pthread_exit(NULL) segfaults

I have a project using pthreads; there's the main thread and a sub-thread and a pipe between them. It's all working fine, except sometimes it doesn't work. The sub thread runs a command interpreter and an ncurses-based GUI shuttles (some) input over the pipe to it.

I create the thread normally (thread is a pthread_t file-scope variable, interp_start is the function)

if (pthread_create(&thread, NULL, interp_start, NULL)) { perror("couldn't create thread"); return; }

Then, if the interpreter thread receives an "exit" command from the user, it calls interp_exit

fclose(output);
pthread_exit(NULL);

The main thread has a select() that examines, among other things, output's FD and calls a function that read()s from the FD:

int num=read(interp_output[0], &ch, 1);
if (num==0) shell_done();
if (num==-1) perror("read");

The intended behavior, which often works, is to close the FILE* in the thread, which makes the select() report ready, which makes the read() happen, which returns 0, which calls shell_done(). This does, after some simple and unrelated cleanup:

//fprintf(stderr, "joining thread\n");
pthread_join(thread, NULL);
//fprintf(stderr, "joined\n");
exit(EXIT_SUCCESS);

All of this sometimes segfaults. Usually it's fine. If I uncomment those two printfs, if it fails I get neither (it segfaults in pthread_exit) or just the first (it segfaults in pthread_join).

I'm not messing with 'thread' at any other point, and I'm only dealing with null pointers. What's the deal? I'd be looking elsewhere except I'm consistently having pr开发者_运维百科oblems at one of those two lines - once even at restore_sem_to_pool. I think it has to be the way I'm killing the thread, but I'm doing just about the simplest thing possible.

Thanks in advance...

---

Try using valgrind (specifically the "memcheck" part). It can quickly help you to pinpoint invalid memory accesses at runtime, sometimes even including runs of your program that do not crash.

---

I suspect it's a combination of all of these, but the problem went away. It only happened when I typed 'quit' too soon after the program started, but I don't have time to properly debug this now and the assignment is due already anyway. @John - I actually was setting up valgrind for just this purpose. [EDIT] I ran it earlier and it didn't help me with this problem, but I definitely found some memory I wasn't freeing [/EDIT] Everybody else - I think you're right, it's probably some sort of memory corruption. @caf - I forgot to check other thread... but I was seeing the problem in both threads. GDB was pointing me at the one that was actually segfaulting. [EDIT] I checked the other threads and it was either waiting on the pt_join (if the segfault was with the pt_exit) or somewhere just before the pt_exit (if the segfault was in the pt_join) [/EDIT]

Votes to all, but I don't think it's fair to accept an answer here.