mirror of
https://github.com/JGRennison/OpenTTD-patches.git
synced 2024-11-16 00:12:51 +00:00
eb78cdb2d4
- Supports trains, road vehicles and ships. - Uses A* pathfinding (same codebase as the new ai). - Currently unlimited search depth, so might perform badly on large maps/networks (especially ships). - Will always find a route if there is one. - Allows custom penalties for obstacles to be set in openttd.cfg (npf_ values). - With NPF enabled, ships can have orders that are very far apart. Be careful, this will break (ships get lost) when the old pathfinder is used again. - Feature: Disabling 90 degree turns for trains and ships. - Requires NPF to be enabled. - Ships and trains can no longer make weird 90 degree turns on tile borders. - Codechange: Removed table/directions.h. - table/directions.h contained ugly static tables but was included more than once. The tables, along with a few new ones are in npf.[ch] now. Better suggestions for a location? - Fix: Binary heap in queue.c did not allocate enough space, resulting in a segfault. - Codechange: Rewritten FindFirstBit2x64, added KillFirstBit2x64. - Codechange: Introduced constant INVALID_TILE, to replace the usage of 0 as an invalid tile. Also replaces TILE_WRAPPED. - Codechange: Moved TileAddWrap() to map.[ch] - Add TileIndexDiffCByDir(), TileIndexDiffCByDir(). - Codechange: Moved IsTrainStationTile() to station.h - Add: IsRoadStationTile() and GetRoadStationDir().
673 lines
16 KiB
C
673 lines
16 KiB
C
#include "stdafx.h"
|
|
#include "ttd.h"
|
|
#include "queue.h"
|
|
|
|
static void Stack_Clear(Queue* q, bool free_values)
|
|
{
|
|
uint i;
|
|
if (free_values)
|
|
for (i=0;i<q->data.stack.size;i++)
|
|
free(q->data.stack.elements[i]);
|
|
q->data.stack.size = 0;
|
|
}
|
|
|
|
static void Stack_Free(Queue* q, bool free_values)
|
|
{
|
|
q->clear(q, free_values);
|
|
free(q->data.stack.elements);
|
|
if (q->freeq)
|
|
free(q);
|
|
}
|
|
|
|
static bool Stack_Push(Queue* q, void* item, int priority)
|
|
{
|
|
if (q->data.stack.size == q->data.stack.max_size)
|
|
return false;
|
|
q->data.stack.elements[q->data.stack.size++] = item;
|
|
return true;
|
|
}
|
|
|
|
static void* Stack_Pop(Queue* q)
|
|
{
|
|
void* result;
|
|
if (q->data.stack.size == 0)
|
|
return NULL;
|
|
result = q->data.stack.elements[--q->data.stack.size];
|
|
|
|
return result;
|
|
}
|
|
|
|
static bool Stack_Delete(Queue* q, void* item, int priority)
|
|
{
|
|
return false;
|
|
}
|
|
|
|
static Queue* init_stack(Queue* q, uint max_size)
|
|
{
|
|
q->push = Stack_Push;
|
|
q->pop = Stack_Pop;
|
|
q->del = Stack_Delete;
|
|
q->clear = Stack_Clear;
|
|
q->free = Stack_Free;
|
|
q->data.stack.max_size = max_size;
|
|
q->data.stack.size = 0;
|
|
q->data.stack.elements = malloc(max_size * sizeof(void*));
|
|
q->freeq = false;
|
|
return q;
|
|
}
|
|
|
|
Queue* new_Stack(uint max_size)
|
|
{
|
|
Queue* q = malloc(sizeof(Queue));
