Signals, sockets and threads

[Robert Schuster]

Hi all,
the title looks like fun, eh? :)

In an attempt to get gnu/testlet/java/net/ServerSocket/ReturnOnClose to succeed
on Cacao with the new an shiny VMChannel implementation I found out the Cacao's
Thread.interrupt() does not cause blocking system calls to be interrupted. A
short glimpse a GCJ's code (where the above testcase runs fine) revealed that I
needed to add a pthread_kill (and an obligatory sigaction) to get things 
working.

With that in place I tried to implement a correct VMChannel.accept. The
interesting part looks like this:

  do
    {
      ret = cpnio_accept (fd, (struct sockaddr *) &addr, &alen);
      tmp_errno = errno;

      /* Check if interrupted by Thread.interrupt(). If not then some
       * other unrelated signal interrupted the system function and
       * we should start over again.
       */
      if (tmp_errno == EINTR && JCL_thread_interrupted(env, clazz))
        {
          JCL_ThrowException (env, "java/net/SocketException", strerror
(tmp_errno));
          return -1;
        }
    }

  while (ret == -1);

  if (ret == -1)
    {
      if (EWOULDBLOCK != tmp_errno && EAGAIN != tmp_errno)
        JCL_ThrowException (env, "java/net/SocketException", strerror 
(tmp_errno));
    }

  return ret;

tmp_errno is an int which is set to 0 in the beginning.

The special bit is JCL_thread_interrupted which calls back into the VM
to ask it whether the current thread has its interrupted flag set. My idea is
that a delibarate Thread.interrupt() from inside the VM should stop
VMChannel.accept() while a user running: "killall -SIGUSR1 cacao" from the
console should not.

Is that implementation OK. If not please give me some pointers to make it 
better.

cya
Robert

signature.asc
Description: PGP signature

signature.asc
Description: OpenPGP digital signature