uthash 是C的比较优秀的开源代码,它实现了常见的hash操作函数,例如查找、插入、删除等待。该套开源代码采用宏的方式实现hash函数的相关功能,支持C语言的任意数据结构最为key值,甚至可以采用多个值作为key,无论是自定义的struct还是基本数据类型,需要注意的是不同类型的key其操作接口方式略有不通。
使用uthash代码时只需要包含头文件"uthash.h"即可。由于该代码采用宏的方式实现,所有的实现代码都在uthash.h文件中,因此只需要在自己的代码中包含该头文件即可。
1.uthash的效率
uthash的插入、查找、删除的操作时间都是常量,当然这个常量的值收到key以及所选择的hash函数的影响,uthash共提供了7中函数函数,一般情况下选择默认的即可。如果对效率要求特别高时,可以再根据自己的需求选择适合自己的hash函数。
2、uthash的使用
在hash操作中,都是按照“键-值“对的方式进行插、查等操作,在uthash中,其基本数据结构就是一个包含“键-值“对的结构体,另外,该结构体中还包含一个uthash内部使用的hash处理句柄,如下代码所示:
#include"uthash.h" struct my_struct { int id; /* key */ char name[10]; UT_hash_handle hh; /* makes this structure hashable */ };
其中:
- id是键(key);
- name是值,即自己要保存的数据域,这里可以根据自己的需要让它变成结构体指针或者其他类型都可以;
- hh是内部使用的hash处理句柄,在使用过程中,只需要在结构体中定义一个UT_hash_handle类型的变量即可,不需要为该句柄变量赋值,但必须在该结构体中定义该变量。
- Uthash所实现的hash表中可以提供类似于双向链表的操作,可以通过结构体成员hh的
hh.prev
和hh.next获取当前节点的上一个节点或者下一个节点。
3.Key类型为int的简单示例
1)定义一个键为int类型的hash结构体:
#include "uthash.h" struct my_struct { int ikey; /* key */ char value[10]; UT_hash_handle hh; }; struct my_struct *g_users = NULL;
这里需要注意:
- key的类型为int,key的类型不一样,后面的插入、查找调用的接口函数就不一样,因此要求确保key的类型与uthash的接口函数一致。
- 必须提供
UT_hash_handle
变量hh
,无需为其初始化。 - 定义一个hash结构的空指针
users
,用于指向保存数据的hash表,必须初始化为空,在后面的查、插等操作中,uthash内部会根据其是否为空而进行不同的操作。
2)实现自己的查找接口函数:
struct my_struct *find_user(int ikey) { struct my_struct *s; HASH_FIND_INT(g_users, &ikey, s ); return s; }
其实现过程就是先定义一个hash结构体指针变量,然后通过HASH_FIND_INT
接口找到该key所对应的hash结构体。这里需要注意:
- Uthash为整型key提供的查找接口为
HASH_FIND_INT
; - 传给接口
HASH_FIND_INT
的第一个参数就是在1)中定义的指向hash表的指针,传入的第二个参数是整型变量ikey的地址。
3)实现自己的插入接口函数:
void add_user(int ikey, char *value_buf) { struct my_struct *s; HASH_FIND_INT(g_users, &ikey, s); /* 插入前先查看key值是否已经在hash表g_users里面了 */ if (s==NULL) { s = (struct my_struct*)malloc(sizeof(struct my_struct)); s->ikey = ikey; HASH_ADD_INT(g_users, ikey, s ); /* 这里必须明确告诉插入函数,自己定义的hash结构体中键变量的名字 */ } strcpy(s-> value, value_buf); }
由于uthash要求键(key)必须唯一,而uthash内部未对key值得唯一性进行很好的处理,因此它要求外部在插入操作时要确保其key值不在当前的hash表中,这就需要,在插入操作时,先查找hash表看其值是否已经存在,不存在在时再进行插入操作,在这里需要特别注意以下两点:
- 插入时,先查找,当键不在当前的hash表中时再进行插入,以确保键的唯一性。
- 需调用插入接口函数时需要明确告诉接口函数,自己定义的键变量的名字是什么。
4)实现删除接口
void delete_user(int ikey) { struct my_struct *s = NULL; HASH_FIND_INT(g_users, &ikey, s); if (s==NULL) { HASH_DEL(g_users, s); free(s); } }
删除操作的接口函数为HASH_DEL
,只需要告诉该接口要释放哪个hash表(这里是g_users
)里的哪个节点(这里是s),需要注意:释放申请的hash结构体变量,uthash函数只将结构体从hash表中移除,并未释放该结构体所占据的内存。
5)清空hash表
void delete_all() { struct my_struct *current_user, *tmp; HASH_ITER(hh, users, current_user, tmp) { HASH_DEL(g_users,current_user); free(current_user); } }
这里需要注意:uthash内部提供了另外一个清空函数:
HASH_CLEAR(hh, g_users);
函数,但它不释放各节点的内存,因此尽量不要使用它,
6)统计hash表中的已经存在的元素数
该操作使用函数HASH_COUNT
即可获取到当前hash表中的元素数,其用法为:
unsigned int num_users; num_users = HASH_COUNT(g_users); printf("there are %u items\n", num_users);
7、遍历元素
在开发过程中,可能需要对整个hash表进行遍历,这里可以通过hh.next
获取当前元素的下一个元素。具体遍历方法为:
struct my_struct *s, *tmp; HASH_ITER(hh, g_users, s, tmp) { printf("user ikey %d: value %s\n", s->ikey, s->value); /* ... it is safe to delete and free s here */ }
另外还有一种不安全的删除方法,尽量避免使用它:
void print_users() { struct my_struct *s; for(s=g_users; s != NULL; s=s->hh.next) { printf("user ikey %d: value %s\n", s->ikey, s->value); } }
4. 其他类型key的使用
本节主要关于key值类型为其他任意类型,例如整型、字符串、指针、结构体等时的用法。
注意:在使用key值为浮点类型时,由于浮点类型的比较受到精度的影响,例如:1.0000000002被认为与1相等,这些问题在uthash中也存在。
4.1. int类型key
前面就是以int类型的key作为示例,总结int类型key使用方法,可以看到其查找和插入分别使用专用接口:HASH_FIND_INT和ASH_ADD_INT。
4.2. 字符指针char*类型key与字符数组char key[100]类型key
特别注意在Strting类型中,uthash对指针char*和字符数组(例如char key[100])做了区分,这两种情况下使用的接口函数时不一样的。在添加的时候,key的类型为指针时使用接口函数HASH_ADD_KEYPTR,key的类型为字符数组时,使用接口函数HASH_ADD_STR,除了添加的接口不一样外,其他的查找、删除、变量等接口函数都是一样的。
4.3.使用地址作为key
在uthash中也可使用地址做key进行hash操作,使用地址作为key值时,其类型为void*,这样它就可以支持任意类型的地址了。在使用地址作为key时,插入和查找的专用接口函数为HASH_ADD_PTR
和HASH_FIND_PTR
,其余接口是一样的。
4.3.其他非常用类型key
在uthash中还可使用结构体作为key,甚至可以采用组合的方式让多个值作为key,这些在其官方的网站张均有较详细的使用示例。在使用uthash需要注意以下几点:
- 在定义hash结构体时不要忘记定义UT_hash_handle的变量
- 需确保key值唯一,如果插入key-value对时,key值已经存在,再插入的时候就会出错。
- 不同的key值,其增加和查找调用的接口函数不一样,具体可见第4节。一般情况下,不通类型的key,其插入和查找接口函数是不一样的,删除、遍历、元素统计接口是通用的,特殊情况下,字符数组和字符串作为key值时,其插入接口函数不一样,但是查找接口是一样的。
#ifndef UTHASH_H
#define UTHASH_H
#include <string.h> /* memcmp,strlen */
#include <stddef.h> /* ptrdiff_t */
//
stddef.h定义了一些标准宏以及类型.
