingleListNode<T> smaller;
//Using <= instead of < gives us sort stability
if (rightEmpty || (!leftEmpty && left.Value.CompareTo(right.Value) <= 0))
{
smaller = left; left = left.Next; --leftSize;
leftEmpty = leftSize == 0;
}
else
{
smaller = right; right = right.Next; --rightSize;
rightEmpty = rightSize == 0 || right == null;
}
//Update new list
if (tail != null)
tail.Next = smaller;
else
head = smaller;
tail = smaller;
}
//right now points to next block for left
left = right;
}
//terminate new list, take care of case when input list is null
if (tail != null)
tail.Next = null;
//Lg n iterations
blockSize <<= 1;
} while (blockCount > 1);
return head;
}
Points of interest
There is no special handling for cases like null list of list of 1 etc required. These cases "just works".
Lot of "standard" algorithms texts have two loops to go over leftover elements to handle the case when one list is shorter than other. Above code eliminates need for it.
We make sure sort is stable
The inner while loop which is a hot spot evaluates 3 expressions per iteration on average which I think is minimum one can do.
Update: #ideasman42 has translated above code to C/C++ along with suggestions for fixing comments and bit more improvement. Above code is now up to date with these.
I decided to test the examples here, and also one more approach, originally written by Jonathan Cunningham in Pop-11. I coded all the approaches in C# and did a comparison with a range of different list sizes. I compared the Mono eglib approach by Raja R Harinath, the C# code by Shital Shah, the Java approach by Jayadev, the recursive and non-recursive versions by David Gamble, the first C code by Ed Wynn (this crashed with my sample dataset, I didn't debug), and Cunningham's version. Full code here: https://gist.github.com/314e572808f29adb0e41.git.
Mono eglib is based on a similar idea to Cunningham's and is of comparable speed, unless the list happens to be sorted already, in which case Cunningham's approach is much much faster (if its partially sorted, the eglib is slightly faster). The eglib code uses a fixed table to hold the merge sort recursion, whereas Cunningham's approach works by using increasing levels of recursion - so it starts out using no recursion, then 1-deep recursion, then 2-deep recursion and so on, according to how many steps are needed to do the sort. I find the Cunningham code a little easier to follow and there is no guessing involved in how big to make the recursion table, so it gets my vote. The other approaches I tried from this page were two or more times slower.
Here is the C# port of the Pop-11 sort:
/// <summary>
/// Sort a linked list in place. Returns the sorted list.
/// Originally by Jonathan Cunningham in Pop-11, May 1981.
/// Ported to C# by Jon Meyer.
/// </summary>
public class ListSorter<T> where T : IComparable<T> {
SingleListNode<T> workNode = new SingleListNode<T>(default(T));
SingleListNode<T> list;
/// <summary>
/// Sorts a linked list. Returns the sorted list.
/// </summary>
public SingleListNode<T> Sort(SingleListNode<T> head) {
if (head == null) throw new NullReferenceException("head");
list = head;
var run = GetRun(); // get first run
// As we progress, we increase the recursion depth.
var n = 1;
while (list != null) {
var run2 = GetSequence(n);
run = Merge(run, run2);
n++;
}
return run;
}
// Get the longest run of ordered elements from list.
// The run is returned, and list is updated to point to the
// first out-of-order element.
SingleListNode<T> GetRun() {
var run = list; // the return result is the original list
var prevNode = list;
var prevItem = list.Value;
list = list.Next; // advance to the next item
while (list != null) {
var comp = prevItem.CompareTo(list.Value);
if (comp > 0) {
// reached end of sequence
prevNode.Next = null;
break;
}
prevItem = list.Value;
prevNode = list;
list = list.Next;
}
return run;
}
// Generates a sequence of Merge and GetRun() operations.
// If n is 1, returns GetRun()
// If n is 2, returns Merge(GetRun(), GetRun())
// If n is 3, returns Merge(Merge(GetRun(), GetRun()),
// Merge(GetRun(), GetRun()))
// and so on.
SingleListNode<T> GetSequence(int n) {
if (n < 2) {
return GetRun();
} else {
n--;
var run1 = GetSequence(n);
if (list == null) return run1;
var run2 = GetSequence(n);
return Merge(run1, run2);
}
}
// Given two ordered lists this returns a list that is the
// result of merging the two lists in-place (modifying the pairs
// in list1 and list2).
