# 2020q3 Homework (quiz1)
contributed by < `hankluo6` >
## Linked List
### `add_entry`
* 將 `new_node` 加入到 linked list 的尾端,使用 pointer to pointer 來指向要加入的節點位置的「地址」。
```c
void add_entry(node_t **head, int new_value)
{
node_t **indirect = head;
node_t *new_node = malloc(sizeof(node_t));
new_node->value = new_value;
new_node->next = NULL;
AA1
while (*indirect)
indirect = &(*indirect)->next;
AA2
}
```
#### `AA1`
- [x] `(a)` `assert(new_node)`
先使用 `assert` 來檢驗 `malloc` 是否成功。
#### `AA2`
- [x] `(b)` `*indirect = new_node`
當找到位置後,將該位址的值設為 `new_node`。
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "pointer to pointer";
indirect;
}
subgraph cluster_1 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node1;
node2;
}
indirect -> head;
head -> node1;
node1 -> node2;
new_node
}
```
* 先將 `indirect` 設定為 `head` 的 address
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "pointer to pointer";
indirect;
}
subgraph cluster_1 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node1;
node2;
}
indirect -> node1 [constraint = none];
head -> node1;
node1 -> node2;
new_node
}
```
* 在 `while` 中持續尋找有 `NULL` 值的位置,`indirect` 為 `&node->next`,也就是存有 node 值的 address
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "pointer to pointer";
indirect;
}
subgraph cluster_1 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node1;
node2;
}
indirect -> node2;
head -> node1;
node1 -> node2;
new_node
}
```
* 持續尋找 `NULL` 值的位置
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "pointer to pointer";
indirect;
}
subgraph cluster_1 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node1;
node2;
NULL;
}
indirect -> NULL;
head -> node1;
node1 -> node2;
node2 -> NULL
new_node
}
```
* 找到後跳出迴圈,`indirect` 為 `NULL` 位置的 address
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "pointer to pointer";
indirect;
}
subgraph cluster_1 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node1;
node2;
new_node;
}
indirect -> new_node;
head -> node1;
node1 -> node2;
node2 -> new_node
}
```
* 將 `new_node` 設定為該 address 的內容,完成插入在 linked list 尾端
#### 討論
如果 `malloc` 無法分配記憶體,`malloc` 會回傳 `NULL`。`NULL` 的定義在規格書中有寫道:
:::info
An integer constant expression with the value 0, or such an expression cast to type void *, is called a null pointer constant.
:::
得知 `NULL` 定義為 `0` 或為 `(void*)0`。對 `NULL` 取其成員,應為未定義行為。
透過程式碼驗證
```c
typedef struct my_struct {
int val;
} node;
int main() {
node *head = NULL;
printf("%d\n", head->val);
}
```
出現 segmentation fault,與預期的一樣。
再看看 `assert` 的作用,從 man page 中寫道:
:::info
If expression is false (i.e., compares equal to zero), assert()
prints an error message to standard error and terminates the program
by calling abort(3).
:::
可推知當 expression 為 `NULL (0)` 時,會呼叫 `abort(3)`。
如果這邊 `assert` 的作用是用來檢驗 `malloc` 是否成功,我認為應放在 `new_node->value` 前比較合適,否則沒有作用,`new_node` 為 NULL 時,`new_node->value` 就會丟出 segmentation fault。
### `find_entry`
```c
node_t *find_entry(node_t *head, int value)
{
node_t *current = head;
for (; current && current->value != value; current = current->next)
/* interate */;
return current;
}
```
`find_entry` 函式用來尋找指定 value 的節點
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node1;
node2;
}
current -> head;
head -> node1;
node1 -> node2;
