Home > Backend Development > PHP Tutorial > About the analysis of HashTable in PHP source code

About the analysis of HashTable in PHP source code

不言
Release: 2023-04-01 21:38:01
Original
1698 people have browsed it

这篇文章主要介绍了关于PHP源码中HashTable的解析,有着一定的参考价值,现在分享给大家,有需要的朋友可以参考一下

PHP源码中HashTable的简单示例 前些日子看了那篇对hasttable的介绍,于是也想自己运行一下,可是对于源码的调试不是太在行。 所以想了个办法:自己把PHP源码中的一些简单操作提取出来,自己运行一下,查看输出或调试。 于是花费了三天的空闲时间把一些相关的东西提取出来,主要是Zend目录下的zend_alloc.c,zend_alloc.h,zend_hash.c,zend_hash.h四个文件。 将与PHP相关的内存分配去掉,默认使用系统自带的内存分配方式。 另外:一些注释是http://www.phppan.com/2009/12/zend-hashtable/中所引用文章中的相关信息。 作者地址:http://www.phpinternals.com 下面的代码是一个可以运行的C程序,它初始化一个容量为50的hashtable(实际上分配了64个),然后将30到68,写入hash table,并将这个hash table 打印出来。 相信这会给一些想学习源码的童鞋一些帮助。 源代码如下:

  <!-- #include <stdio.h-->#include #include typedef unsigned long ulong;typedef unsigned int uint;typedef unsigned char zend_bool;typedef unsigned int size_t;typedef void (*dtor_func_t)(void *pDest);typedef ulong (*hash_func_t)(char *arKey, uint nKeyLength);#define SUCCESS 0#define FAILURE -1 /* this MUST stay a negative number, or it may affect functions! */ #define HASH_UPDATE (1&lt;&lt;0)#define HASH_ADD (1&lt;&lt;1)#define HASH_NEXT_INSERT(1&lt;&lt;2) #define HASH_DEL_KEY 0 #define perealloc_recoverable(ptr, size, persistent) (__zend_realloc((ptr), (size)))#define pefree_rel(ptr, persistent)(free(ptr))//此处省略了使用PHP的内存分配函数#define pemalloc_rel(size, persistent) (__zend_malloc(size))#define perealloc_rel(ptr, size, persistent) (__zend_realloc((ptr), (size)))#define pemalloc(size, persistent) (__zend_malloc(size))#define pefree(ptr, persistent)  (free(ptr)) inline static void * __zend_malloc(size_t len) {
    void *tmp = malloc(len);
    if (tmp) {
        return tmp;
    }  
    fprintf(stderr, "Out of memory\n");
    exit(1);} 
 inline static void * __zend_realloc(void *p, size_t len) {
    p = realloc(p, len);
    if (p) {
        return p;
    }  
    fprintf(stderr, "Out of memory\n");
    exit(1);} 
 typedef struct bucket {
    ulong h;       /* Used for numeric indexing */
    uint nKeyLength;     /* key 长度 */
    void *pData;      /* 指向Bucket中保存的数据的指针 */
    void *pDataPtr;     /* 指针数据 */
    struct bucket *pListNext;   /* 指向HashTable桶列中下一个元素 */
    struct bucket *pListLast;    /* 指向HashTable桶列中前一个元素 */
    struct bucket *pNext;    /* 指向具有同一个hash值的桶列的后一个元素 */
    struct bucket *pLast;    /* 指向具有同一个hash值的桶列的前一个元素 */
    char arKey[1];      /* 必须是最后一个成员,key名称*/} Bucket; 
 typedef struct _hashtable {
    uint nTableSize;/*指定了HashTable的大小,同时它限定了HashTable中能保存Bucket的最大数量
此 数越大,系统为HashTable分配的内存就越多。为了提高计算效率,
系统自动会将nTableSize调整到最小一个不小于nTableSize的2 的整数次方*/
    uint nTableMask;/*nTableMask的值永远是nTableSize – 1,引入这个字段的主要目的是为了提高计算效率*/
    uint nNumOfElements;/*记录HashTable当前保存的数据元素的个数*/
    ulong nNextFreeElement;/*记录HashTable中下一个可用于插入数据元素的arBuckets的索引*/
    Bucket *pInternalPointer;/* Used for element traversal */
    Bucket *pListHead;/*Bucket双向链表的第一个元素*/
    Bucket *pListTail;/*Bucket双向链表的最后一元素*/
    Bucket **arBuckets;/*存储Bucket双向链表*/
    dtor_func_t pDestructor;/*函数指针,在HashTable的增加、修改、删除Bucket时自动调用,用于处理相关数据的清理工作*/
    zend_bool persistent;/*指出了Bucket内存分配的方式。如果persisient为TRUE,则使用操作系统本身的内存分配函数为Bucket分配内存,否则使用PHP的内存分配函数。*/
    unsigned char nApplyCount;/*nApplyCount与bApplyProtection结合提供了一个防止在遍历HashTable时进入递归循环时的一种机制*/
    zend_bool bApplyProtection;} HashTable; 
 