SingleListNode<T> Merge(SingleListNode<T> list1, SingleListNode<T> list2) {
// we reuse a single work node to hold the result.
// Simplifies the number of test cases in the code below.
var prevNode = workNode;
while (true) {
if (list1.Value.CompareTo(list2.Value) <= 0) {
// list1 goes first
prevNode.Next = list1;
prevNode = list1;
if ((list1 = list1.Next) == null) {
// reached end of list1 - join list2 to prevNode
prevNode.Next = list2;
break;
}
} else { // same but for list2
prevNode.Next = list2;
prevNode = list2;
if ((list2 = list2.Next) == null) {
prevNode.Next = list1;
break;
}
}
}
// the result is in the back of the workNode
return workNode.Next;
}
}
Here is my implementation of Knuth's "List merge sort" (Algorithm 5.2.4L from Vol.3 of TAOCP, 2nd ed.). I'll add some comments at the end, but here's a summary:
On random input, it runs a bit faster than Simon Tatham's code (see Dave Gamble's non-recursive answer, with a link) but a bit slower than Dave Gamble's recursive code. It's harder to understand than either. At least in my implementation, it requires each element to have TWO pointers to elements. (An alternative would be one pointer and a boolean flag.) So, it's probably not a useful approach. However, one distinctive point is that it runs very fast if the input has long stretches that are already sorted.
element *knuthsort(element *list)
{ /* This is my attempt at implementing Knuth's Algorithm 5.2.4L "List merge sort"
from Vol.3 of TAOCP, 2nd ed. */
element *p, *pnext, *q, *qnext, head1, head2, *s, *t;
if(!list) return NULL;
L1: /* This is the clever L1 from exercise 12, p.167, solution p.647. */
head1.next=list;
t=&head2;
for(p=list, pnext=p->next; pnext; p=pnext, pnext=p->next) {
if( cmp(p,pnext) > 0 ) { t->next=NULL; t->spare=pnext; t=p; }
}
t->next=NULL; t->spare=NULL; p->spare=NULL;
head2.next=head2.spare;
L2: /* begin a new pass: */
t=&head2;
q=t->next;
if(!q) return head1.next;
s=&head1;
p=s->next;
L3: /* compare: */
if( cmp(p,q) > 0 ) goto L6;
L4: /* add p onto the current end, s: */
if(s->next) s->next=p; else s->spare=p;
s=p;
if(p->next) { p=p->next; goto L3; }
else p=p->spare;
L5: /* complete the sublist by adding q and all its successors: */
s->next=q; s=t;
for(qnext=q->next; qnext; q=qnext, qnext=q->next);
t=q; q=q->spare;
goto L8;
L6: /* add q onto the current end, s: */
if(s->next) s->next=q; else s->spare=q;
s=q;
if(q->next) { q=q->next; goto L3; }
else q=q->spare;
L7: /* complete the sublist by adding p and all its successors: */
s->next=p;
s=t;
for(pnext=p->next; pnext; p=pnext, pnext=p->next);
t=p; p=p->spare;
L8: /* is this end of the pass? */
if(q) goto L3;
if(s->next) s->next=p; else s->spare=p;
t->next=NULL; t->spare=NULL;
goto L2;
}
There's a non-recursive linked-list mergesort in mono eglib.
The basic idea is that the control-loop for the various merges parallels the bitwise-increment of a binary integer. There are O(n) merges to "insert" n nodes into the merge tree, and the rank of those merges corresponds to the binary digit that gets incremented. Using this analogy, only O(log n) nodes of the merge-tree need to be materialized into a temporary holding array.
A tested, working C++ version of single linked list, based on the highest voted answer.