NULL Null 指针常量
offsetof(type, member-desginator) 获得字段在结构体中的偏移量
ptrdiff_t 带符号的整数类型, 用来表示指针相减的结果类型
wchar_t 宽字符类型
size_t 无符号整数类型, 用来表示sizeof操作符的结果类型
//
#include <stdlib.h> /* exit() */
//
stdlib 头文件即standard library标准库头文件
stdlib 头文件里包含了C、C++语言的最常用的系统函数
该文件包含了C语言标准库函数的定义
stdlib.h里面定义了五种类型、一些宏和通用工具函数。 类型例如size_t、wchar_t、div_t、ldiv_t和lldiv_t; 宏例如EXIT_FAILURE、EXIT_SUCCESS、RAND_MAX和MB_CUR_MAX等等; 常用的函数如malloc()、calloc()、realloc()、free()、system()、atoi()(str-待转换为长整型的字符串)、atol()、rand()、srand()、exit()等等。
//
/* These macros use decltype(在C++中,decltype作为操作符,用于查询表达式的数据类型) or the earlier __typeof GNU extension(表达式).
As decltype is only available in newer compilers (VS2010 or gcc 4.3+
when compiling c++ source) this code uses whatever method is needed
or, for VS2008 where neither is available, uses casting workarounds. */
//编译器环境处理
#ifdef _MSC_VER /* MS compiler */
#if _MSC_VER >= 1600 && defined(__cplusplus) /* VS2010 or newer in C++ mode */
#define DECLTYPE(x) (decltype(x))
#else /* VS2008 or older (or VS2010 in C mode) */
#define NO_DECLTYPE
#define DECLTYPE(x)
#endif
#else /* GNU, Sun and other compilers */
#define DECLTYPE(x) (__typeof(x))
#endif
#ifdef NO_DECLTYPE
#define DECLTYPE_ASSIGN(dst,src) \
do { \
char **_da_dst = (char**)(&(dst)); \
*_da_dst = (char*)(src); \
} while(0)
#else
#define DECLTYPE_ASSIGN(dst,src) \
do { \
(dst) = DECLTYPE(dst)(src); \
} while(0)
#endif
/* a number of the hash function use uint32_t which isn't defined on win32 */
#ifdef _MSC_VER
//
在程序中加入_MSC_VER宏,编译器就能够在调用的时根据其版本自动选择可以链接的lib库版本
_MSC_VER 定义编译器的版本。下面是一些编译器版本的_MSC_VER值(参见扩展阅读中的参考文献2的链接)
MS VC++ 11.0_MSC_VER = 1700 (Visual C++ 2012)
MS VC++ 10.0 _MSC_VER = 1600
MS VC++ 9.0 _MSC_VER = 1500
MS VC++ 8.0 _MSC_VER = 1400
MS VC++ 7.1 _MSC_VER = 1310
MS VC++ 7.0 _MSC_VER = 1300
MS VC++ 6.0 _MSC_VER = 1200
MS VC++ 5.0 _MSC_VER = 1100
typedef unsigned int uint32_t;
typedef unsigned char uint8_t;
#else
#include <inttypes.h> /* uint32_t */
//
inttypes.h
#ifndef __INTTYPES_H_
#define __INTTYPES_H_
typedef signed char int8_t;
typedef unsigned char uint8_t;
typedef int int16_t;
typedef unsigned int uint16_t;
typedef long int32_t;
typedef unsigned long uint32_t;
typedef long long int64_t;
typedef unsigned long long uint64_t;
typedef int16_t intptr_t;
typedef uint16_t uintptr_t;
#endif
//
#endif
#define UTHASH_VERSION 1.9.8
#ifndef uthash_fatal
#define uthash_fatal(msg) exit(-1) /* fatal error (out of memory,etc) */
#endif
#ifndef uthash_malloc
#define uthash_malloc(sz) malloc(sz) /* malloc fcn */
#endif
#ifndef uthash_free
#define uthash_free(ptr,sz) free(ptr) /* free fcn */
#endif
#ifndef uthash_noexpand_fyi
#define uthash_noexpand_fyi(tbl) /* can be defined to log noexpand */
#endif
#ifndef uthash_expand_fyi
#define uthash_expand_fyi(tbl) /* can be defined to log expands */
#endif
/* initial number of buckets */
#define HASH_INITIAL_NUM_BUCKETS 32 /* initial number of buckets */
#define HASH_INITIAL_NUM_BUCKETS_LOG2 5 /* lg2 of initial number of buckets */
#define HASH_BKT_CAPACITY_THRESH 10 /* expand when bucket count reaches */
/* calculate the element whose hash handle address is hhe */
#define ELMT_FROM_HH(tbl,hhp) ((void*)(((char*)(hhp)) - ((tbl)->hho)))
#define HASH_FIND(hh,head,keyptr,keylen,out) \ //
do { \
unsigned _hf_bkt,_hf_hashv; \
out=NULL; \
if (head) { \
HASH_FCN(keyptr,keylen, (head)->hh.tbl->num_buckets, _hf_hashv, _hf_bkt); \
if (HASH_BLOOM_TEST((head)->hh.tbl, _hf_hashv)) { \
HASH_FIND_IN_BKT((head)->hh.tbl, hh, (head)->hh.tbl->buckets[ _hf_bkt ], \
keyptr,keylen,out); \
} \
} \
} while (0)
#ifdef HASH_BLOOM
#define HASH_BLOOM_BITLEN (1ULL << HASH_BLOOM)
#define HASH_BLOOM_BYTELEN (HASH_BLOOM_BITLEN/8) + ((HASH_BLOOM_BITLEN%8) ? 1:0)