}
```
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node1;
node2;
}
current -> node1;
head -> node1;
node1 -> node2;
}
```
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node1;
node2;
}
current -> node2;
head -> node1;
node1 -> node2;
}
```
### `remove_entry`
```c
void remove_entry(node_t **head, node_t *entry)
{
node_t **indirect = head;
while ((*indirect) != entry)
indirect = &(*indirect)->next;
*indirect = entry->next;
free(entry);
}
```
`remove_entry` 使用 pointer to pointer,可以減少刪除 `head` 時需要的判斷式
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "pointer to pointer";
indirect;
}
subgraph cluster_1 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
entry;
node2;
}
indirect -> entry;
head -> entry;
entry -> node2;
}
```
* `while` 迴圈找到 `entry` 的位置
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "pointer to pointer";
indirect;
}
subgraph cluster_1 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node2;
}
indirect -> node2;
head -> node2;
}
```
* 因為 `indirect` 為 `&head->next`,所以直接將該 address 上的值更新為 `node2` 即可
### `swap_pair`
```c
node_t *swap_pair(node_t *head)
{
for (node_t **node = &head; *node && (*node)->next; BB1) {
node_t *tmp = *node;
BB2;
tmp->next = (*node)->next;
(*node)->next = tmp;
}
return head;
}
```
#### `BB1`
- [x] `(e)` `node = &(*node)->next->next`
可以先把 type 不合的選項刪掉,`(a), (d), (f)` 的 pointer type 不合先移除。接著分析迴圈作用,從 `*node && (*node)->next` 可猜測每次迴圈應為移動兩個 node。注意迴圈內的程式碼,可看出我們沒設定 `*node` 前節點的 `next`,這是因為我們 `node` 是以之前 `node->next` 的 address 來更新,所以答案為 `(e)`
:::warning
這邊需要兩次 `next` 的原因是因為 `*node` 已被設定到 `tmp` 的前面
:::
#### `BB2`
- [x] `(c)` `*node = (*node)->next`
我們需要將 `node` 往後移動一個節點,型態吻合的只有 `(b), (c)` 兩個選項。
深入討論 `(b)` 選項,先執行 `node_t *tmp = *node`,`tmp` 目前為第一個節點,假如執行 `node = &(*node)->next` 的話,`node` 現在為第一個節點的 `next` 的 address,再來, `tmp->next = (*node)->next` 這行會讓第一個節點 `tmp` 的 `next` 指向 `*node` 的 `next`。但 `&tmp->next == node`,對 `tmp->next` 的改動會影響到 `*node` 的值,所以 `tmp->next = (*node)->next` 事實上與 `*node = (*node)->next` 等效。造成迴圈的 `node = &(*node)->next->next` 有不正確的結果。
故答案應為 `(c)`
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "pointer to pointer";
node0 [label = "node"];
}
subgraph cluster_1 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node1
node2;
node3
}
node0 -> head;
head -> node1;
node1 ->node2;
node2 -> node3;
tmp -> head;
}
```
* 先將 `*tmp` 指向 `*node`,`tmp` 為要 swap 的第一個節點
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "pointer to pointer";
node0 [label = "node"];
}
subgraph cluster_1 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node1
node2;
node3
}
node0 -> head [style=invis];
node0 -> node1;
head -> node1;
node1 ->node2;
node2 -> node3;
tmp -> head;
}
```
* `*node = (*node)->next` 將 node 位址的值前進一個節點,現在 `*node` 為要 swap 的第二個節點
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "pointer to pointer";
node0 [label = "node"];
}
subgraph cluster_1 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node1
node2;
node3
}
node0 -> head [style=invis];
node0 -> node1;
head -> node1 [style=invis];
node1 -> node2;
node2 -> node3;
tmp -> head;
head -> node2;
}
```
* `tmp` 的 `next` 指向 `*node` 的後方,也就是將 swap 的第一個 node 的 next 指向第三個節點