|
|
init_stack(q, max_size);
|
|
q->freeq = true;
|
|
return q;
|
|
}
|
|
|
|
/*
|
|
* Fifo
|
|
*/
|
|
|
|
static void Fifo_Clear(Queue* q, bool free_values)
|
|
{
|
|
uint head, tail;
|
|
if (free_values) {
|
|
head = q->data.fifo.head;
|
|
tail = q->data.fifo.tail; /* cache for speed */
|
|
while (head != tail) {
|
|
free(q->data.fifo.elements[tail]);
|
|
tail = (tail + 1) % q->data.fifo.max_size;
|
|
}
|
|
}
|
|
q->data.fifo.head = q->data.fifo.tail = 0;
|
|
}
|
|
|
|
static void Fifo_Free(Queue* q, bool free_values)
|
|
{
|
|
q->clear(q, free_values);
|
|
free(q->data.fifo.elements);
|
|
if (q->freeq)
|
|
free(q);
|
|
}
|
|
|
|
static bool Fifo_Push(Queue* q, void* item, int priority)
|
|
{
|
|
uint next = (q->data.fifo.head + 1) % q->data.fifo.max_size;
|
|
if (next == q->data.fifo.tail)
|
|
return false;
|
|
q->data.fifo.elements[q->data.fifo.head] = item;
|
|
|
|
|
|
q->data.fifo.head = next;
|
|
return true;
|
|
}
|
|
|
|
static void* Fifo_Pop(Queue* q)
|
|
{
|
|
void* result;
|
|
if (q->data.fifo.head == q->data.fifo.tail)
|
|
return NULL;
|
|
result = q->data.fifo.elements[q->data.fifo.tail];
|
|
|
|
|
|
q->data.fifo.tail = (q->data.fifo.tail + 1) % q->data.fifo.max_size;
|
|
return result;
|
|
}
|
|
|
|
static bool Fifo_Delete(Queue* q, void* item, int priority)
|
|
{
|
|
return false;
|
|
}
|
|
|
|
static Queue* init_fifo(Queue* q, uint max_size)
|
|
{
|
|
q->push = Fifo_Push;
|
|
q->pop = Fifo_Pop;
|
|
q->del = Fifo_Delete;
|
|
q->clear = Fifo_Clear;
|
|
q->free = Fifo_Free;
|
|
q->data.fifo.max_size = max_size;
|
|
q->data.fifo.head = 0;
|
|
q->data.fifo.tail = 0;
|
|
q->data.fifo.elements = malloc(max_size * sizeof(void*));
|
|
q->freeq = false;
|
|
return q;
|
|
}
|
|
|
|
Queue* new_Fifo(uint max_size)
|
|
{
|
|
Queue* q = malloc(sizeof(Queue));
|
|
init_fifo(q, max_size);
|
|
q->freeq = true;
|
|
return q;
|
|
}
|
|
|
|
|
|
/*
|
|
* Insertion Sorter
|
|
*/
|
|
|
|
static void InsSort_Clear(Queue* q, bool free_values)
|
|
{
|
|
InsSortNode* node = q->data.inssort.first;
|
|
InsSortNode* prev;
|
|
while (node != NULL) {
|
|
if (free_values)
|
|
free(node->item);
|
|
prev = node;
|
|
node = node->next;
|
|
free(prev);
|
|
|
|
}
|
|
q->data.inssort.first = NULL;
|
|
}
|
|
|
|
static void InsSort_Free(Queue* q, bool free_values)
|
|
{
|
|
q->clear(q, free_values);
|
|
if (q->freeq)
|
|
free(q);
|
|
}
|
|
|
|
static bool InsSort_Push(Queue* q, void* item, int priority)
|
|
{
|
|
InsSortNode* newnode = malloc(sizeof(InsSortNode));
|
|
if (newnode == NULL) return false;
|
|
newnode->item = item;
|
|
newnode->priority = priority;
|
|
if (q->data.inssort.first == NULL || q->data.inssort.first->priority >= priority) {
|
|
newnode->next = q->data.inssort.first;
|
|
q->data.inssort.first = newnode;
|
|
} else {
|
|
InsSortNode* node = q->data.inssort.first;
|
|
while( node != NULL ) {
|
|