#define HASH_BLOOM_MAKE(tbl) \ //
do { \
(tbl)->bloom_nbits = HASH_BLOOM; \
(tbl)->bloom_bv = (uint8_t*)uthash_malloc(HASH_BLOOM_BYTELEN); \
if (!((tbl)->bloom_bv)) { uthash_fatal( "out of memory"); } \
memset((tbl)->bloom_bv, 0, HASH_BLOOM_BYTELEN); \
(tbl)->bloom_sig = HASH_BLOOM_SIGNATURE; \
} while (0)
#define HASH_BLOOM_FREE(tbl) \
do { \
uthash_free((tbl)->bloom_bv, HASH_BLOOM_BYTELEN); \
} while (0)
#define HASH_BLOOM_BITSET(bv,idx) (bv[(idx)/8] |= (1U << ((idx)%8)))
#define HASH_BLOOM_BITTEST(bv,idx) (bv[(idx)/8] & (1U << ((idx)%8)))
#define HASH_BLOOM_ADD(tbl,hashv) \
HASH_BLOOM_BITSET((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1)))
#define HASH_BLOOM_TEST(tbl,hashv) \
HASH_BLOOM_BITTEST((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1)))
#else
#define HASH_BLOOM_MAKE(tbl)
#define HASH_BLOOM_FREE(tbl)
#define HASH_BLOOM_ADD(tbl,hashv)
#define HASH_BLOOM_TEST(tbl,hashv) (1)
#define HASH_BLOOM_BYTELEN 0
#endif
#define HASH_MAKE_TABLE(hh,head) \
do { \
(head)->hh.tbl = (UT_hash_table*)uthash_malloc( \
sizeof(UT_hash_table)); \
if (!((head)->hh.tbl)) { uthash_fatal( "out of memory"); } \
memset((head)->hh.tbl, 0, sizeof(UT_hash_table)); \
(head)->hh.tbl->tail = &((head)->hh); \
(head)->hh.tbl->num_buckets = HASH_INITIAL_NUM_BUCKETS; \
(head)->hh.tbl->log2_num_buckets = HASH_INITIAL_NUM_BUCKETS_LOG2; \
(head)->hh.tbl->hho = (char*)(&(head)->hh) - (char*)(head); \
(head)->hh.tbl->buckets = (UT_hash_bucket*)uthash_malloc( \
HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \
if (! (head)->hh.tbl->buckets) { uthash_fatal( "out of memory"); } \
memset((head)->hh.tbl->buckets, 0, \
HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \
HASH_BLOOM_MAKE((head)->hh.tbl); \
(head)->hh.tbl->signature = HASH_SIGNATURE; \
} while(0)
#define HASH_ADD(hh,head,fieldname,keylen_in,add) \
HASH_ADD_KEYPTR(hh,head,&((add)->fieldname),keylen_in,add)
#define HASH_REPLACE(hh,head,fieldname,keylen_in,add,replaced) \
do { \
replaced=NULL; \
HASH_FIND(hh,head,&((add)->fieldname),keylen_in,replaced); \
if (replaced!=NULL) { \
HASH_DELETE(hh,head,replaced); \
}; \
HASH_ADD(hh,head,fieldname,keylen_in,add); \
} while(0)
#define HASH_ADD_KEYPTR(hh,head,keyptr,keylen_in,add) \
do { \
unsigned _ha_bkt; \
(add)->hh.next = NULL; \
(add)->hh.key = (char*)keyptr; \
(add)->hh.keylen = (unsigned)keylen_in; \
if (!(head)) { \
head = (add); \
(head)->hh.prev = NULL; \
HASH_MAKE_TABLE(hh,head); \
} else { \
(head)->hh.tbl->tail->next = (add); \
(add)->hh.prev = ELMT_FROM_HH((head)->hh.tbl, (head)->hh.tbl->tail); \
(head)->hh.tbl->tail = &((add)->hh); \
} \
(head)->hh.tbl->num_items++; \
(add)->hh.tbl = (head)->hh.tbl; \
HASH_FCN(keyptr,keylen_in, (head)->hh.tbl->num_buckets, \
(add)->hh.hashv, _ha_bkt); \
HASH_ADD_TO_BKT((head)->hh.tbl->buckets[_ha_bkt],&(add)->hh); \
HASH_BLOOM_ADD((head)->hh.tbl,(add)->hh.hashv); \
HASH_EMIT_KEY(hh,head,keyptr,keylen_in); \
HASH_FSCK(hh,head); \
} while(0)
#define HASH_TO_BKT( hashv, num_bkts, bkt ) \
do { \
bkt = ((hashv) & ((num_bkts) - 1)); \
} while(0)
/* delete "delptr" from the hash table.
* "the usual" patch-up process for the app-order doubly-linked-list.
* The use of _hd_hh_del below deserves special explanation.
* These used to be expressed using (delptr) but that led to a bug
* if someone used the same symbol for the head and deletee, like
* HASH_DELETE(hh,users,users);
* We want that to work, but by changing the head (users) below
* we were forfeiting our ability to further refer to the deletee (users)
* in the patch-up process. Solution: use scratch space to
* copy the deletee pointer, then the latter references are via that
* scratch pointer rather than through the repointed (users) symbol.