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "pointer to pointer";
node0 [label = "node"];
}
subgraph cluster_1 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node1
node2;
node3
}
node0 -> head [style=invis];
node0 -> node1;
head -> node1 [style=invis];
node1 -> node2 [style=invis];
node1 -> head;
node2 -> node3;
tmp -> head;
head -> node2;
}
```
* `(*node)->next` 要指回 `tmp`,實現 swap 的部分
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "pointer to pointer";
node0 [label = "node"];
}
subgraph cluster_1 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node1
node2;
node3
}
node0 -> head [style=invis];
node0 -> node2;
head -> node1 [style=invis];
node1 -> node2 [style=invis];
node1 -> head;
node2 -> node3;
tmp -> head;
head -> node2;
}
```
* 因為 `*node` 與 `(*node)->next` 已經完成 swap,將 `node` 更新為下兩個節點的位置,根據 loop invariant,可以兩兩 swap 每個節點
#### 討論
用 gdb 來檢驗上述 BB2 選項的分析是否正確
假設將 BB2 設為 `node = &(*node)->next`
```
(gdb) n
32 tmp->next = (*node)->next;
(gdb) p node
$1 = (node_t **) 0x555555756268
(gdb) p *node
$2 = (node_t *) 0x555555756280
(gdb) p &(*node)->next
$3 = (struct __node **) 0x555555756288
(gdb) p (*node)->next
$4 = (struct __node *) 0x5555557562a0
(gdb) n
33 (*node)->next = tmp;
(gdb) p node
$5 = (node_t **) 0x555555756268
(gdb) p *node
$6 = (node_t *) 0x5555557562a0
```
可以看到 node 的值在 `tmp->next = (*node)->next` 後被更新為下個節點,與推測符合,所以該選項為非。
### `reverse`
```c
node_t *reverse(node_t *head)
{
node_t *cursor = NULL;
while (head) {
node_t *next = head->next;
CCC;
head = next;
}
return cursor;
}
```
#### `CCC`
- [x] `(b)` `head->next = cursor; cursor = head`
從現有程式碼可以看到,`node_t *next = head->next;` 是將 `next` 移動到 `head` 後方,而 `head = next` 則是將 `head` 向後移動。所以缺少的部分應為將 `next` 反轉的部分。反轉的目標從 linked list 的頭之 `prev` 開始,隨著 `head` 移動而跟著移動。可以推測出反轉的目標應為 `cursor`。再從選項中找有執行上面步驟的程式碼 `(b)`。
```graphviz
digraph linked_list{
rankdir=LR;
node [shape=record];
a [label="{ <data> a | <ref> }"];
b [label="{ <data> b | <ref> }"];
c [label="{ <data> c | <ref> }"];
d [label="{ <data> d | <ref> }"];
a:ref:c -> b:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
b:ref:c -> c:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
c:ref:c -> d:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
cursor -> NULL;
head -> a;
}
```
* 進入迴圈前的狀態,`cursor` 為 `NULL` 表示 `head` 的前一個節點為 `NULL`
```graphviz
digraph linked_list{
rankdir=LR;
node [shape=record];
a [label="{ <data> a | <ref> }"];
b [label="{ <data> b | <ref> }"];
c [label="{ <data> c | <ref> }"];
d [label="{ <data> d | <ref> }"];
a:ref:c -> b:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
b:ref:c -> c:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
c:ref:c -> d:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
cursor -> NULL;
head -> a;
next -> b;
}
```
* 新增 `next` 指標用來指向 `head->next`,用來在 `head` 反轉後,還能知道原本 `head->next` 的值
```graphviz
digraph linked_list{
rankdir=LR;
node [shape=record];
a [label="{ <data> a | <ref> }"];
b [label="{ <data> b | <ref> }"];
c [label="{ <data> c | <ref> }"];
d [label="{ <data> d | <ref> }"];
a:ref:c -> NULL [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
b:ref:c -> c:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
c:ref:c -> d:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
