Pete's Log: silly pthreads

Entry #1073, Sat, July 28, 2001, 14:52 EST (Coding, Hacking, & CS stuff)
(posted when I was 23 years old.)

---

I found myself rather worried today that I was leaking threads in some unimaginable fashion in the pilsner. But then I discovered it was all just due to an oddity in the linux pthreads implementation. please forgive the fact that I'm synchronizing with sleep statements, and consider the following code:

#include <stdio.h>
#include <pthread.h>
#include <unistd.h>

void* foo(void* f)
{
  printf("now 2\n");
  sleep(3);
  printf("now 3\n");
  return NULL;
}

int
main()
{
  pthread_t t;
  printf("now 1\n");
  sleep(3);
  pthread_create(&t, NULL, foo, NULL);
  sleep(6);
  pthread_join(t, NULL);
  printf("now 4\n");
  sleep(3);
  return 0;
}

If you compile this code and call the binary threadfu, for example, and then run it and do a "ps aux | grep threadfu" each time it prints out a "now #" ... you may find the same odd results I did. At "now 1" there was one threadfu process. At "now 2" there were three. At "now 3" I see two threadfu processes, and at "now 4" I still see two threadfu processes. At every point, other than "now 1" there was one threadfu process more than I expected. Odd. But at least this should indicate that I'm not leaking threads. The linux pthread implementation seems to be doing that. But at least if I throw more pthread_create's into the code there's only one extra process and not one per thread. I'm guessing the extra process probably serves some useful purpose, such as reaping dead threads.

---