*/
#define HASH_DELETE(hh,head,delptr) \
do { \
unsigned _hd_bkt; \
struct UT_hash_handle *_hd_hh_del; \
if ( ((delptr)->hh.prev == NULL) && ((delptr)->hh.next == NULL) ) { \
uthash_free((head)->hh.tbl->buckets, \
(head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \
HASH_BLOOM_FREE((head)->hh.tbl); \
uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \
head = NULL; \
} else { \
_hd_hh_del = &((delptr)->hh); \
if ((delptr) == ELMT_FROM_HH((head)->hh.tbl,(head)->hh.tbl->tail)) { \
(head)->hh.tbl->tail = \
(UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \
(head)->hh.tbl->hho); \
} \
if ((delptr)->hh.prev) { \
((UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \
(head)->hh.tbl->hho))->next = (delptr)->hh.next; \
} else { \
DECLTYPE_ASSIGN(head,(delptr)->hh.next); \
} \
if (_hd_hh_del->next) { \
((UT_hash_handle*)((ptrdiff_t)_hd_hh_del->next + \
(head)->hh.tbl->hho))->prev = \
_hd_hh_del->prev; \
} \
HASH_TO_BKT( _hd_hh_del->hashv, (head)->hh.tbl->num_buckets, _hd_bkt); \
HASH_DEL_IN_BKT(hh,(head)->hh.tbl->buckets[_hd_bkt], _hd_hh_del); \
(head)->hh.tbl->num_items--; \
} \
HASH_FSCK(hh,head); \
} while (0)
/* convenience forms of HASH_FIND/HASH_ADD/HASH_DEL */
#define HASH_FIND_STR(head,findstr,out) \
HASH_FIND(hh,head,findstr,strlen(findstr),out)
#define HASH_ADD_STR(head,strfield,add) \
HASH_ADD(hh,head,strfield,strlen(add->strfield),add)
#define HASH_REPLACE_STR(head,strfield,add,replaced) \
HASH_REPLACE(hh,head,strfield,strlen(add->strfield),add,replaced)
#define HASH_FIND_INT(head,findint,out) \
HASH_FIND(hh,head,findint,sizeof(int),out)
#define HASH_ADD_INT(head,intfield,add) \
HASH_ADD(hh,head,intfield,sizeof(int),add)
#define HASH_REPLACE_INT(head,intfield,add,replaced) \
HASH_REPLACE(hh,head,intfield,sizeof(int),add,replaced)
#define HASH_FIND_PTR(head,findptr,out) \
HASH_FIND(hh,head,findptr,sizeof(void *),out)
#define HASH_ADD_PTR(head,ptrfield,add) \
HASH_ADD(hh,head,ptrfield,sizeof(void *),add)
#define HASH_REPLACE_PTR(head,ptrfield,add) \
HASH_REPLACE(hh,head,ptrfield,sizeof(void *),add,replaced)
#define HASH_DEL(head,delptr) \
HASH_DELETE(hh,head,delptr)
/* HASH_FSCK checks hash integrity on every add/delete when HASH_DEBUG is defined.
* This is for uthash developer only; it compiles away if HASH_DEBUG isn't defined.
*/
#ifdef HASH_DEBUG
#define HASH_OOPS(...) do { fprintf(stderr,__VA_ARGS__); exit(-1); } while (0)
#define HASH_FSCK(hh,head) \
do { \
unsigned _bkt_i; \
unsigned _count, _bkt_count; \
char *_prev; \
struct UT_hash_handle *_thh; \
if (head) { \
_count = 0; \
for( _bkt_i = 0; _bkt_i < (head)->hh.tbl->num_buckets; _bkt_i++) { \
_bkt_count = 0; \
_thh = (head)->hh.tbl->buckets[_bkt_i].hh_head; \
_prev = NULL; \
while (_thh) { \
if (_prev != (char*)(_thh->hh_prev)) { \
HASH_OOPS("invalid hh_prev %p, actual %p\n", \
_thh->hh_prev, _prev ); \
} \
_bkt_count++; \
_prev = (char*)(_thh); \
_thh = _thh->hh_next; \
} \
_count += _bkt_count; \
if ((head)->hh.tbl->buckets[_bkt_i].count != _bkt_count) { \
HASH_OOPS("invalid bucket count %d, actual %d\n", \
(head)->hh.tbl->buckets[_bkt_i].count, _bkt_count); \
} \
} \
if (_count != (head)->hh.tbl->num_items) { \
HASH_OOPS("invalid hh item count %d, actual %d\n", \
(head)->hh.tbl->num_items, _count ); \
} \
/* traverse hh in app order; check next/prev integrity, count */ \
_count = 0; \
_prev = NULL; \
_thh = &(head)->hh; \
while (_thh) { \
_count++; \
if (_prev !=(char*)(_thh->prev)) { \
HASH_OOPS("invalid prev %p, actual %p\n", \
_thh->prev, _prev ); \
} \
_prev = (char*)ELMT_FROM_HH((head)->hh.tbl, _thh); \
_thh = ( _thh->next ? (UT_hash_handle*)((char*)(_thh->next) + \
(head)->hh.tbl->hho) : NULL ); \
} \
if (_count != (head)->hh.tbl->num_items) { \
HASH_OOPS("invalid app item count %d, actual %d\n", \
(head)->hh.tbl->num_items, _count ); \
} \
} \
} while (0)
#else
#define HASH_FSCK(hh,head)
#endif
/* When compiled with -DHASH_EMIT_KEYS, length-prefixed keys are emitted to
* the descriptor to which this macro is defined for tuning the hash function.
* The app can #include <unistd.h> to get the prototype for write(2). */
#ifdef HASH_EMIT_KEYS
#define HASH_EMIT_KEY(hh,head,keyptr,fieldlen) \
do { \
unsigned _klen = fieldlen; \
write(HASH_EMIT_KEYS, &_klen, sizeof(_klen)); \
write(HASH_EMIT_KEYS, keyptr, fieldlen); \
} while (0)
#else
#define HASH_EMIT_KEY(hh,head,keyptr,fieldlen)
#endif
/* default to Jenkin's hash unless overridden e.g. DHASH_FUNCTION=HASH_SAX */
#ifdef HASH_FUNCTION
#define HASH_FCN HASH_FUNCTION
#else
#define HASH_FCN HASH_JEN
#endif
/* The Bernstein hash function, used in Perl prior to v5.6 */
#define HASH_BER(key,keylen,num_bkts,hashv,bkt) \
do { \
unsigned _hb_keylen=keylen; \