cursor -> NULL;
head -> a;
next -> b;
}
```
* 將 `head->next` 反轉,指向為 `cursor`,這邊 `cursor` 為 `head` 前節點 `NULL`
```graphviz
digraph linked_list{
rankdir=LR;
node [shape=record];
a [label="{ <data> a | <ref> }"];
b [label="{ <data> b | <ref> }"];
c [label="{ <data> c | <ref> }"];
d [label="{ <data> d | <ref> }"];
a:ref:c -> NULL [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
b:ref:c -> c:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
c:ref:c -> d:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
cursor -> a;
head -> a;
next -> b;
}
```
`cursor` 需移動到 `head`,紀錄下次反轉需指向的節點
```graphviz
digraph linked_list{
rankdir=LR;
node [shape=record];
a [label="{ <data> a | <ref> }"];
b [label="{ <data> b | <ref> }"];
c [label="{ <data> c | <ref> }"];
d [label="{ <data> d | <ref> }"];
a:ref:c -> NULL [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
b:ref:c -> c:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
c:ref:c -> d:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
cursor -> a;
head -> b;
next -> b;
}
```
* 將 `head` 往後移動,準備反轉第二個節點 `b`,而 `a` 已經反轉完畢,根據 loop invariant,最後將反轉整個 linked list
## 改寫 `swap_pair` 與 `reverse`
```c
void swap_pair(node_t **head)
{
for (node_t **node = head; *node && (*node)->next; node = &((*node)->next->next)) {
node_t *tmp = *node;
*node = (*node)->next;
tmp->next = (*node)->next;
(*node)->next = tmp;
}
}
```
只需將傳進去的參數 `head` 改為 pointer to pointer 的形式就行了,因為本身 `**node` 就已經在 `&head` 上操作,所以其餘部分皆不用改動。
```c
void reverse(node_t **head)
{
node_t *cursor = NULL;
while (*head) {
node_t *next = (*head)->next;
(*head)->next = cursor;
cursor = *head;
*head = next;
}
*head = cursor;
}
```
將 `head` 改為 pointer to pointer,迴圈內每個步驟只會改變 `*head` 的值,並固定真正的記憶體位置,就不需要回傳 `head`。須注意迴圈的判斷式為 `*head` 是否為 `NULL`,故跳出迴圈後,`*head == NULL`,需將 `*head` 設定為最後一個節點 `cursor`。
## 以遞迴方式實現 `reverse`
```c
node_t *rev_recursive(node_t *prev, node_t *head)
{
if (!head)
return prev;
node_t *next = head->next;
head->next = prev;
return rev_recursive(head, next);
}
void reverse(node_t **head)
{
*head = rev_recursive(NULL, *head);
}
```
遞迴的方式很簡單,每次進入 `rev_recursive` 時,保證 `prev` 前的 node 皆已經 `reverse`,此時再將 `head->next = prev` 就能讓 `head` 以前皆完成 `reverse`,在繼續遞迴下去。終止條件為 `head` 為空,也就表示 `prev` 以前已經完成 `reverse`。
## Fisher–Yates shuffle
參考 [wiki](https://en.wikipedia.org/wiki/Fisher%E2%80%93Yates_shuffle) 上的 pseudocode
```
-- To shuffle an array a of n elements (indices 0..n-1):
for i from 0 to n−2 do
j ← random integer such that i ≤ j < n
exchange a[i] and a[j]
```
### Version 1.
```c
void shuffle(node_t **head)
{
int size = 0;
node_t *current = *head;
node_t *target = *head;
node_t *prev_current = NULL;
node_t *prev_target = NULL;
for (node_t *tmp = *head; tmp; tmp = tmp->next, ++size)
/* calculate linked list size */;
for (int i = 0; i < size - 1; ++i) {
prev_target = target;
target = current;
int pos = rand() % (size - i);
while (pos--) {
prev_target = target;
target = target->next;
}
if (prev_current)
prev_current->next = target;
if (prev_target)
prev_target->next = current;
node_t *tmp = current->next;
current->next = target->next;
target->next = tmp;
prev_current = target;
current = target->next;
if (!i)
*head = target;
