xkb implementation seems to behave differently from its C counterpart

[acid-bong]

xcffib v1.7.1

I'm trying to reimplement the event loop that https://github.com/xkbmon/xkbmon does, but can't replicate the behaviour, specifically -- the filtering of events. I'm only interested in those that signal about the keyboard layout change, and xkbmon does it like this (the comments are added by me):

    xcb_generic_event_t *event;
    while ((event = xcb_wait_for_event(connection))) {
        // filter 1: it selects the `pad0` field of `event`
        // xcffib's event doesn't have that field
        switch (event->pad0) {
            // and thus can't be identified as a StateNotify type
            // all events have type NewKeyboardNotifyEvent
            case XCB_XKB_STATE_NOTIFY: {
                xcb_xkb_state_notify_event_t *ne = (void *)event;
                // filter 2: it bitwise-compares `changed` field to a constant
                // good news: the constant is present under xcffib.xkb.StatePart.GroupState
                // bad news: the `changed` field doesn't hold any value except 0, thus events can't be filtered by it
                if (ne->changed & XCB_XKB_STATE_PART_GROUP_STATE) {
                    print_layout(connection);
                }

                break;
            }
            default:
                break;
        }
        free(event);
    }

At the moment, the only reliable way to check if the layout has changed is to check the value of event.oldMaxKeyCode, which gets the same index as the result of xkb.GetState(xcffib.xkb.ID.UseCoreKbd).reply().lockedGroup (where xkb = conn(xcffib.xkb.key)). On one hand, it might be somewhat cheaper to use that value myself to get the layout by index than to rerun GetState(), on the other -- it still differs from the C implementation.

Btw, the code I come up with (with some help from, Lord forgive me, ChatGPT (except from the event filter, I found it out myself)) is such: https://pastebin.com/F2QQiUL0

[tych0]

IIUC you're not using the API in exactly the same way (i.e. checking changed masks), which is perhaps why it's not doing the same thing. Does something like this (untested) help?

~ diff -U5 sample.py.orig sample.py
--- sample.py.orig	2025-02-19 07:15:35.736883265 -0700
+++ sample.py	2025-02-19 07:20:34.049897473 -0700
@@ -37,19 +37,16 @@
         0,
         xcffib.xkb.EventType.StateNotify,
         0,
         0,
         0,
-    )
+    ).check()

-    old_index = xkb.GetState(xcffib.xkb.ID.UseCoreKbd).reply().lockedGroup
     print_layout(conn, index)

     while event := conn.wait_for_event():
-        new_index = event.oldMaxKeyCode
-        if new_index != old_index:
+        if isinstance(event, xcffib.xkb.StateNotifyEvent) and event.changed & xcffib.xkb.StatePart.GroupState:
             print_layout(conn, index)
-            old_index = new_index


 if __name__ == "__main__":
-    main()
\ No newline at end of file
+    main()

[acid-bong]

Sadly, adding .check() to xkb.SelectEvents() makes it worse:

Traceback (most recent call last):
  File "/tmp/./xkb.py", line 69, in <module>
    main()
  File "/tmp/./xkb.py", line 44, in main
    ).check()
      ^^^^^^^
  File "/nix/store/77iryi5f198r99gk79dj5n1mw4xnb73h-python3-3.12.8-env/lib/python3.12/site-packages/xcffib/__init__.py", line 339, in check
    assert self.is_checked and self.reply_type is None, (
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AssertionError: Request is not void and checked

Maybe there's a different thing that requires checking?

[acid-bong]

Btw, if I don't retrieve a layout index or the symbols string, conn.wait_for_event() hangs forever and can only be killed with SIGKILL

[tych0]

Oh, sorry, you can drop the .check(), that was me being pedantic.

[tych0]

Yeah, waiting for event should only give you the events you've selected, so i think that's expected. You'll still need the SelectEvents bit

[acid-bong]

Ight, sorry for the pause. The SelectEvents isn't what i'm complaining about, but the event contents: changed attr doesn't take any values other than 0, and the type isn't xcffib.xkb.StateNotifyEvent. That's where it differs from the C impl

[tych0]

What is the type?

[acid-bong]

NewKeyboardNotifyEvent (I thought i mentioned it earlier, but apparently i didn't, my bad)

[tych0]

Ok, and if you xtrace it, do you see the events you expect? I can imagine NewKeyboardNotifyEvents are sent, but perhaps in addition to StateChangeNotify?