char *_hb_key=(char*)(key); \
(hashv) = 0; \
while (_hb_keylen--) { (hashv) = ((hashv) * 33) + *_hb_key++; } \
bkt = (hashv) & (num_bkts-1); \
} while (0)
/* SAX/FNV/OAT/JEN hash functions are macro variants of those listed at
*http://eternallyconfuzzled.com/tuts/algorithms/jsw_tut_hashing.aspx */
#define HASH_SAX(key,keylen,num_bkts,hashv,bkt) \
do { \
unsigned _sx_i; \
char *_hs_key=(char*)(key); \
hashv = 0; \
for(_sx_i=0; _sx_i < keylen; _sx_i++) \
hashv ^= (hashv << 5) + (hashv >> 2) + _hs_key[_sx_i]; \
bkt = hashv & (num_bkts-1); \
} while (0)
#define HASH_FNV(key,keylen,num_bkts,hashv,bkt) \
do { \
unsigned _fn_i; \
char *_hf_key=(char*)(key); \
hashv = 2166136261UL; \
for(_fn_i=0; _fn_i < keylen; _fn_i++) \
hashv = (hashv * 16777619) ^ _hf_key[_fn_i]; \
bkt = hashv & (num_bkts-1); \
} while(0)
#define HASH_OAT(key,keylen,num_bkts,hashv,bkt) \
do { \
unsigned _ho_i; \
char *_ho_key=(char*)(key); \
hashv = 0; \
for(_ho_i=0; _ho_i < keylen; _ho_i++) { \
hashv += _ho_key[_ho_i]; \
hashv += (hashv << 10); \
hashv ^= (hashv >> 6); \
} \
hashv += (hashv << 3); \
hashv ^= (hashv >> 11); \
hashv += (hashv << 15); \
bkt = hashv & (num_bkts-1); \
} while(0)
#define HASH_JEN_MIX(a,b,c) \
do { \
a -= b; a -= c; a ^= ( c >> 13 ); \
b -= c; b -= a; b ^= ( a << 8 ); \
c -= a; c -= b; c ^= ( b >> 13 ); \
a -= b; a -= c; a ^= ( c >> 12 ); \
b -= c; b -= a; b ^= ( a << 16 ); \
c -= a; c -= b; c ^= ( b >> 5 ); \
a -= b; a -= c; a ^= ( c >> 3 ); \
b -= c; b -= a; b ^= ( a << 10 ); \
c -= a; c -= b; c ^= ( b >> 15 ); \
} while (0)
#define HASH_JEN(key,keylen,num_bkts,hashv,bkt) \
do { \
unsigned _hj_i,_hj_j,_hj_k; \
unsigned char *_hj_key=(unsigned char*)(key); \
hashv = 0xfeedbeef; \
_hj_i = _hj_j = 0x9e3779b9; \
_hj_k = (unsigned)keylen; \
while (_hj_k >= 12) { \
_hj_i += (_hj_key[0] + ( (unsigned)_hj_key[1] << 8 ) \
+ ( (unsigned)_hj_key[2] << 16 ) \
+ ( (unsigned)_hj_key[3] << 24 ) ); \
_hj_j += (_hj_key[4] + ( (unsigned)_hj_key[5] << 8 ) \
+ ( (unsigned)_hj_key[6] << 16 ) \
+ ( (unsigned)_hj_key[7] << 24 ) ); \
hashv += (_hj_key[8] + ( (unsigned)_hj_key[9] << 8 ) \
+ ( (unsigned)_hj_key[10] << 16 ) \
+ ( (unsigned)_hj_key[11] << 24 ) ); \
\
HASH_JEN_MIX(_hj_i, _hj_j, hashv); \
\
_hj_key += 12; \
_hj_k -= 12; \
} \
hashv += keylen; \
switch ( _hj_k ) { \
case 11: hashv += ( (unsigned)_hj_key[10] << 24 ); \
case 10: hashv += ( (unsigned)_hj_key[9] << 16 ); \
case 9: hashv += ( (unsigned)_hj_key[8] << 8 ); \
case 8: _hj_j += ( (unsigned)_hj_key[7] << 24 ); \
case 7: _hj_j += ( (unsigned)_hj_key[6] << 16 ); \
case 6: _hj_j += ( (unsigned)_hj_key[5] << 8 ); \
case 5: _hj_j += _hj_key[4]; \
case 4: _hj_i += ( (unsigned)_hj_key[3] << 24 ); \
case 3: _hj_i += ( (unsigned)_hj_key[2] << 16 ); \
case 2: _hj_i += ( (unsigned)_hj_key[1] << 8 ); \
case 1: _hj_i += _hj_key[0]; \
} \
HASH_JEN_MIX(_hj_i, _hj_j, hashv); \
bkt = hashv & (num_bkts-1); \
} while(0)
/* The Paul Hsieh hash function */
#undef get16bits
#if (defined(__GNUC__) && defined(__i386__)) || defined(__WATCOMC__) \
|| defined(_MSC_VER) || defined (__BORLANDC__) || defined (__TURBOC__)
#define get16bits(d) (*((const uint16_t *) (d)))
#endif
#if !defined (get16bits)
#define get16bits(d) ((((uint32_t)(((const uint8_t *)(d))[1])) << 8) \
+(uint32_t)(((const uint8_t *)(d))[0]) )
#endif
#define HASH_SFH(key,keylen,num_bkts,hashv,bkt) \
do { \
unsigned char *_sfh_key=(unsigned char*)(key); \
uint32_t _sfh_tmp, _sfh_len = keylen; \
\
int _sfh_rem = _sfh_len & 3; \
_sfh_len >>= 2; \
hashv = 0xcafebabe; \
\
/* Main loop */ \
for (;_sfh_len > 0; _sfh_len--) { \
hashv += get16bits (_sfh_key); \
_sfh_tmp = (uint32_t)(get16bits (_sfh_key+2)) << 11 ^ hashv; \
hashv = (hashv << 16) ^ _sfh_tmp; \
_sfh_key += 2*sizeof (uint16_t); \
hashv += hashv >> 11; \
} \
\
/* Handle end cases */ \
switch (_sfh_rem) { \
case 3: hashv += get16bits (_sfh_key); \
hashv ^= hashv << 16; \
hashv ^= (uint32_t)(_sfh_key[sizeof (uint16_t)] << 18); \
hashv += hashv >> 11; \
break; \
case 2: hashv += get16bits (_sfh_key); \
hashv ^= hashv << 11; \
hashv += hashv >> 17; \
break; \
case 1: hashv += *_sfh_key; \
hashv ^= hashv << 10; \
hashv += hashv >> 1; \
} \
\
/* Force "avalanching" of final 127 bits */ \
hashv ^= hashv << 3; \
hashv += hashv >> 5; \
hashv ^= hashv << 4; \
hashv += hashv >> 17; \
hashv ^= hashv << 25; \
hashv += hashv >> 6; \
bkt = hashv & (num_bkts-1); \
} while(0)
#ifdef HASH_USING_NO_STRICT_ALIASING
/* The MurmurHash exploits some CPU's (x86,x86_64) tolerance for unaligned reads.
* For other types of CPU's (e.g. Sparc) an unaligned read causes a bus error.
* MurmurHash uses the faster approach only on CPU's where we know it's safe.