}
}
```
紀錄要交換的兩個 node 的 `node->prev` 與 `node`,再互相 swap 就行了。
既然有 `node->prev` 的話,我們可以不用特別紀錄需要交換的 `node` 的值,只要找要交換的 `node` 的 `prev` 就行了。
### Version 2.
```c
void swap(node_t **x, node_t **y)
{
node_t *tmp = *x;
*x = *y;
*y = tmp;
}
void shuffle(node_t **head)
{
int size = 0;
node_t *dummy = malloc(sizeof(node_t));
dummy->next = *head;
node_t *current = dummy;
node_t *target = dummy;
for (node_t *tmp = *head; tmp; tmp = tmp->next, ++size)
/* calculate linked list size */;
for (int i = 0; i < size; ++i) {
target = current;
int pos = rand() % (size - i);
while (pos-- > 0) {
target = target->next;
}
swap(¤t->next, &target->next);
current = current->next;
target = target->next;
swap(¤t->next, &target->next);
}
*head = dummy->next;
free(dummy);
}
```
增加 `dummy` 節點來減少 `target` 選到 `head` 的時的判斷式,程式的邏輯變得很簡單,找到要交換的兩節點之 `prev` 後,swap 目前兩節點之 `next`,再將節點往後移動,並再一次 swap 兩邊的 `next`。
```graphviz
digraph linked_list{
rankdir=LR;
node [shape=record];
dummy [label="{ <data> dummy | <ref> }"];
a [label="{ <data> a | <ref> }"];
b [label="{ <data> b | <ref> }"];
c [label="{ <data> c | <ref> }"];
d [label="{ <data> d | <ref> }"];
dummy:ref:c -> a:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
a:ref:c -> b:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
b:ref:c -> c:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
c:ref:c -> d:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
current -> dummy;
target -> dummy;
}
```
* 將 `current` 與 `target` 設置為 `dummy`,一開始要交換的 node 為 `a`,只是我們的 `current` 目前指向要交換的節點之 `prev`,`target` 也一樣指向為要交換的節點之 `prev`
```graphviz
digraph linked_list{
rankdir=LR;
node [shape=record];
dummy [label="{ <data> dummy | <ref> }"];
a [label="{ <data> a | <ref> }"];
b [label="{ <data> b | <ref> }"];
c [label="{ <data> c | <ref> }"];
d [label="{ <data> d | <ref> }"];
dummy:ref:c -> a:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
a:ref:c -> b:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
b:ref:c -> c:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
c:ref:c -> d:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
current -> dummy;
target -> b;
}
```
* `target` 選到 `b`,所以 `a` 與 `c` 需要交換
```graphviz
digraph linked_list{
rankdir=LR;
node [shape=record];
dummy [label="{ <data> dummy | <ref> }"];
a [label="{ <data> a | <ref> }"];
b [label="{ <data> b | <ref> }"];
c [label="{ <data> c | <ref> }"];
d [label="{ <data> d | <ref> }"];
dummy:ref:c -> c:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
a:ref:c -> b:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
b:ref:c -> a:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
c:ref:c -> d:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
current -> dummy;
target -> b;
}
```
* 將 `a->prev` 的 `next` 與 `c->prev` 的 `next` 互換
```graphviz
digraph linked_list{
rankdir=LR;
node [shape=record];
dummy [label="{ <data> dummy | <ref> }"];
a [label="{ <data> a | <ref> }"];
b [label="{ <data> b | <ref> }"];
c [label="{ <data> c | <ref> }"];
d [label="{ <data> d | <ref> }"];
dummy:ref:c -> c:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
a:ref:c -> b:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
b:ref:c -> a:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
c:ref:c -> d:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
current -> c;
target -> a;
}
```
* 移動 `target` 與 `current` 至它們的 `next` 節點
```graphviz
digraph linked_list{
rankdir=LR;
node [shape=record];
dummy [label="{ <data> dummy | <ref> }"];
a [label="{ <data> a | <ref> }"];
b [label="{ <data> b | <ref> }"];
c [label="{ <data> c | <ref> }"];
d [label="{ <data> d | <ref> }"];
dummy:ref:c -> c:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
a:ref:c -> d:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
b:ref:c -> a:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
c:ref:c -> b:data [arrowhead=vee, arrowtail=dot, dir=both, tailclip=false];
current -> c;
target -> a;