singlelinkedlist.h:
#pragma once
#include <stdexcept>
#include <iostream>
#include <initializer_list>
namespace ythlearn{
template<typename T>
class Linkedlist{
public:
class Node{
public:
Node* next;
T elem;
};
Node head;
int _size;
public:
Linkedlist(){
head.next = nullptr;
_size = 0;
}
Linkedlist(std::initializer_list<T> init_list){
head.next = nullptr;
_size = 0;
for(auto s = init_list.begin(); s!=init_list.end(); s++){
push_left(*s);
}
}
int size(){
return _size;
}
bool isEmpty(){
return size() == 0;
}
bool isSorted(){
Node* n_ptr = head.next;
while(n_ptr->next != nullptr){
if(n_ptr->elem > n_ptr->next->elem)
return false;
n_ptr = n_ptr->next;
}
return true;
}
Linkedlist& push_left(T elem){
Node* n = new Node;
n->elem = elem;
n->next = head.next;
head.next = n;
++_size;
return *this;
}
void print(){
Node* loopPtr = head.next;
while(loopPtr != nullptr){
std::cout << loopPtr->elem << " ";
loopPtr = loopPtr->next;
}
std::cout << std::endl;
}
void call_merge(){
head.next = merge_sort(head.next);
}
Node* merge_sort(Node* n){
if(n == nullptr || n->next == nullptr)
return n;
Node* middle = getMiddle(n);
Node* left_head = n;
Node* right_head = middle->next;
middle->next = nullptr;
return merge(merge_sort(left_head), merge_sort(right_head));
}
Node* getMiddle(Node* n){
if(n == nullptr)
return n;
Node* slow, *fast;
slow = fast = n;
while(fast->next != nullptr && fast->next->next != nullptr){
slow = slow->next;
fast = fast->next->next;
}
return slow;
}
Node* merge(Node* a, Node* b){
Node dummyHead;
Node* current = &dummyHead;
while(a != nullptr && b != nullptr){
if(a->elem < b->elem){
current->next = a;
a = a->next;
}else{
current->next = b;
b = b->next;
}
current = current->next;
}
current->next = (a == nullptr) ? b : a;
return dummyHead.next;
}
Linkedlist(const Linkedlist&) = delete;
Linkedlist& operator=(const Linkedlist&) const = delete;
~Linkedlist(){
Node* node_to_delete;
Node* ptr = head.next;
while(ptr != nullptr){
node_to_delete = ptr;
ptr = ptr->next;
delete node_to_delete;
}
}
};
}
main.cpp:
#include <iostream>
#include <cassert>
#include "singlelinkedlist.h"
using namespace std;
using namespace ythlearn;
int main(){
Linkedlist<int> l = {3,6,-5,222,495,-129,0};
l.print();
l.call_merge();
l.print();
assert(l.isSorted());
return 0;
}
Simplest Java Implementation:
Time Complexity: O(nLogn) n = number of nodes. Each iteration through the linked list doubles the size of the sorted smaller linked lists. For example, after the first iteration, the linked list will be sorted into two halves. After the second iteration, the linked list will be sorted into four halves. It will keep sorting up to the size of the linked list. This will take O(logn) doublings of the small linked lists' sizes to reach the original linked list size. The n in nlogn is there because each iteration of the linked list will take time proportional to the number of nodes in the originial linked list.
class Node {
int data;
Node next;
Node(int d) {
data = d;
}
}
class LinkedList {
Node head;
public Node mergesort(Node head) {
if(head == null || head.next == null) return head;
Node middle = middle(head), middle_next = middle.next;
middle.next = null;
Node left = mergesort(head), right = mergesort(middle_next), node = merge(left, right);
return node;
}
public Node merge(Node first, Node second) {
Node node = null;
if (first == null) return second;
else if (second == null) return first;
else if (first.data <= second.data) {
node = first;
node.next = merge(first.next, second);
} else {
node = second;
node.next = merge(first, second.next);
}
return node;
}
public Node middle(Node head) {
if (head == null) return head;
Node second = head, first = head.next;
while(first != null) {
first = first.next;
if (first != null) {
second = second.next;
first = first.next;
}
}
return second;
}
}
Another example of a non-recursive merge sort for linked lists, where the functions are not part of a class. This example code and HP / Microsoft std::list::sort both use the same basic algorithm. A bottom up, non-recursive, merge sort that uses a small (26 to 32) array of pointers to the first nodes of a list, where array[i] is either 0 or points to a list of size 2 to the power i. On my system, Intel 2600K 3.4ghz, it can sort 4 million nodes with 32 bit unsigned integers as data in about 1 second.