*
* Note the preprocessor built-in defines can be emitted using:
*
* gcc -m64 -dM -E - < /dev/null (on gcc)
* cc -## a.c (where a.c is a simple test file) (Sun Studio)
*/
#if (defined(__i386__) || defined(__x86_64__) || defined(_M_IX86))
#define MUR_GETBLOCK(p,i) p[i]
#else /* non intel */
#define MUR_PLUS0_ALIGNED(p) (((unsigned long)p & 0x3) == 0)
#define MUR_PLUS1_ALIGNED(p) (((unsigned long)p & 0x3) == 1)
#define MUR_PLUS2_ALIGNED(p) (((unsigned long)p & 0x3) == 2)
#define MUR_PLUS3_ALIGNED(p) (((unsigned long)p & 0x3) == 3)
#define WP(p) ((uint32_t*)((unsigned long)(p) & ~3UL))
#if (defined(__BIG_ENDIAN__) || defined(SPARC) || defined(__ppc__) || defined(__ppc64__))
#define MUR_THREE_ONE(p) ((((*WP(p))&0x00ffffff) << 8) | (((*(WP(p)+1))&0xff000000) >> 24))
#define MUR_TWO_TWO(p) ((((*WP(p))&0x0000ffff) <<16) | (((*(WP(p)+1))&0xffff0000) >> 16))
#define MUR_ONE_THREE(p) ((((*WP(p))&0x000000ff) <<24) | (((*(WP(p)+1))&0xffffff00) >> 8))
#else /* assume little endian non-intel */
#define MUR_THREE_ONE(p) ((((*WP(p))&0xffffff00) >> 8) | (((*(WP(p)+1))&0x000000ff) << 24))
#define MUR_TWO_TWO(p) ((((*WP(p))&0xffff0000) >>16) | (((*(WP(p)+1))&0x0000ffff) << 16))
#define MUR_ONE_THREE(p) ((((*WP(p))&0xff000000) >>24) | (((*(WP(p)+1))&0x00ffffff) << 8))
#endif
#define MUR_GETBLOCK(p,i) (MUR_PLUS0_ALIGNED(p) ? ((p)[i]) : \
(MUR_PLUS1_ALIGNED(p) ? MUR_THREE_ONE(p) : \
(MUR_PLUS2_ALIGNED(p) ? MUR_TWO_TWO(p) : \
MUR_ONE_THREE(p))))
#endif
#define MUR_ROTL32(x,r) (((x) << (r)) | ((x) >> (32 - (r))))
#define MUR_FMIX(_h) \
do { \
_h ^= _h >> 16; \
_h *= 0x85ebca6b; \
_h ^= _h >> 13; \
_h *= 0xc2b2ae35l; \
_h ^= _h >> 16; \
} while(0)
#define HASH_MUR(key,keylen,num_bkts,hashv,bkt) \
do { \
const uint8_t *_mur_data = (const uint8_t*)(key); \
const int _mur_nblocks = (keylen) / 4; \
uint32_t _mur_h1 = 0xf88D5353; \
uint32_t _mur_c1 = 0xcc9e2d51; \
uint32_t _mur_c2 = 0x1b873593; \
uint32_t _mur_k1 = 0; \
const uint8_t *_mur_tail; \
const uint32_t *_mur_blocks = (const uint32_t*)(_mur_data+_mur_nblocks*4); \
int _mur_i; \
for(_mur_i = -_mur_nblocks; _mur_i; _mur_i++) { \
_mur_k1 = MUR_GETBLOCK(_mur_blocks,_mur_i); \
_mur_k1 *= _mur_c1; \
_mur_k1 = MUR_ROTL32(_mur_k1,15); \
_mur_k1 *= _mur_c2; \
\
_mur_h1 ^= _mur_k1; \
_mur_h1 = MUR_ROTL32(_mur_h1,13); \
_mur_h1 = _mur_h1*5+0xe6546b64; \
} \
_mur_tail = (const uint8_t*)(_mur_data + _mur_nblocks*4); \
_mur_k1=0; \
switch((keylen) & 3) { \
case 3: _mur_k1 ^= _mur_tail[2] << 16; \
case 2: _mur_k1 ^= _mur_tail[1] << 8; \
case 1: _mur_k1 ^= _mur_tail[0]; \
_mur_k1 *= _mur_c1; \
_mur_k1 = MUR_ROTL32(_mur_k1,15); \
_mur_k1 *= _mur_c2; \
_mur_h1 ^= _mur_k1; \
} \
_mur_h1 ^= (keylen); \
MUR_FMIX(_mur_h1); \
hashv = _mur_h1; \
bkt = hashv & (num_bkts-1); \
} while(0)
#endif /* HASH_USING_NO_STRICT_ALIASING */
/* key comparison function; return 0 if keys equal */
#define HASH_KEYCMP(a,b,len) memcmp(a,b,len)
/* iterate over items in a known bucket to find desired item */
#define HASH_FIND_IN_BKT(tbl,hh,head,keyptr,keylen_in,out) \
do { \
if (head.hh_head) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,head.hh_head)); \
else out=NULL; \
while (out) { \
if ((out)->hh.keylen == keylen_in) { \
if ((HASH_KEYCMP((out)->hh.key,keyptr,keylen_in)) == 0) break; \
} \
if ((out)->hh.hh_next) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,(out)->hh.hh_next)); \
else out = NULL; \
} \
} while(0)
/* add an item to a bucket */
#define HASH_ADD_TO_BKT(head,addhh) \
do { \
head.count++; \
(addhh)->hh_next = head.hh_head; \
(addhh)->hh_prev = NULL; \
if (head.hh_head) { (head).hh_head->hh_prev = (addhh); } \
(head).hh_head=addhh; \
if (head.count >= ((head.expand_mult+1) * HASH_BKT_CAPACITY_THRESH) \
&& (addhh)->tbl->noexpand != 1) { \
HASH_EXPAND_BUCKETS((addhh)->tbl); \
} \
} while(0)
/* remove an item from a given bucket */
#define HASH_DEL_IN_BKT(hh,head,hh_del) \
(head).count--; \
if ((head).hh_head == hh_del) { \
(head).hh_head = hh_del->hh_next; \
} \
if (hh_del->hh_prev) { \
hh_del->hh_prev->hh_next = hh_del->hh_next; \
} \
if (hh_del->hh_next) { \
hh_del->hh_next->hh_prev = hh_del->hh_prev; \
}
/* Bucket expansion has the effect of doubling the number of buckets
* and redistributing the items into the new buckets. Ideally the
* items will distribute more or less evenly into the new buckets
* (the extent to which this is true is a measure of the quality of
* the hash function as it applies to the key domain).
*
* With the items distributed into more buckets, the chain length
* (item count) in each bucket is reduced. Thus by expanding buckets
* the hash keeps a bound on the chain length. This bounded chain
* length is the essence of how a hash provides constant time lookup.
*
* The calculation of tbl->ideal_chain_maxlen below deserves some
* explanation. First, keep in mind that we're calculating the ideal
* maximum chain length based on the *new* (doubled) bucket count.