if (node->next == NULL || node->next->priority >= priority) {
|
|
newnode->next = node->next;
|
|
node->next = newnode;
|
|
break;
|
|
}
|
|
node = node->next;
|
|
}
|
|
}
|
|
return true;
|
|
}
|
|
|
|
static void* InsSort_Pop(Queue* q)
|
|
{
|
|
InsSortNode* node = q->data.inssort.first;
|
|
void* result;
|
|
if (node == NULL)
|
|
return NULL;
|
|
result = node->item;
|
|
q->data.inssort.first = q->data.inssort.first->next;
|
|
if (q->data.inssort.first)
|
|
assert(q->data.inssort.first->priority >= node->priority);
|
|
free(node);
|
|
return result;
|
|
}
|
|
|
|
static bool InsSort_Delete(Queue* q, void* item, int priority)
|
|
{
|
|
return false;
|
|
}
|
|
|
|
void init_InsSort(Queue* q) {
|
|
q->push = InsSort_Push;
|
|
q->pop = InsSort_Pop;
|
|
q->del = InsSort_Delete;
|
|
q->clear = InsSort_Clear;
|
|
q->free = InsSort_Free;
|
|
q->data.inssort.first = NULL;
|
|
q->freeq = false;
|
|
}
|
|
|
|
Queue* new_InsSort(void)
|
|
{
|
|
Queue* q = malloc(sizeof(Queue));
|
|
init_InsSort(q);
|
|
q->freeq = true;
|
|
return q;
|
|
}
|
|
|
|
|
|
/*
|
|
* Binary Heap
|
|
* For information, see: http://www.policyalmanac.org/games/binaryHeaps.htm
|
|
*/
|
|
|
|
#define BINARY_HEAP_BLOCKSIZE (1 << BINARY_HEAP_BLOCKSIZE_BITS)
|
|
#define BINARY_HEAP_BLOCKSIZE_MASK (BINARY_HEAP_BLOCKSIZE-1)
|
|
|
|
// To make our life easy, we make the next define
|
|
// Because Binary Heaps works with array from 1 to n,
|
|
// and C with array from 0 to n-1, and we don't like typing
|
|
// q->data.binaryheap.elements[i-1] every time, we use this define.
|
|
#define BIN_HEAP_ARR(i) q->data.binaryheap.elements[((i)-1) >> BINARY_HEAP_BLOCKSIZE_BITS][((i)-1) & BINARY_HEAP_BLOCKSIZE_MASK]
|
|
|
|
static void BinaryHeap_Clear(Queue* q, bool free_values)
|
|
{
|
|
/* Free all items if needed and free all but the first blocks of
|
|
* memory */
|
|
uint i,j;
|
|
for (i=0;i<q->data.binaryheap.blocks;i++) {
|
|
if (q->data.binaryheap.elements[i] == NULL) {
|
|
/* No more allocated blocks */
|
|
break;
|
|
}
|
|
/* For every allocated block */
|
|
if (free_values)
|
|
for (j=0;j<(1<<BINARY_HEAP_BLOCKSIZE_BITS);j++) {
|
|
/* For every element in the block */
|
|
if ((q->data.binaryheap.size >> BINARY_HEAP_BLOCKSIZE_BITS) == i
|
|
&& (q->data.binaryheap.size & BINARY_HEAP_BLOCKSIZE_MASK) == j)
|
|
break; /* We're past the last element */
|
|
free(q->data.binaryheap.elements[i][j].item);
|
|
}
|
|
if (i != 0) {
|
|
/* Leave the first block of memory alone */
|
|
free(q->data.binaryheap.elements[i]);
|
|
q->data.binaryheap.elements[i] = NULL;
|
|
}
|
|
}
|
|
q->data.binaryheap.size = 0;
|
|
q->data.binaryheap.blocks = 1;
|
|
}
|
|
|
|
static void BinaryHeap_Free(Queue* q, bool free_values)
|
|
{
|
|
uint i;
|
|
q->clear(q, free_values);
|
|
for (i=0;i<q->data.binaryheap.blocks;i++) {
|
|
if (q->data.binaryheap.elements[i] == NULL)
|
|
break;
|
|
free(q->data.binaryheap.elements[i]);
|
|
}
|
|
if (q->freeq)