 typedef struct _zend_hash_key {
    char *arKey;
    uint nKeyLength;
    ulong h;} zend_hash_key; typedef zend_bool (*merge_checker_func_t)(HashTable *target_ht, void *source_data, zend_hash_key *hash_key, void *pParam); 
 #define CONNECT_TO_BUCKET_DLLIST(element, list_head) \
(element)-&gt;pNext = (list_head); \
(element)-&gt;pLast = NULL; \
if ((element)-&gt;pNext) { \
    (element)-&gt;pNext-&gt;pLast = (element); \
} #define CONNECT_TO_GLOBAL_DLLIST(element, ht) \
(element)-&gt;pListLast = (ht)-&gt;pListTail; \
(ht)-&gt;pListTail = (element); \
(element)-&gt;pListNext = NULL; \
if ((element)-&gt;pListLast != NULL) { \
    (element)-&gt;pListLast-&gt;pListNext = (element); \
} \
if (!(ht)-&gt;pListHead) { \
    (ht)-&gt;pListHead = (element); \
} \
if ((ht)-&gt;pInternalPointer == NULL) { \
    (ht)-&gt;pInternalPointer = (element); \
} #define ZEND_HASH_IF_FULL_DO_RESIZE(ht) \
if ((ht)-&gt;nNumOfElements &gt; (ht)-&gt;nTableSize) {\
    zend_hash_do_resize(ht); \
} int zend_hash_rehash(HashTable *ht) {
    Bucket *p;
    uint nIndex; 
     memset(ht-&gt;arBuckets, 0, ht-&gt;nTableSize * sizeof(Bucket *));
    p = ht-&gt;pListHead;
    while (p != NULL) {
        nIndex = p-&gt;h &amp; ht-&gt;nTableMask;
        CONNECT_TO_BUCKET_DLLIST(p, ht-&gt;arBuckets[nIndex]);
        ht-&gt;arBuckets[nIndex] = p;
        p = p-&gt;pListNext;
    }
    return SUCCESS;} static int zend_hash_do_resize(HashTable *ht) {
    Bucket **t;     if ((ht-&gt;nTableSize &lt;&lt; 1) &gt; 0) {/* Let&#39;s double the table size */
        t = (Bucket **) perealloc_recoverable(ht-&gt;arBuckets, (ht-&gt;nTableSize &lt;&lt; 1) * sizeof(Bucket *), ht-&gt;persistent);
        if (t) {
            ht-&gt;arBuckets = t;
            ht-&gt;nTableSize = (ht-&gt;nTableSize &lt;&lt; 1);
            ht-&gt;nTableMask = ht-&gt;nTableSize - 1;
            zend_hash_rehash(ht);
            return SUCCESS;
        }
        return FAILURE;
    }
    return SUCCESS;} 
 
 
 #define UPDATE_DATA(ht, p, pData, nDataSize) \
if (nDataSize == sizeof(void*)) { \
if ((p)-&gt;pData != &amp;(p)-&gt;pDataPtr) { \
pefree_rel((p)-&gt;pData, (ht)-&gt;persistent); \
} \
memcpy(&amp;(p)-&gt;pDataPtr, pData, sizeof(void *)); \
(p)-&gt;pData = &amp;(p)-&gt;pDataPtr; \
} else { \
    if ((p)-&gt;pData == &amp;(p)-&gt;pDataPtr) { \
        (p)-&gt;pData = (void *) pemalloc_rel(nDataSize, (ht)-&gt;persistent); \
        (p)-&gt;pDataPtr=NULL; \
    } else { \
        (p)-&gt;pData = (void *) perealloc_rel((p)-&gt;pData, nDataSize, (ht)-&gt;persistent);\
/* (p)-&gt;pDataPtr is already NULL so no need to initialize it */ \
    } \
    memcpy((p)-&gt;pData, pData, nDataSize); \
} #define INIT_DATA(ht, p, pData, nDataSize); \
if (nDataSize == sizeof(void*)) { \
memcpy(&amp;(p)-&gt;pDataPtr, pData, sizeof(void *)); \
(p)-&gt;pData = &amp;(p)-&gt;pDataPtr; \
} else { \
    (p)-&gt;pData = (void *) pemalloc_rel(nDataSize, (ht)-&gt;persistent);\
    if (!(p)-&gt;pData) { \
        pefree_rel(p, (ht)-&gt;persistent); \
        return FAILURE; \
    } \
    memcpy((p)-&gt;pData, pData, nDataSize); \
    (p)-&gt;pDataPtr=NULL; \
} 
 