* In fractions this is just n/b (n=number of items,b=new num buckets).
* Since the ideal chain length is an integer, we want to calculate
* ceil(n/b). We don't depend on floating point arithmetic in this
* hash, so to calculate ceil(n/b) with integers we could write
*
* ceil(n/b) = (n/b) + ((n%b)?1:0)
*
* and in fact a previous version of this hash did just that.
* But now we have improved things a bit by recognizing that b is
* always a power of two. We keep its base 2 log handy (call it lb),
* so now we can write this with a bit shift and logical AND:
*
* ceil(n/b) = (n>>lb) + ( (n & (b-1)) ? 1:0)
*
*/
#define HASH_EXPAND_BUCKETS(tbl) \
do { \
unsigned _he_bkt; \
unsigned _he_bkt_i; \
struct UT_hash_handle *_he_thh, *_he_hh_nxt; \
UT_hash_bucket *_he_new_buckets, *_he_newbkt; \
_he_new_buckets = (UT_hash_bucket*)uthash_malloc( \
2 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \
if (!_he_new_buckets) { uthash_fatal( "out of memory"); } \
memset(_he_new_buckets, 0, \
2 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \
tbl->ideal_chain_maxlen = \
(tbl->num_items >> (tbl->log2_num_buckets+1)) + \
((tbl->num_items & ((tbl->num_buckets*2)-1)) ? 1 : 0); \
tbl->nonideal_items = 0; \
for(_he_bkt_i = 0; _he_bkt_i < tbl->num_buckets; _he_bkt_i++) \
{ \
_he_thh = tbl->buckets[ _he_bkt_i ].hh_head; \
while (_he_thh) { \
_he_hh_nxt = _he_thh->hh_next; \
HASH_TO_BKT( _he_thh->hashv, tbl->num_buckets*2, _he_bkt); \
_he_newbkt = &(_he_new_buckets[ _he_bkt ]); \
if (++(_he_newbkt->count) > tbl->ideal_chain_maxlen) { \
tbl->nonideal_items++; \
_he_newbkt->expand_mult = _he_newbkt->count / \
tbl->ideal_chain_maxlen; \
} \
_he_thh->hh_prev = NULL; \
_he_thh->hh_next = _he_newbkt->hh_head; \
if (_he_newbkt->hh_head) _he_newbkt->hh_head->hh_prev = \
_he_thh; \
_he_newbkt->hh_head = _he_thh; \
_he_thh = _he_hh_nxt; \
} \
} \
uthash_free( tbl->buckets, tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \
tbl->num_buckets *= 2; \
tbl->log2_num_buckets++; \
tbl->buckets = _he_new_buckets; \
tbl->ineff_expands = (tbl->nonideal_items > (tbl->num_items >> 1)) ? \
(tbl->ineff_expands+1) : 0; \
if (tbl->ineff_expands > 1) { \
tbl->noexpand=1; \
uthash_noexpand_fyi(tbl); \
} \
uthash_expand_fyi(tbl); \
} while(0)
/* This is an adaptation of Simon Tatham's O(n log(n)) mergesort */
/* Note that HASH_SORT assumes the hash handle name to be hh.
* HASH_SRT was added to allow the hash handle name to be passed in. */
#define HASH_SORT(head,cmpfcn) HASH_SRT(hh,head,cmpfcn)
#define HASH_SRT(hh,head,cmpfcn) \
do { \
unsigned _hs_i; \
unsigned _hs_looping,_hs_nmerges,_hs_insize,_hs_psize,_hs_qsize; \
struct UT_hash_handle *_hs_p, *_hs_q, *_hs_e, *_hs_list, *_hs_tail; \
if (head) { \
_hs_insize = 1; \
_hs_looping = 1; \
_hs_list = &((head)->hh); \
while (_hs_looping) { \
_hs_p = _hs_list; \
_hs_list = NULL; \
_hs_tail = NULL; \
_hs_nmerges = 0; \
while (_hs_p) { \
_hs_nmerges++; \
_hs_q = _hs_p; \
_hs_psize = 0; \
for ( _hs_i = 0; _hs_i < _hs_insize; _hs_i++ ) { \
_hs_psize++; \
_hs_q = (UT_hash_handle*)((_hs_q->next) ? \
((void*)((char*)(_hs_q->next) + \
(head)->hh.tbl->hho)) : NULL); \
if (! (_hs_q) ) break; \
} \
_hs_qsize = _hs_insize; \
while ((_hs_psize > 0) || ((_hs_qsize > 0) && _hs_q )) { \
if (_hs_psize == 0) { \
_hs_e = _hs_q; \
_hs_q = (UT_hash_handle*)((_hs_q->next) ? \
((void*)((char*)(_hs_q->next) + \
(head)->hh.tbl->hho)) : NULL); \
_hs_qsize--; \
} else if ( (_hs_qsize == 0) || !(_hs_q) ) { \
_hs_e = _hs_p; \
_hs_p = (UT_hash_handle*)((_hs_p->next) ? \
((void*)((char*)(_hs_p->next) + \
(head)->hh.tbl->hho)) : NULL); \
_hs_psize--; \
} else if (( \
cmpfcn(DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_p)), \
DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_q))) \
) <= 0) { \
_hs_e = _hs_p; \
_hs_p = (UT_hash_handle*)((_hs_p->next) ? \
((void*)((char*)(_hs_p->next) + \
(head)->hh.tbl->hho)) : NULL); \
_hs_psize--; \
} else { \
_hs_e = _hs_q; \
_hs_q = (UT_hash_handle*)((_hs_q->next) ? \
((void*)((char*)(_hs_q->next) + \
(head)->hh.tbl->hho)) : NULL); \
_hs_qsize--; \
} \
if ( _hs_tail ) { \
_hs_tail->next = ((_hs_e) ? \
ELMT_FROM_HH((head)->hh.tbl,_hs_e) : NULL); \
} else { \
_hs_list = _hs_e; \
} \
_hs_e->prev = ((_hs_tail) ? \
ELMT_FROM_HH((head)->hh.tbl,_hs_tail) : NULL); \
_hs_tail = _hs_e; \
} \
_hs_p = _hs_q; \
} \
_hs_tail->next = NULL; \
if ( _hs_nmerges <= 1 ) { \
_hs_looping=0; \
(head)->hh.tbl->tail = _hs_tail; \
DECLTYPE_ASSIGN(head,ELMT_FROM_HH((head)->hh.tbl, _hs_list)); \
} \
_hs_insize *= 2; \
} \
HASH_FSCK(hh,head); \
} \
} while (0)
/* This function selects items from one hash into another hash.