typedef struct NODE_{
struct NODE_ * next;
uint32_t data;
}NODE;
NODE * MergeLists(NODE *, NODE *); /* prototype */
/* sort a list using array of pointers to list */
/* aList[i] == NULL or ptr to list with 2^i nodes */
#define NUMLISTS 32 /* number of lists */
NODE * SortList(NODE *pList)
{
NODE * aList[NUMLISTS]; /* array of lists */
NODE * pNode;
NODE * pNext;
int i;
if(pList == NULL) /* check for empty list */
return NULL;
for(i = 0; i < NUMLISTS; i++) /* init array */
aList[i] = NULL;
pNode = pList; /* merge nodes into array */
while(pNode != NULL){
pNext = pNode->next;
pNode->next = NULL;
for(i = 0; (i < NUMLISTS) && (aList[i] != NULL); i++){
pNode = MergeLists(aList[i], pNode);
aList[i] = NULL;
}
if(i == NUMLISTS) /* don't go beyond end of array */
i--;
aList[i] = pNode;
pNode = pNext;
}
pNode = NULL; /* merge array into one list */
for(i = 0; i < NUMLISTS; i++)
pNode = MergeLists(aList[i], pNode);
return pNode;
}
/* merge two already sorted lists */
/* compare uses pSrc2 < pSrc1 to follow the STL rule */
/* of only using < and not <= */
NODE * MergeLists(NODE *pSrc1, NODE *pSrc2)
{
NODE *pDst = NULL; /* destination head ptr */
NODE **ppDst = &pDst; /* ptr to head or prev->next */
if(pSrc1 == NULL)
return pSrc2;
if(pSrc2 == NULL)
return pSrc1;
while(1){
if(pSrc2->data < pSrc1->data){ /* if src2 < src1 */
*ppDst = pSrc2;
pSrc2 = *(ppDst = &(pSrc2->next));
if(pSrc2 == NULL){
*ppDst = pSrc1;
break;
}
} else { /* src1 <= src2 */
*ppDst = pSrc1;
pSrc1 = *(ppDst = &(pSrc1->next));
if(pSrc1 == NULL){
*ppDst = pSrc2;
break;
}
}
}
return pDst;
}
Visual Studio 2015 changed std::list::sort to be based on iterators instead of lists, and also changed to a top down merge sort, which requires the overhead of scanning. I initially assumed that the switch to top down was needed to work with the iterators, but when it was asked about again, I investigated this and determined that the switch to the slower top down method was not needed, and bottom up could be implemented using the same iterator based logic. The answer in this link explains this and provide a stand-alone example as well as a replacement for VS2019's std::list::sort() in the include file "list".
`std::list<>::sort()` - why the sudden switch to top-down strategy?
This is the entire Piece of code which shows how we can create linklist in java and sort it using Merge sort. I am creating node in MergeNode class and there is another class MergesortLinklist where there is divide and merge logic.
class MergeNode {
Object value;
MergeNode next;
MergeNode(Object val) {
value = val;
next = null;
}
MergeNode() {
value = null;
next = null;
}
public Object getValue() {
return value;
}
public void setValue(Object value) {
this.value = value;
}
public MergeNode getNext() {
return next;
}
public void setNext(MergeNode next) {
this.next = next;
}
#Override
public String toString() {
return "MergeNode [value=" + value + ", next=" + next + "]";
}
}
public class MergesortLinkList {
MergeNode head;
static int totalnode;
public MergeNode getHead() {
return head;
}
public void setHead(MergeNode head) {
this.head = head;
}
MergeNode add(int i) {
// TODO Auto-generated method stub
if (head == null) {
head = new MergeNode(i);
// System.out.println("head value is "+head);
return head;
}
MergeNode temp = head;
while (temp.next != null) {
temp = temp.next;
}
temp.next = new MergeNode(i);
return head;
}
MergeNode mergesort(MergeNode nl1) {
// TODO Auto-generated method stub
if (nl1.next == null) {
return nl1;
}
int counter = 0;
MergeNode temp = nl1;
while (temp != null) {
counter++;
temp = temp.next;
}
System.out.println("total nodes " + counter);
int middle = (counter - 1) / 2;
temp = nl1;
MergeNode left = nl1, right = nl1;
int leftindex = 0, rightindex = 0;
if (middle == leftindex) {
right = left.next;
}
while (leftindex < middle) {
leftindex++;
left = left.next;
right = left.next;
}
left.next = null;
left = nl1;
System.out.println(left.toString());
System.out.println(right.toString());
MergeNode p1 = mergesort(left);
MergeNode p2 = mergesort(right);
MergeNode node = merge(p1, p2);
return node;
}
MergeNode merge(MergeNode p1, MergeNode p2) {
// TODO Auto-generated method stub
MergeNode L = p1;