 static inline ulong zend_inline_hash_func(char *arKey, uint nKeyLength) {
    register ulong hash = 5381; /* variant with the hash unrolled eight times */
    for (; nKeyLength &gt;= 8; nKeyLength -= 8) {
        hash = ((hash &lt;&lt; 5) + hash) + *arKey++;
        hash = ((hash &lt;&lt; 5) + hash) + *arKey++;
        hash = ((hash &lt;&lt; 5) + hash) + *arKey++;
        hash = ((hash &lt;&lt; 5) + hash) + *arKey++;
        hash = ((hash &lt;&lt; 5) + hash) + *arKey++;
        hash = ((hash &lt;&lt; 5) + hash) + *arKey++;
        hash = ((hash &lt;&lt; 5) + hash) + *arKey++;
        hash = ((hash &lt;&lt; 5) + hash) + *arKey++;
    }
    switch (nKeyLength) {
        case 7: hash = ((hash &lt;&lt; 5) + hash) + *arKey++; /* fallthrough... */
        case 6: hash = ((hash &lt;&lt; 5) + hash) + *arKey++; /* fallthrough... */
        case 5: hash = ((hash &lt;&lt; 5) + hash) + *arKey++; /* fallthrough... */
        case 4: hash = ((hash &lt;&lt; 5) + hash) + *arKey++; /* fallthrough... */
        case 3: hash = ((hash &lt;&lt; 5) + hash) + *arKey++; /* fallthrough... */
        case 2: hash = ((hash &lt;&lt; 5) + hash) + *arKey++; /* fallthrough... */
        case 1: hash = ((hash &lt;&lt; 5) + hash) + *arKey++; break;
        case 0: break;
    }
    return hash;}ulong zend_hash_func(char *arKey, uint nKeyLength) {
    return zend_inline_hash_func(arKey, nKeyLength);} //省略了int zend_hash_init(HashTable *ht, uint nSize, hash_func_t pHashFunction, dtor_func_t pDestructor) {
    uint i = 3;
    Bucket **tmp;
    zend_bool persistent = 1; 
     if (nSize &gt;= 0x80000000) {/* prevent overflow */
        ht-&gt;nTableSize = 0x80000000;
    } else {
        while ((1U &lt;&lt; i) &lt; nSize) {
            i++;
        }
        ht-&gt;nTableSize = 1 &lt;&lt; i;
    } 
    ht-&gt;nTableMask = ht-&gt;nTableSize - 1;
    ht-&gt;pDestructor = pDestructor;
    ht-&gt;arBuckets = NULL;
    ht-&gt;pListHead = NULL;
    ht-&gt;pListTail = NULL;
    ht-&gt;nNumOfElements = 0;
    ht-&gt;nNextFreeElement = 0;
    ht-&gt;pInternalPointer = NULL;
    ht-&gt;persistent = persistent;
    ht-&gt;nApplyCount = 0;
    ht-&gt;bApplyProtection = 1; 
 
    tmp = (Bucket **) calloc(ht-&gt;nTableSize, sizeof(Bucket *));
    if (!tmp) {
        return FAILURE;
    }
    ht-&gt;arBuckets = tmp; 
     return SUCCESS;} int zend_hash_add_or_update(HashTable *ht, char *arKey, uint nKeyLength, void *pData, uint nDataSize, void **pDest, int flag) {
    ulong h;
    uint nIndex;
    Bucket *p; 
     if (nKeyLength &lt;= 0) {
        return FAILURE;
    } 
    h = zend_inline_hash_func(arKey, nKeyLength);
    nIndex = h &amp; ht-&gt;nTableMask; 
    p = ht-&gt;arBuckets[nIndex];     while (p != NULL) {
        if ((p-&gt;h == h) &amp;&amp; (p-&gt;nKeyLength == nKeyLength)) {
            if (!memcmp(p-&gt;arKey, arKey, nKeyLength)) {
                if (flag &amp; HASH_ADD) {
                    return FAILURE;
                }                 if (ht-&gt;pDestructor) {
                    ht-&gt;pDestructor(p-&gt;pData);
                }
                UPDATE_DATA(ht, p, pData, nDataSize);
                if (pDest) {
                *pDest = p-&gt;pData;
            }
            return SUCCESS;
        }
    }
    p = p-&gt;pNext;} 
p = (Bucket *) pemalloc(sizeof(Bucket) - 1 + nKeyLength, ht-&gt;persistent);if (!p) {
    return FAILURE;}memcpy(p-&gt;arKey, arKey, nKeyLength);p-&gt;nKeyLength = nKeyLength;INIT_DATA(ht, p, pData, nDataSize);p-&gt;h = h;CONNECT_TO_BUCKET_DLLIST(p, ht-&gt;arBuckets[nIndex]);if (pDest) {*pDest = p-&gt;pData;} 
 