* The end result is that the selected items have dual presence
* in both hashes. There is no copy of the items made; rather
* they are added into the new hash through a secondary hash
* hash handle that must be present in the structure. */
#define HASH_SELECT(hh_dst, dst, hh_src, src, cond) \
do { \
unsigned _src_bkt, _dst_bkt; \
void *_last_elt=NULL, *_elt; \
UT_hash_handle *_src_hh, *_dst_hh, *_last_elt_hh=NULL; \
ptrdiff_t _dst_hho = ((char*)(&(dst)->hh_dst) - (char*)(dst)); \
if (src) { \
for(_src_bkt=0; _src_bkt < (src)->hh_src.tbl->num_buckets; _src_bkt++) { \
for(_src_hh = (src)->hh_src.tbl->buckets[_src_bkt].hh_head; \
_src_hh; \
_src_hh = _src_hh->hh_next) { \
_elt = ELMT_FROM_HH((src)->hh_src.tbl, _src_hh); \
if (cond(_elt)) { \
_dst_hh = (UT_hash_handle*)(((char*)_elt) + _dst_hho); \
_dst_hh->key = _src_hh->key; \
_dst_hh->keylen = _src_hh->keylen; \
_dst_hh->hashv = _src_hh->hashv; \
_dst_hh->prev = _last_elt; \
_dst_hh->next = NULL; \
if (_last_elt_hh) { _last_elt_hh->next = _elt; } \
if (!dst) { \
DECLTYPE_ASSIGN(dst,_elt); \
HASH_MAKE_TABLE(hh_dst,dst); \
} else { \
_dst_hh->tbl = (dst)->hh_dst.tbl; \
} \
HASH_TO_BKT(_dst_hh->hashv, _dst_hh->tbl->num_buckets, _dst_bkt); \
HASH_ADD_TO_BKT(_dst_hh->tbl->buckets[_dst_bkt],_dst_hh); \
(dst)->hh_dst.tbl->num_items++; \
_last_elt = _elt; \
_last_elt_hh = _dst_hh; \
} \
} \
} \
} \
HASH_FSCK(hh_dst,dst); \
} while (0)
#define HASH_CLEAR(hh,head) \
do { \
if (head) { \
uthash_free((head)->hh.tbl->buckets, \
(head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket)); \
HASH_BLOOM_FREE((head)->hh.tbl); \
uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \
(head)=NULL; \
} \
} while(0)
#define HASH_OVERHEAD(hh,head) \
(size_t)((((head)->hh.tbl->num_items * sizeof(UT_hash_handle)) + \
((head)->hh.tbl->num_buckets * sizeof(UT_hash_bucket)) + \
(sizeof(UT_hash_table)) + \
(HASH_BLOOM_BYTELEN)))
#ifdef NO_DECLTYPE
#define HASH_ITER(hh,head,el,tmp) \
for((el)=(head), (*(char**)(&(tmp)))=(char*)((head)?(head)->hh.next:NULL); \
el; (el)=(tmp),(*(char**)(&(tmp)))=(char*)((tmp)?(tmp)->hh.next:NULL))
#else
#define HASH_ITER(hh,head,el,tmp) \
for((el)=(head),(tmp)=DECLTYPE(el)((head)?(head)->hh.next:NULL); \
el; (el)=(tmp),(tmp)=DECLTYPE(el)((tmp)?(tmp)->hh.next:NULL))
#endif
/* obtain a count of items in the hash */
#define HASH_COUNT(head) HASH_CNT(hh,head)
#define HASH_CNT(hh,head) ((head)?((head)->hh.tbl->num_items):0)
typedef struct UT_hash_bucket {
struct UT_hash_handle *hh_head;
unsigned count;
/* expand_mult is normally set to 0. In this situation, the max chain length
* threshold is enforced at its default value, HASH_BKT_CAPACITY_THRESH. (If
* the bucket's chain exceeds this length, bucket expansion is triggered).
* However, setting expand_mult to a non-zero value delays bucket expansion
* (that would be triggered by additions to this particular bucket)
* until its chain length reaches a *multiple* of HASH_BKT_CAPACITY_THRESH.
* (The multiplier is simply expand_mult+1). The whole idea of this
* multiplier is to reduce bucket expansions, since they are expensive, in
* situations where we know that a particular bucket tends to be overused.
* It is better to let its chain length grow to a longer yet-still-bounded
* value, than to do an O(n) bucket expansion too often.
*/
unsigned expand_mult;
} UT_hash_bucket;
/* random signature used only to find hash tables in external analysis */
#define HASH_SIGNATURE 0xa0111fe1
#define HASH_BLOOM_SIGNATURE 0xb12220f2
typedef struct UT_hash_table { /
UT_hash_bucket *buckets;
unsigned num_buckets, log2_num_buckets;
unsigned num_items;
struct UT_hash_handle *tail; /* tail hh in app order, for fast append */
ptrdiff_t hho; /* hash handle offset (byte pos of hash handle in element */
/* in an ideal situation (all buckets used equally), no bucket would have
* more than ceil(#items/#buckets) items. that's the ideal chain length. */
unsigned ideal_chain_maxlen;
/* nonideal_items is the number of items in the hash whose chain position
* exceeds the ideal chain maxlen. these items pay the penalty for an uneven
* hash distribution; reaching them in a chain traversal takes >ideal steps */
unsigned nonideal_items;
/* ineffective expands occur when a bucket doubling was performed, but
* afterward, more than half the items in the hash had nonideal chain
* positions. If this happens on two consecutive expansions we inhibit any
* further expansion, as it's not helping; this happens when the hash
* function isn't a good fit for the key domain. When expansion is inhibited
* the hash will still work, albeit no longer in constant time. */
unsigned ineff_expands, noexpand;
uint32_t signature; /* used only to find hash tables in external analysis */
#ifdef HASH_BLOOM
uint32_t bloom_sig; /* used only to test bloom exists in external analysis */
uint8_t *bloom_bv;
char bloom_nbits;
#endif
} UT_hash_table;
typedef struct UT_hash_handle {
struct UT_hash_table *tbl;
void *prev; /* prev element in app order */
void *next; /* next element in app order */
struct UT_hash_handle *hh_prev; /* previous hh in bucket order */
struct UT_hash_handle *hh_next; /* next hh in bucket order */
void *key; /* ptr to enclosing struct's key */
unsigned keylen; /* enclosing struct's key len */
unsigned hashv; /* result of hash-fcn(key) */
} UT_hash_handle;
#endif /* UTHASH_H */