链接：https://pan.baidu.com/s/1VKgfZo60mTdxvXrtMvkR6g 

提取码：8d0q 

 

在代码的头部注释解释了gdk_system.c的实现作用。

This file contains a wrapper layer for threading, hence the underscore convention MT_x (Multi-Threading).  As all platforms that MonetDB runs on now support POSIX Threads (pthreads), this wrapping layer has become rather thin.

这是thread的wrapper层，所有thread转为MT_x封装。Monetdb支持POSIX的threads（就是pthreads）

  In the late 1990s when multi-threading support was introduced in MonetDB, pthreads was just emerging as a standard API and not widely adopted yet.  The earliest MT implementation focused on SGI Unix and provided multi- threading using multiple processses, and shared memory.

在1990年，monetdb支持了多线程。Pthreads作为标准api，早期MT实现采用SGI Unix，采用共享内存的多进程，

One of the relics of this model, namely the need to pre-allocate locks and semaphores, and consequently a maximum number of them, has been removed in the latest iteration of this layer.

模型最大的问题是预先分配锁和信号量，因此有一个最大的数量，这在最新的迭代版本中已经移除。

 

2数据结构分析

 

在gdk_system.h中，定义了锁、信号量数据结构，每个锁、信号量都增加了name用于标识。

 

typedef struct MT_Lock {

pthread_mutex_t lock;

char name[MT_NAME_LEN];

#ifdef LOCK_STATS

size_t count;

ATOMIC_TYPE contention;

ATOMIC_TYPE sleep;

struct MT_Lock *volatile next;

struct MT_Lock *volatile prev;

const char *locker;

const char *thread;

#endif

} MT_Lock;

 

/* this is the normal implementation of our pthreads-based read-write lock */

typedef struct MT_RWLock {

pthread_rwlock_t lock;

char name[MT_NAME_LEN];

} MT_RWLock;

 

typedef struct {

sem_t sema;

char name[MT_NAME_LEN];

} MT_Sema;

 

在gdk_system.c中定义了线程数据结构，主线程数据结构。把线程封装起来，同时记录线程的状态、线程的依赖关系。 程序采用宏定义的方式区分win和linux中的线程实现。

static struct winthread {

struct winthread *next;

HANDLE hdl;

DWORD tid;

void (*func) (void *);

void *data;

MT_Lock *lockwait; /* lock we're waiting for */

MT_Sema *semawait; /* semaphore we're waiting for */

struct winthread *joinwait; /* process we are joining with */

const char *working; /* what we're currently doing */

char algorithm[512]; /* the algorithm used in the last operation */

size_t algolen; /* length of string in .algorithm */

ATOMIC_TYPE exited;

bool detached:1, waiting:1;

char threadname[MT_NAME_LEN];

QryCtx *qry_ctx;

} *winthreads = NULL;

static struct winthread mainthread = {

.threadname = "main thread",

.exited = ATOMIC_VAR_INIT(0),

};

static CRITICAL_SECTION winthread_cs;

static DWORD threadslot = TLS_OUT_OF_INDEXES;