|
|
free(q);
|
|
}
|
|
|
|
static bool BinaryHeap_Push(Queue* q, void* item, int priority)
|
|
{
|
|
#ifdef QUEUE_DEBUG
|
|
printf("[BinaryHeap] Pushing an element. There are %d elements left\n", q->data.binaryheap.size);
|
|
#endif
|
|
if (q->data.binaryheap.size == q->data.binaryheap.max_size)
|
|
return false;
|
|
assert(q->data.binaryheap.size < q->data.binaryheap.max_size);
|
|
|
|
if (q->data.binaryheap.elements[q->data.binaryheap.size >> BINARY_HEAP_BLOCKSIZE_BITS] == NULL) {
|
|
/* The currently allocated blocks are full, allocate a new one */
|
|
assert((q->data.binaryheap.size & BINARY_HEAP_BLOCKSIZE_MASK) == 0);
|
|
q->data.binaryheap.elements[q->data.binaryheap.size >> BINARY_HEAP_BLOCKSIZE_BITS] = malloc(BINARY_HEAP_BLOCKSIZE * sizeof(BinaryHeapNode));
|
|
q->data.binaryheap.blocks++;
|
|
#ifdef QUEUE_DEBUG
|
|
printf("[BinaryHeap] Increasing size of elements to %d nodes\n",q->data.binaryheap.blocks * BINARY_HEAP_BLOCKSIZE);
|
|
#endif
|
|
}
|
|
|
|
// Add the item at the end of the array
|
|
BIN_HEAP_ARR(q->data.binaryheap.size+1).priority = priority;
|
|
BIN_HEAP_ARR(q->data.binaryheap.size+1).item = item;
|
|
q->data.binaryheap.size++;
|
|
|
|
// Now we are going to check where it belongs. As long as the parent is
|
|
// bigger, we switch with the parent
|
|
{
|
|
int i, j;
|
|
BinaryHeapNode temp;
|
|
i = q->data.binaryheap.size;
|
|
while (i > 1) {
|
|
// Get the parent of this object (divide by 2)
|
|
j = i / 2;
|
|
// Is the parent bigger then the current, switch them
|
|
if (BIN_HEAP_ARR(i).priority <= BIN_HEAP_ARR(j).priority) {
|
|
temp = BIN_HEAP_ARR(j);
|
|
BIN_HEAP_ARR(j) = BIN_HEAP_ARR(i);
|
|
BIN_HEAP_ARR(i) = temp;
|
|
i = j;
|
|
} else {
|
|
// It is not, we're done!
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool BinaryHeap_Delete(Queue* q, void* item, int priority)
|
|
{
|
|
#ifdef QUEUE_DEBUG
|
|
printf("[BinaryHeap] Deleting an element. There are %d elements left\n", q->data.binaryheap.size);
|
|
#endif
|
|
uint i = 0;
|
|
// First, we try to find the item..
|
|
do {
|
|
if (BIN_HEAP_ARR(i+1).item == item) break;
|
|
i++;
|
|
} while (i < q->data.binaryheap.size);
|
|
// We did not find the item, so we return false
|
|
if (i == q->data.binaryheap.size) return false;
|
|
|
|
// Now we put the last item over the current item while decreasing the size of the elements
|
|
q->data.binaryheap.size--;
|
|
BIN_HEAP_ARR(i+1) = BIN_HEAP_ARR(q->data.binaryheap.size+1);
|
|
|
|
// Now the only thing we have to do, is resort it..
|
|
// On place i there is the item to be sorted.. let's start there
|
|
{
|
|
uint j;
|
|
BinaryHeapNode temp;
|
|
// Because of the fast that Binary Heap uses array from 1 to n, we need to increase
|
|
// i with 1
|
|
i++;
|
|
|
|
for (;;) {
|
|
j = i;
|
|
// Check if we have 2 childs
|
|
if (2*j+1 <= q->data.binaryheap.size) {
|
|
// Is this child smaller than the parent?
