[SOLVED] pthread_cancel doesn't work

sshh2010_2020 · 12-12-2010, 10:14 PM

Hi, i am writing a thread programming on the linux(federol core 8), but i really don't what is wrong with it.
The code is below:

Code:

#include<iostream>
#include<unistd.h>
#include<pthread.h>
using namespace std;

pthread_t pid[3];

void* thread_run_1(void* arg)
{
   if (pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, NULL) != 0)
       return NULL;

    cout<<"Now in the thread 1"<<endl;

    int state, oldstate;

    state = PTHREAD_CANCEL_DEFERRED;
    pthread_setcancelstate(state, &oldstate);

    cout<<"oldstate is "<<(state == oldstate? "Deferred":"Async")<<endl;


	
    for (int i = 1; i<= 1e9; ++i);  //wait for thread2 to cancel this thread	
    //sleep(2);			//wait for thread2 to cancel this thread
   
    cout<<"before testcancel"<<endl;
    pthread_setcancelstate(PTHREAD_CANCEL_ENABLE,NULL);
    pthread_testcancel();
    pthread_setcancelstate(PTHREAD_CANCEL_DISABLE,NULL);

    cout<<"after testcancel"<<endl;
    cout<<"thread 1 done!"<<endl;
	
    return NULL;

}

void* thread_run_2(void* arg)
{
    cout<<"Now in the thread 2"<<endl;
    pthread_cancel(pid[1]);
    sleep(6);				//wait for thread1 to end
    cout<<"thread 2 done!"<<endl;

    return NULL;
}

int main()
{
     pid[0] = pthread_self();
    if (pthread_create(&pid[1],NULL,thread_run_1,NULL) != 0)
    {
        cout<<"error create thread 1"<<endl;
        return -1;
    }
    if (pthread_create(&pid[2],NULL,thread_run_2,NULL) != 0)
    {
        cout<<"error create thread 2"<<endl;
        return -1;
    }

    pthread_join(pid[1], NULL);
    pthread_join(pid[2], NULL);

    cout<<"Main thread done!"<<endl;
    return 0;
}

while the output is:

Now in the thread 1
oldstate is Async
Now in the thread 2
FATAL: exception not rethrown
before testcancel
Aborted

howerver, the output what's I expected is below:
Now in the thread 1
oldstate is Deferred
before testcancel
Now in the thread 2
thread 2 done!
Main thread done!

I am a newbie that i don't know how to modify it, could anybody help me, thx

neonsignal · 12-13-2010, 06:01 AM

The main issue is that the cancel type is set to DEFERRED using the wrong function (it should have been setcanceltype, not setcancelstate).

The exact sequence of the output can vary, because you start the two threads at nearly the same time, you cannot guarantee the ordering of these. And the thread 2 may complete before the thread 1 loop finishes and reaches a point where the cancellation will occur (depending on which takes longer, the 1e9 loop or the 6 second delay). Also, the 'before testcancel' occurs after the thread 2 has started (because the 1e9 loop takes quite a while), but I suspect that is what you intended.

sshh2010_2020 · 12-13-2010, 08:03 AM

Quote:

Originally Posted by neonsignal

The main issue is that the cancel type is set to DEFERRED using the wrong function (it should have been setcanceltype, not setcancelstate).

The exact sequence of the output can vary, because you start the two threads at nearly the same time, you cannot guarantee the ordering of these. And the thread 2 may complete before the thread 1 loop finishes and reaches a point where the cancellation will occur (depending on which takes longer, the 1e9 loop or the 6 second delay). Also, the 'before testcancel' occurs after the thread 2 has started (because the 1e9 loop takes quite a while), but I suspect that is what you intended.

hi, thanks for your patience in helping me.
The output is what's I wanted after I change the funtion setcancelstate() to setcanceltype().

howerver, there is another question confuses me, and I don't know why.
That is when I comment out 4 lines in the function thread_run_1(), the output is as same as the first wrong output.
The whole changed code is below:

Code:

#include<iostream>
#include<unistd.h>
#include<pthread.h>
using namespace std;

pthread_t pid[3];

void* thread_run_1(void* arg)
{
   //if (pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, NULL) != 0)
   //    return NULL;

    cout<<"Now in the thread 1"<<endl;

    int state, oldstate;

    state = PTHREAD_CANCEL_DEFERRED;
    pthread_setcanceltype(state, &oldstate);

    cout<<"oldstate is "<<(state == oldstate? "Deferred":"Async")<<endl;

	
    for (int i = 1; i<= 1e9; ++i);  //wait for thread2 to cancel this thread	
    //sleep(2);			//wait for thread2 to cancel this thread
   
    cout<<"before testcancel"<<endl;
    //pthread_setcancelstate(PTHREAD_CANCEL_ENABLE,NULL);
    pthread_testcancel();
    //pthread_setcancelstate(PTHREAD_CANCEL_DISABLE,NULL);

    cout<<"after testcancel"<<endl;
    cout<<"thread 1 done!"<<endl;
	
    return NULL;

}

void* thread_run_2(void* arg)
{
    cout<<"Now in the thread 2"<<endl;
    pthread_cancel(pid[1]);
    sleep(6);				//wait for thread1 to end
    cout<<"thread 2 done!"<<endl;

    return NULL;
}

int main()
{
     pid[0] = pthread_self();
    if (pthread_create(&pid[1],NULL,thread_run_1,NULL) != 0)
    {
        cout<<"error create thread 1"<<endl;
        return -1;
    }
    if (pthread_create(&pid[2],NULL,thread_run_2,NULL) != 0)
    {
        cout<<"error create thread 2"<<endl;
        return -1;
    }

    pthread_join(pid[1], NULL);
    pthread_join(pid[2], NULL);

    cout<<"Main thread done!"<<endl;
    return 0;
}

I just really don't know why we had to use the funtion pthread_setcancelstate() 3 times.

neonsignal · 12-16-2010, 05:04 AM

To reduce the confusion, it would be worth having the two threads write to different streams (since streams are not threadsafe). For trial purposes, perhaps just use cout and cerr. In a real application, you would need to mutex protect the shared stream.

Alternatively, just have a small (eg 1 second) delay before you start the second thread; this will separate the debug output enough that you won't get the crashes caused by using the same stream in two threads. But keep in mind that such code is inherently unsafe.

Quote:

I just really don't know why we had to use the function pthread_setcancelstate() 3 times.

I don't know why the cancel was disabled at all. I thought you wrote this code?

sshh2010_2020 · 12-16-2010, 08:16 AM

Quote:

Originally Posted by neonsignal

To reduce the confusion, it would be worth having the two threads write to different streams (since streams are not threadsafe). For trial purposes, perhaps just use cout and cerr. In a real application, you would need to mutex protect the shared stream.

Alternatively, just have a small (eg 1 second) delay before you start the second thread; this will separate the debug output enough that you won't get the crashes caused by using the same stream in two threads. But keep in mind that such code is inherently unsafe.

I have tried the 2 metheds you mentioned above, however, both of them failed.
Thank you very much all the same for helping me.
I have been learning the Interprocess Communication just for few days, so I guess I won't figure out the problem encountered right row.

Quote:

Originally Posted by neonsignal

I don't know why the cancel was disabled at all. I thought you wrote this code?

The method that use the function pthread_setcancelstate() 3 times doesn't been raised by me. I learn it in the net with the link below:
http://www.linuxquestions.org/questi...thread-633072/