CONNECT_TO_GLOBAL_DLLIST(p, ht);ht-&gt;arBuckets[nIndex] = p; 
 
ht-&gt;nNumOfElements++;ZEND_HASH_IF_FULL_DO_RESIZE(ht); /* If the Hash table is full, resize it */return SUCCESS;} void zend_hash_destroy(HashTable *ht) {
    Bucket *p, *q; 
 
    p = ht-&gt;pListHead;
    while (p != NULL) {
        q = p;
        p = p-&gt;pListNext;
        if (ht-&gt;pDestructor) {
            ht-&gt;pDestructor(q-&gt;pData);
        }
        if (q-&gt;pData != &amp;q-&gt;pDataPtr) {
            pefree(q-&gt;pData, ht-&gt;persistent);
        }
        pefree(q, ht-&gt;persistent);
    }
    pefree(ht-&gt;arBuckets, ht-&gt;persistent); } 
 
 int zend_hash_find(HashTable *ht, char *arKey, uint nKeyLength, void **pData) {
    ulong h;
    uint nIndex;
    Bucket *p; 
 
    h = zend_inline_hash_func(arKey, nKeyLength);
    nIndex = h &amp; ht-&gt;nTableMask; 
    p = ht-&gt;arBuckets[nIndex];
    while (p != NULL) {
        if ((p-&gt;h == h) &amp;&amp; (p-&gt;nKeyLength == nKeyLength)) {
            if (!memcmp(p-&gt;arKey, arKey, nKeyLength)) {
            *pData = p-&gt;pData;
            return SUCCESS;
        }
    }
    p = p-&gt;pNext;}return FAILURE;} 
 void zend_hash_display(HashTable *ht) {
    Bucket *p;
    uint i;
    int flag  = 0 ;     for (i = 0; i &lt; ht-&gt;nTableSize; i++) {
        p = ht-&gt;arBuckets[i];
        flag = 0;
        while (p != NULL) {
            printf("(%d %s &lt;==&gt; 0x%lX %d)   ", i, p-&gt;arKey, p-&gt;h, p-&gt;pNext);
            p = p-&gt;pNext;
            flag = 1;
        }
        if (flag == 1) {
            printf("\n");
        }     } 
    p = ht-&gt;pListTail;
    while (p != NULL) {
        printf("%s &lt;==&gt; 0x%lX\n", p-&gt;arKey, p-&gt;h);
        p = p-&gt;pListLast;
    }}int main() {
    int i;
    char ch[20];
    HashTable ht;
    zend_hash_init(&amp;ht, 50, NULL, NULL);
    for (i = 30; i &lt; 68; i++) {
        sprintf(ch, "%d", i);
        ch[strlen(ch) + 1] = &#39;\0&#39;;
        zend_hash_add_or_update(&amp;ht, ch, strlen(ch) + 1, NULL, 0, NULL, 0);
    } 
    zend_hash_display(&amp;ht);
    zend_hash_destroy(&amp;ht);
    return 0;}?&gt;
Copy after login

以上就是本文的全部内容,希望对大家的学习有所帮助,更多相关内容请关注PHP中文网!

相关推荐:

关于PHP源代码中Zend HashTable的解析

The above is the detailed content of About the analysis of HashTable in PHP source code. For more information, please follow other related articles on the PHP Chinese website!

Related labels:
source:php.cn
Statement of this Website
The content of this article is voluntarily contributed by netizens, and the copyright belongs to the original author. This site does not assume corresponding legal responsibility. If you find any content suspected of plagiarism or infringement, please contact admin@php.cn
Popular Tutorials
More>
Latest Downloads
More>
Web Effects
Website Source Code
Website Materials
Front End Template