inline_all_enqueues not working as expected

[xiewei20082008]

Unit tests fail to work on the code where scoped_actor is involved. I tried to add sched.inline_all_enqueues() before, but it still didn't work. Is there any way to make the scheduler run? The sample code is as follows. Thank you a lot.

behavior inc_impl(event_based_actor *self) {
  return {
    [=](int x) {
      return x+1;
    }
  };
}

behavior a_impl(event_based_actor *self, actor &inc_actor) {
  return {
    [=](int x) {
      if (x==1) {
        scoped_actor sc{self->system()};
        cout << "before request" << endl;                            
        sc->request(inc_actor, caf::infinite, 1).receive(    // get stuck here.
          [=](int x) {
            cout << x << endl;
          },
          [=](caf::error &e) {}
        );
        cout << "after request" << endl;
      }
      else if(x==2) {
        cout << "before request" << endl;
        make_function_view(inc_actor)(1);    // get stuck here.
        cout << "after request" << endl;
      }
    }
  };
}

struct test_fixture : test_coordinator_fixture<> {
    test_fixture() {
    }
};

CAF_TEST_FIXTURE_SCOPE(test1, test_fixture)

CAF_TEST(blocked actor) {
  sched.inline_all_enqueues();
  auto inc_actor = sys.spawn(inc_impl);
  auto a_actor = sys.spawn(a_impl, inc_actor);
  self->send(a_actor, 1);
  // self->send(a_actor, 2);
  run();
}

CAF_TEST_FIXTURE_SCOPE_END()

[Neverlord]

Thanks for providing the minimal example. I've reproduced the issue and I've implemented a patch. If the PR gets merged, I'll also provide the patch in the branch topic/0.17.4 (using the master branch is currently not a good idea, since it contains some breaking changes and more are coming for the upcoming 0.18 release).

However, please note that using a scoped actor inside an event-based actor is not safe. Event-based actors are scheduled cooperatively, this means they are supposed to "play nice" with each other.

The scheduler waits for the actor to return from its message handler before it can run other actors. Blocking a scheduler thread can degrade performance and cause starvation or even deadlocks.

Please consider using self->request().then/await instead or switch to using blocking actors instead if you really need a blocking receive.

[xiewei20082008]

Also thank you for the advise on using event-based actors. I refactored my code - it was of great help to me.

[Neverlord]

My pleasure. :)