On 29/08/11 at 13:56 +0900, Lucas Nussbaum wrote:
> On 29/08/11 at 12:43 +0900, KOSAKI Motohiro wrote:
> > > I've just checked, and FreeBSD 8.2 is also affected by this issue.
> > > Test script:
> > > system("false")
> > > sleep 0.5
> > > th = Thread::new { $? }
> > > p th
> > > th.join
> > > p th
> > > p th.value
> > 
> > This is buggy. $? is thread local storage variable. last exit value
> > was set to main thread.
> > because system() was called from it. Therefore it fail even if run on linux.
> 
> Right, sorry.
> 
> > In the other hand, on assert_normal_exit(), io.close is called from
> > sub thread. it's a source
> > of difference.
> 
> Right, it changes the order of termination.

Err, no, it doesn't. the popen process still ends before the thread. I
don't see why io.close would be a source of difference?

In any case, it's a bug in Debian GNU/kfreebsd. With the following test
case:
<-------------
#include <pthread.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/wait.h>

int retval;
int pid;

void * thread_get_retval(void *arg) {
	if (waitpid(pid, &retval, 0) == -1) {
		perror("waitpid");
		exit(1);
	}
	printf("hello\n");
	return(0);
}

int main() {
	pthread_t tid;
	if ((pid = fork()) == 0) {
		return 42;
	}
	printf("PID: %d\n", pid);
	
	if (pthread_create(&tid, NULL, thread_get_retval, NULL)) {
		perror("pthread_create");
		exit(1);
	}

	if (pthread_join( tid, NULL )) {
		perror("pthread_join");
		exit(1);
	}
	printf("Retval: %d\n", WEXITSTATUS(retval));
	return 0;
}
-------------->

Debian GNU/Linux:
PID: 27610
hello
Retval: 42

FreeBSD:
PID: 1172
hello
Retval: 42

Debian GNU/kFreeBSD:
./debian-kfreebsd-amd64:~# ./forkthread 
PID: 719
waitpid: No child processes


Do you still want to apply my patch? I could apply it only in the Debian
package if you prefer.

- Lucas