|
|
if (BIN_HEAP_ARR(j).priority >= BIN_HEAP_ARR(2*j).priority) {i = 2*j; }
|
|
// Yes, we _need_ to use i here, not j, because we want to have the smallest child
|
|
// This way we get that straight away!
|
|
if (BIN_HEAP_ARR(i).priority >= BIN_HEAP_ARR(2*j+1).priority) { i = 2*j+1; }
|
|
// Do we have one child?
|
|
} else if (2*j <= q->data.binaryheap.size) {
|
|
if (BIN_HEAP_ARR(j).priority >= BIN_HEAP_ARR(2*j).priority) { i = 2*j; }
|
|
}
|
|
|
|
// One of our childs is smaller than we are, switch
|
|
if (i != j) {
|
|
temp = BIN_HEAP_ARR(j);
|
|
BIN_HEAP_ARR(j) = BIN_HEAP_ARR(i);
|
|
BIN_HEAP_ARR(i) = temp;
|
|
} else {
|
|
// None of our childs is smaller, so we stay here.. stop :)
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static void* BinaryHeap_Pop(Queue* q)
|
|
{
|
|
#ifdef QUEUE_DEBUG
|
|
printf("[BinaryHeap] Popping an element. There are %d elements left\n", q->data.binaryheap.size);
|
|
#endif
|
|
void* result;
|
|
if (q->data.binaryheap.size == 0)
|
|
return NULL;
|
|
|
|
// The best item is always on top, so give that as result
|
|
result = BIN_HEAP_ARR(1).item;
|
|
// And now we should get ride of this item...
|
|
BinaryHeap_Delete(q,BIN_HEAP_ARR(1).item, BIN_HEAP_ARR(1).priority);
|
|
|
|
return result;
|
|
}
|
|
|
|
void init_BinaryHeap(Queue* q, uint max_size)
|
|
{
|
|
assert(q);
|
|
q->push = BinaryHeap_Push;
|
|
q->pop = BinaryHeap_Pop;
|
|
q->del = BinaryHeap_Delete;
|
|
q->clear = BinaryHeap_Clear;
|
|
q->free = BinaryHeap_Free;
|
|
q->data.binaryheap.max_size = max_size;
|
|
q->data.binaryheap.size = 0;
|
|
// We malloc memory in block of BINARY_HEAP_BLOCKSIZE
|
|
// It autosizes when it runs out of memory
|
|
q->data.binaryheap.elements = calloc(1, ((max_size - 1) / BINARY_HEAP_BLOCKSIZE*sizeof(BinaryHeapNode)) + 1);
|
|
q->data.binaryheap.elements[0] = malloc(BINARY_HEAP_BLOCKSIZE * sizeof(BinaryHeapNode));
|
|
q->data.binaryheap.blocks = 1;
|
|
q->freeq = false;
|
|
#ifdef QUEUE_DEBUG
|
|
printf("[BinaryHeap] Initial size of elements is %d nodes\n",(1024));
|
|
#endif
|
|
}
|
|
|
|
Queue* new_BinaryHeap(uint max_size) {
|
|
Queue* q = malloc(sizeof(Queue));
|
|
init_BinaryHeap(q, max_size);
|
|
q->freeq = true;
|
|
return q;
|
|
}
|
|
|
|
// Because we don't want anyone else to bother with our defines
|
|
#undef BIN_HEAP_ARR
|
|
|
|
/*
|
|
* Hash
|
|
*/
|
|
|
|
void init_Hash(Hash* h, Hash_HashProc* hash, int num_buckets) {