MergeNode R = p2;
int Lcount = 0, Rcount = 0;
MergeNode tempnode = null;
while (L != null && R != null) {
int val1 = (int) L.value;
int val2 = (int) R.value;
if (val1 > val2) {
if (tempnode == null) {
tempnode = new MergeNode(val2);
R = R.next;
} else {
MergeNode store = tempnode;
while (store.next != null) {
store = store.next;
}
store.next = new MergeNode(val2);
R = R.next;
}
} else {
if (tempnode == null) {
tempnode = new MergeNode(val1);
L = L.next;
} else {
MergeNode store = tempnode;
while (store.next != null) {
store = store.next;
}
store.next = new MergeNode(val1);
L = L.next;
}
}
}
MergeNode handle = tempnode;
while (L != null) {
while (handle.next != null) {
handle = handle.next;
}
handle.next = L;
L = null;
}
// Copy remaining elements of L[] if any
while (R != null) {
while (handle.next != null) {
handle = handle.next;
}
handle.next = R;
R = null;
}
System.out.println("----------------sorted value-----------");
System.out.println(tempnode.toString());
return tempnode;
}
public static void main(String[] args) {
MergesortLinkList objsort = new MergesortLinkList();
MergeNode n1 = objsort.add(9);
MergeNode n2 = objsort.add(7);
MergeNode n3 = objsort.add(6);
MergeNode n4 = objsort.add(87);
MergeNode n5 = objsort.add(16);
MergeNode n6 = objsort.add(81);
MergeNode n7 = objsort.add(21);
MergeNode n8 = objsort.add(16);
MergeNode n9 = objsort.add(99);
MergeNode n10 = objsort.add(31);
MergeNode val = objsort.mergesort(n1);
System.out.println("===============sorted values=====================");
while (val != null) {
System.out.println(" value is " + val.value);
val = val.next;
}
}
}
I don't see any C++ solutions posted here. So, here it goes. Hope it helps someone.
class Solution {
public:
ListNode *merge(ListNode *left, ListNode *right){
ListNode *head = NULL, *temp = NULL;
// Find which one is the head node for the merged list
if(left->val <= right->val){
head = left, temp = left;
left = left->next;
}
else{
head = right, temp = right;
right = right->next;
}
while(left && right){
if(left->val <= right->val){
temp->next = left;
temp = left;
left = left->next;
}
else{
temp->next = right;
temp = right;
right = right->next;
}
}
// If some elements still left in the left or the right list
if(left)
temp->next = left;
if(right)
temp->next = right;
return head;
}
ListNode* sortList(ListNode* head){
if(!head || !head->next)
return head;
// Find the length of the list
int length = 0;
ListNode *temp = head;
while(temp){
length++;
temp = temp->next;
}
// Reset temp
temp = head;
// Store half of it in left and the other half in right
// Create two lists and sort them
ListNode *left = temp, *prev = NULL;
int i = 0, mid = length / 2;
// Left list
while(i < mid){
prev = temp;
temp = temp->next;
i++;
}
// The end of the left list should point to NULL
if(prev)
prev->next = NULL;
// Right list
ListNode *right = temp;
// Sort left list
ListNode *sortedLeft = sortList(left);
// Sort right list
ListNode *sortedRight = sortList(right);
// Merge them
ListNode *sortedList = merge(sortedLeft, sortedRight);
return sortedList;
}
};
Here is the Java Implementation of Merge Sort on Linked List:
Time Complexity: O(n.logn)
Space Complexity: O(1) - Merge sort implementation on Linked List avoids the O(n) auxiliary storage cost normally associated with the
algorithm
class Solution
{
public ListNode mergeSortList(ListNode head)
{
if(head == null || head.next == null)
return head;
ListNode mid = getMid(head), second_head = mid.next; mid.next = null;
return merge(mergeSortList(head), mergeSortList(second_head));
}
private ListNode merge(ListNode head1, ListNode head2)
{
ListNode result = new ListNode(0), current = result;
while(head1 != null && head2 != null)
{
if(head1.val < head2.val)
{
current.next = head1;
head1 = head1.next;
}
else
{
current.next = head2;
head2 = head2.next;
}
current = current.next;
}
if(head1 != null) current.next = head1;
if(head2 != null) current.next = head2;
return result.next;
}
private ListNode getMid(ListNode head)
{
ListNode slow = head, fast = head.next;
while(fast != null && fast.next != null)
{
slow = slow.next;
fast = fast.next.next;
}
return slow;
}
}
Hey I know that this is a bit late an answer but got a fast simple one.
