why are events duplicated when polling and reading /dev/input/event file descriptors with or without libevdev

Question

when polling /dev/input/event file descriptors, i am seeing events duplicated, why is that? this happens when using just read, but also seeing this when using libevdev

here is code i was playing with using only poll/read

 while(true) {
    poll(pfds, nfds, -1);
    for (int i = 0; i < nfds; i++) {
      char buffer[100] = "";

      if (pfds[i].revents != 0) {
        if (pfds[i].revents & POLLIN) {
          read(pfds[i].fd, buffer, sizeof(buffer));
          syslog(LOG_INFO, "got data: %s", buffer);
        }
      }
    }
   }

here is code i am playing with when using poll/libdev

while (true) {
    int ready = 0;

    // poll pfds with no timeout
    ready = poll(pfds, num_kbds, -1);

    // check for errors
    if (ready == -1) {
      return -1;
    }

    // handle poll return
    for (int i = 0; i < num_kbds; i++) {
      // check poll result
      if (pfds[i].revents & POLLIN) {
        // get event
        rcs[i] = libevdev_next_event(ldevs[i], LDEVFLAGS, &evs[i]);
        // process event
        process_event(rcs[i], &evs[i]);
      }
    }
  }

i can resolve this issue by reading twice, or calling libevdev_next_event twice in a row, but i don't understand why i am seeing events duplicated

so, the following code using poll/libevdev would resolve the issue

if (pfds[i].revents & POLLIN) {
        // get event
        rcs[i] = libevdev_next_event(ldevs[i], LDEVFLAGS, &evs[i]);
        rcs[i] = libevdev_next_event(ldevs[i], LDEVFLAGS, &evs[i]);
        // process event
        process_event(rcs[i], &evs[i]);
}

it feels weird and unnecessary and broken to solve the issue this way, and i still don't understand what's going on. can anyone shed some light on this?

Answer 1

the events are not duplicated as i thought, there are two sets of events being sent. one for press and one for release, which affects the behavior of poll leading to seemingly strange behavior. i did not understand that evdev press/release would function this way. one solution is using blocking reads with libevdev, then processing events by press & release.