|
|
/* Allocate space for the Hash, the buckets and the bucket flags */
|
|
int i;
|
|
assert(h);
|
|
#ifdef HASH_DEBUG
|
|
debug("Allocated hash: %p", h);
|
|
#endif
|
|
h->hash = hash;
|
|
h->size = 0;
|
|
h->num_buckets = num_buckets;
|
|
h->buckets = malloc(num_buckets * (sizeof(HashNode) + sizeof(bool)));
|
|
#ifdef HASH_DEBUG
|
|
debug("Buckets = %p", h->buckets);
|
|
#endif
|
|
h->buckets_in_use = (bool*)(h->buckets + num_buckets);
|
|
h->freeh = false;
|
|
for (i=0;i<num_buckets;i++)
|
|
h->buckets_in_use[i] = false;
|
|
}
|
|
|
|
Hash* new_Hash(Hash_HashProc* hash, int num_buckets) {
|
|
Hash* h = malloc(sizeof(Hash));
|
|
init_Hash(h, hash, num_buckets);
|
|
h->freeh = true;
|
|
return h;
|
|
}
|
|
|
|
void delete_Hash(Hash* h, bool free_values) {
|
|
uint i;
|
|
/* Iterate all buckets */
|
|
for (i=0;i<h->num_buckets;i++)
|
|
{
|
|
if (h->buckets_in_use[i]) {
|
|
HashNode* node;
|
|
/* Free the first value */
|
|
if (free_values)
|
|
free(h->buckets[i].value);
|
|
node = h->buckets[i].next;
|
|
while (node != NULL) {
|
|
HashNode* prev = node;
|
|
node = node->next;
|
|
/* Free the value */
|
|
if (free_values)
|
|
free(prev->value);
|
|
/* Free the node */
|
|
free(prev);
|
|
}
|
|
}
|
|
}
|
|
free(h->buckets);
|
|
/* No need to free buckets_in_use, it is always allocated in one
|
|
* malloc with buckets */
|
|
#ifdef HASH_DEBUG
|
|
debug("Freeing Hash: %p", h);
|
|
#endif
|
|
if (h->freeh)
|
|
free(h);
|
|
}
|
|
|
|
void clear_Hash(Hash* h, bool free_values)
|
|
{
|
|
uint i;
|
|
HashNode* node;
|
|
/* Iterate all buckets */
|
|
for (i=0;i<h->num_buckets;i++)
|
|
{
|
|
if (h->buckets_in_use[i]) {
|
|
h->buckets_in_use[i] = false;
|
|
/* Free the first value */
|
|
if (free_values)
|
|
free(h->buckets[i].value);
|
|
node = h->buckets[i].next;
|
|
while (node != NULL) {
|
|
HashNode* prev = node;
|
|
node = node->next;
|
|
if (free_values)
|
|
free(prev->value);
|
|
free(prev);
|
|
}
|
|
}
|
|
}
|
|
h->size = 0;
|
|
}
|
|
|
|
/* Finds the node that that saves this key pair. If it is not
|
|
* found, returns NULL. If it is found, *prev is set to the
|
|
* node before the one found, or if the node found was the first in the bucket
|
|
* to NULL. If it is not found, *prev is set to the last HashNode in the
|
|
* bucket, or NULL if it is empty. prev can also be NULL, in which case it is
|
|
* not used for output.