The code is in F# but will goes in any language. Since this is an uncommen language of the ML family, I'll give some point to enhance the readability.
F# are nesting done by tabulating. the last line of code in a function (nested part) are the return value. (x, y) is a tuple, x::xs is a list of the head x and tail xs (where xs can be empty), |> take the result of last line an pipe it as argument to the expression right of it (readability enhancing) and last (fun args -> some expression) are a lambda function.
// split the list into a singleton list
let split list = List.map (fun x -> [x]) lst
// takes to list and merge them into a sorted list
let sort lst1 lst2 =
// nested function to hide accumulator
let rec s acc pair =
match pair with
// empty list case, return the sorted list
| [], [] -> List.rev acc
| xs, [] | [], xs ->
// one empty list case,
// append the rest of xs onto acc and return the sorted list
List.fold (fun ys y -> y :: ys) acc xs
|> List.rev
// general case
| x::xs, y::ys ->
match x < y with
| true -> // cons x onto the accumulator
s (x::acc) (xs,y::ys)
| _ ->
// cons y onto the accumulator
s (y::acc) (x::xs,ys)
s [] (lst1, lst2)
let msort lst =
let rec merge acc lst =
match lst with
| [] ->
match acc with
| [] -> [] // empty list case
| _ -> merge [] acc
| x :: [] -> // single list case (x is a list)
match acc with
| [] -> x // since acc are empty there are only x left, hence x are the sorted list.
| _ -> merge [] (x::acc) // still need merging.
| x1 :: x2 :: xs ->
// merge the lists x1 and x2 and add them to the acummulator. recursiv call
merge (sort x1 x2 :: acc) xs
// return part
split list // expand to singleton list list
|> merge [] // merge and sort recursively.
It is important to notice that this is fully tail recursive so no possibility of stack overflow, and by first expanding the list to a singleton list list in one go we, lower the constant factor on the worst cost. Since merge are working on list of list, we can recursively merge and sort the inner list until we get to the fix point where all inner list are sorted into one list and then we return that list, hence collapsing from a list list to a list again.
Here is the solution using Swift Programming Language.
//Main MergeSort Function
func mergeSort(head: Node?) -> Node? {
guard let head = head else { return nil }
guard let _ = head.next else { return head }
let middle = getMiddle(head: head)
let left = head
let right = middle.next
middle.next = nil
return merge(left: mergeSort(head: left), right: mergeSort(head: right))
}
//Merge Function
func merge(left: Node?, right: Node?) -> Node? {
guard let left = left, let right = right else { return nil}
let dummyHead: Node = Node(value: 0)
var current: Node? = dummyHead
var currentLeft: Node? = left
var currentRight: Node? = right
while currentLeft != nil && currentRight != nil {
if currentLeft!.value < currentRight!.value {
current?.next = currentLeft
currentLeft = currentLeft!.next
} else {
current?.next = currentRight
currentRight = currentRight!.next
}
current = current?.next
}
if currentLeft != nil {
current?.next = currentLeft
}
if currentRight != nil {
current?.next = currentRight
}
return dummyHead.next!
}
And here are the Node Class & getMiddle Method
class Node {
//Node Class which takes Integers as value
var value: Int
var next: Node?
init(value: Int) {
self.value = value
}
}
func getMiddle(head: Node) -> Node {
guard let nextNode = head.next else { return head }
var slow: Node = head
var fast: Node? = head
while fast?.next?.next != nil {
slow = slow.next!
fast = fast!.next?.next
}
return slow
}
public int[] msort(int[] a) {
if (a.Length > 1) {
int min = a.Length / 2;
int max = min;
int[] b = new int[min];
int[] c = new int[max]; // dividing main array into two half arrays
for (int i = 0; i < min; i++) {
b[i] = a[i];
}
for (int i = min; i < min + max; i++) {
c[i - min] = a[i];
}
b = msort(b);
c = msort(c);
int x = 0;
int y = 0;
int z = 0;
while (b.Length != y && c.Length != z) {
if (b[y] < c[z]) {
a[x] = b[y];
//r--
x++;
y++;
} else {
a[x] = c[z];
x++;
z++;
}
}
while (b.Length != y) {
a[x] = b[y];
x++;
y++;
}
while (c.Length != z) {
a[x] = c[z];
x++;
z++;
}
}
return a;
}

Resources