}
```
* 重複上面的 swap 步驟,將 `a` 的 `next` 與 `c` 的 `next` 互換
### Version 3.
```c
void swap(node_t **x, node_t **y)
{
node_t *tmp = *x;
*x = *y;
*y = tmp;
}
void shuffle(node_t **head)
{
int size = 0;
node_t **current = head;
node_t **target;
for (node_t *tmp = *head; tmp; tmp = tmp->next, ++size)
/* calculate linked list size */;
for (int i = 0; i < size - 1; ++i, current = &(*current)->next) {
target = current;
int pos = rand() % (size - i);
while (pos-- > 0)
target = &(*target)->next;
swap(target, current);
swap(&(*target)->next, &(*current)->next);
}
}
```
使用 pointer to pointer 來取代 `dummy` head,減少所需記憶體。
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "pointer to pointer";
cur [label = "current"];
tar [label = "target"];
}
subgraph cluster_1 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node1
node2;
node3
}
cur -> head;
tar -> head;
head -> node1;
node1 ->node2;
node2 -> node3;
}
```
* 一開始先將 `current` 與 `target` 指向 `head` 的 address
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "pointer to pointer";
cur [label = "current"];
tar [label = "target"];
}
subgraph cluster_1 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node1
node2;
node3
}
cur -> head;
tar -> node2;
head -> node1;
node1 ->node2;
node2 -> node3;
}
```
* `target` 選到 `node2`,指向 `node2` 的 adress
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "pointer to pointer";
cur [label = "current"];
tar [label = "target"];
}
subgraph cluster_1 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node1
node2;
node3
}
cur -> node2;
tar -> head;
head -> node1;
node2 -> node3;
node1 -> head;
}
```
* 將 `head` 與 `node2` 交換,注意這邊交換的是節點,而 `current` 與 `target` 指向的位置不變
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "pointer to pointer";
cur [label = "current"];
tar [label = "target"];
}
subgraph cluster_1 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node1
node2;
node3
}
cur -> node2;
tar -> head;
head -> node3;
node2 -> node1;
node1 -> head;
}
```
* 將 `target->next` 與 `current->next` 交換,完成 swap
```graphviz
digraph linked_list{
rankdir = LR;
node [shape = record];
subgraph cluster_0 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "pointer to pointer";
cur [label = "current"];
tar [label = "target"];
}
subgraph cluster_1 {
style = "filled";
color = "black";
fillcolor = "lightyellow";
node [style = filled, fillcolor = white, color = black];
label = "linked list";
head;
node1
node2;
node3
}
cur -> node1;
tar -> node1;
head -> node3;
node2 -> node1;
node1 -> head;
}
```
* 最後將 `current` 往後移動,進入下輪迴圈
### Valgrind
使用 valgrind 來比較三種版本的記憶體
測試程式
```c
int main(int argc, char const *argv[])
{
node_t *head = NULL;
for (int i = 0; i < 100000; ++i)
add_entry(&head, rand() % 100);
shuffle(&head);
return 0;
}
```
使用 massif 分析,記得開啟 stacks 紀錄堆疊區塊
```
valgrind --tool=massif --stacks=yes ./a.out
```
為了版面整潔,只列出程式結束前幾個 snapshots 的記憶體情形
* #### Version 1.
```
--------------------------------------------------------------------------------
n time(i) total(B) useful-heap(B) extra-heap(B) stacks(B)
--------------------------------------------------------------------------------
46 58,905,268,954 2,400,376 1,600,000 800,000 376
47 60,293,938,629 2,400,376 1,600,000 800,000 376
48 61,682,533,376 2,400,376 1,600,000 800,000 376
49 63,534,166,248 2,400,376 1,600,000 800,000 376
50 64,287,194,889 2,400,376 1,600,000 800,000 376
51 65,040,246,906 2,400,376 1,600,000 800,000 376
```
* #### Version 2.
```
--------------------------------------------------------------------------------
n time(i) total(B) useful-heap(B) extra-heap(B) stacks(B)
--------------------------------------------------------------------------------
88 58,440,902,265 2,400,384 1,600,016 800,008 360
89 58,903,802,669 2,400,384 1,600,016 800,008 360
90 59,366,668,507 2,400,384 1,600,016 800,008 360
91 59,829,453,069 2,400,384 1,600,016 800,008 360
```
* #### Version 3.
```
--------------------------------------------------------------------------------
n time(i) total(B) useful-heap(B) extra-heap(B) stacks(B)
--------------------------------------------------------------------------------
88 58,443,219,786 2,400,360 1,600,000 800,000 360
89 58,906,109,686 2,400,360 1,600,000 800,000 360
90 59,368,915,633 2,400,360 1,600,000 800,000 360
91 59,831,815,688 2,400,360 1,600,000 800,000 360
92 60,294,835,559 2,400,360 1,600,000 800,000 360
93 60,757,641,570 2,400,360 1,600,000 800,000 360
94 61,220,406,226 2,400,360 1,600,000 800,000 360
95 61,683,176,372 2,400,360 1,600,000 800,000 360
96 62,145,982,621 2,400,360 1,600,000 800,000 360
```
* 從 useful-heap 來看,因為 version 2 有額外 `malloc` 一個 `dummy` node,所以比其他兩種版本多了 16 bytes 的 heap。
* 從 stacks 來看,version 1 較其他兩種版本多了兩個指標變數,在實驗環境中 ( 64位元 ),pointer 為 8 bytes,所以 version 1 多了 16 bytes。
最理想的版本為 version 3,使用到的 total memory 最小,為 2,400,360 bytes。
###### tags: `linux2020`