|
|
*/
|
|
static HashNode* Hash_FindNode(Hash* h, uint key1, uint key2, HashNode** prev_out)
|
|
{
|
|
uint hash = h->hash(key1, key2);
|
|
HashNode* result = NULL;
|
|
#ifdef HASH_DEBUG
|
|
debug("Looking for %u, %u", key1, key2);
|
|
#endif
|
|
/* Check if the bucket is empty */
|
|
if (!h->buckets_in_use[hash]) {
|
|
if (prev_out)
|
|
*prev_out = NULL;
|
|
result = NULL;
|
|
/* Check the first node specially */
|
|
} else if (h->buckets[hash].key1 == key1 && h->buckets[hash].key2 == key2) {
|
|
/* Save the value */
|
|
result = h->buckets + hash;
|
|
if (prev_out)
|
|
*prev_out = NULL;
|
|
#ifdef HASH_DEBUG
|
|
debug("Found in first node: %p", result);
|
|
#endif
|
|
/* Check all other nodes */
|
|
} else {
|
|
HashNode* prev = h->buckets + hash;
|
|
HashNode* node = prev->next;
|
|
while (node != NULL) {
|
|
if (node->key1 == key1 && node->key2 == key2) {
|
|
/* Found it */
|
|
result = node;
|
|
#ifdef HASH_DEBUG
|
|
debug("Found in other node: %p", result);
|
|
#endif
|
|
break;
|
|
}
|
|
prev = node;
|
|
node = node->next;
|
|
}
|
|
if (prev_out)
|
|
*prev_out = prev;
|
|
}
|
|
#ifdef HASH_DEBUG
|
|
if (result == NULL)
|
|
debug("Not found");
|
|
#endif
|
|
return result;
|
|
}
|
|
|
|
void* Hash_Delete(Hash* h, uint key1, uint key2) {
|
|
void* result;
|
|
HashNode* prev; /* Used as output var for below function call */
|
|
HashNode* node = Hash_FindNode(h, key1, key2, &prev);
|
|
|
|
if (node == NULL) {
|
|
/* not found */
|
|
result = NULL;
|
|
} else if (prev == NULL) {
|
|
/* It is in the first node, we can't free that one, so we free
|
|
* the next one instead (if there is any)*/
|
|
/* Save the value */
|
|
result = node->value;
|
|
if (node->next != NULL) {
|
|
HashNode* next = node->next;
|
|
/* Copy the second to the first */
|
|
*node = *next;
|
|
/* Free the second */
|
|
#ifndef NOFREE
|
|
free(next);
|
|
#endif
|
|
} else {
|
|
/* This was the last in this bucket */
|
|
/* Mark it as empty */
|
|
uint hash = h->hash(key1, key2);
|
|
h->buckets_in_use[hash] = false;
|
|
}
|
|
} else {
|
|
/* It is in another node */
|
|
/* Save the value */
|
|
result = node->value;
|
|
/* Link previous and next nodes */
|
|
prev->next = node->next;
|
|
/* Free the node */
|
|
#ifndef NOFREE
|
|
free(node);
|
|
#endif
|
|
}
|
|
if (result != NULL)
|
|
h->size--;
|
|
return result;
|
|
}
|
|
|
|
|
|
void* Hash_Set(Hash* h, uint key1, uint key2, void* value) {
|
|
HashNode* prev;
|
|
HashNode* node = Hash_FindNode(h, key1, key2, &prev);
|
|
void* result = NULL;
|
|
if (node != NULL) {
|
|
/* Found it */
|
|
result = node->value;
|
|
node->value = value;
|
|
return result;
|
|
}
|
|
/* It is not yet present, let's add it */
|
|
if (prev == NULL) {
|
|
/* The bucket is still empty */
|
|
uint hash = h->hash(key1, key2);
|
|
h->buckets_in_use[hash] = true;
|
|
node = h->buckets + hash;
|
|
} else {
|
|
/* Add it after prev */
|
|
node = malloc(sizeof(HashNode));
|
|
prev->next = node;
|
|
}
|
|
node->next = NULL;
|
|
node->key1 = key1;
|
|
node->key2 = key2;
|
|
node->value = value;
|
|
h->size++;
|
|
return NULL;
|
|
}
|
|
|
|
void* Hash_Get(Hash* h, uint key1, uint key2) {
|
|
HashNode* node = Hash_FindNode(h, key1, key2, NULL);
|
|
#ifdef HASH_DEBUG
|
|
debug("Found node: %p", node);
|
|
#endif
|
|
if (node == NULL) {
|
|
/* Node not found */
|
|
return NULL;
|
|
} else {
|
|
return node->value;
|
|
}
|
|
}
|
|
|
|
uint Hash_Size(Hash* h) {
|
|
return h->size;
|
|
}
|