- /* I was asked to die (either with kill() or because of a failed element) */
- if (this->iwannadie) {
- this->iwannadie = 0;
- JNIEnv *env = get_current_thread_env();
- XBT_DEBUG("Gonna launch Killed Error");
- // When the process wants to stop before its regular end, we should cut its call stack quickly.
- // The easiest way to do so is to raise an exception that will be catched in its top calling level.
- //
- // For that, we raise a ProcessKilledError that is catched in Process::run() (in msg/Process.java)
- //
- // Throwing a Java exception to stop the actor may be an issue for pure C actors
- // (as the ones created for the VM migration). The Java exception will not be catched anywhere.
- // Bad things happen currently if these actors get killed, unfortunately.
- jxbt_throw_by_name(env, "org/simgrid/msg/ProcessKilledError",
- std::string("Process ") + this->process()->get_cname() + " killed from file JavaContext.cpp");
-
- // (remember that throwing a java exception from C does not break the C execution path.
- // Instead, it marks the exception to be raised when returning to the Java world and
- // continues to execute the C function until it ends or returns).
-
- // Once the Java stack is marked to be unrolled, a C cancel_error is raised to kill the simcall
- // on which the killed actor is blocked (if any).
- // Not doing so would prevent the actor to notice that it's dead, leading to segfaults when it wakes up.
- // This is dangerous: if the killed actor is not actually blocked, the cancel_error will not get catched.
- // But it should be OK in most cases:
- // - If I kill myself, I must do so with Process.kill().
- // The binding of this function in jmsg_process.cpp adds a try/catch around the MSG_process_kill() leading to us
- // - If I kill someone else that is blocked, the cancel_error will unblock it.
- //
- // A problem remains probably if I kill a process that is ready_to_run in the same scheduling round.
- // I guess that this will kill the whole simulation because the victim does not catch the exception.
- // The only solution I see to that problem would be to completely rewrite the process killing sequence
- // (also in C) so that it's based on regular C++ exceptions that would be catched anyway.
- // In other words, we need to do in C++ what we do in Java for sake of uniformity.
- //
- // Plus, C++ RAII would work in that case, too.
-
- XBT_DEBUG("Trigger a cancel error at the C level");
- THROWF(cancel_error, 0, "process cancelled");
- } else {
- Context::stop();
- /* detach the thread and kills it */
- JNIEnv *env = this->jenv;
- env->DeleteGlobalRef(this->jprocess);