Keypad number are translate to letter windows managed by Xpra

[carlos4313]

When using the keypad with NumLock on the number from the keypad do not display
the good information on the terminal managed by Xpra. I.e pressing 5 display E.

[Antoine Martin]

From the attachments:

• server is Ubuntu 19.10
• version 3.0.8-r25889
• server command: /usr/bin/xpra start :5
• client is Microsoft Windows 10
• version 3.0.8-r25879
• client command: xpra attach --encoding=rgb --opengl=no --keyboard-sync=no --min-quality=30 ssh:carlan2@vmfpgacarlan:5
• layout us

I can reproduce the problem with Ubuntu 19.10, not with Fedora.
Also, as I suggested on the mailing list: ​Keypads number output wrong value: fix things by starting your server with:

XPRA_SIMULATE_MODIFIERS=0 xpra start ...

This regression is caused by #2301

[Antoine Martin]

786 client   1 @21.296 mask_to_names(<flags 0 of type Gdk.ModifierType>) GetKeyState(VK_NUMLOCK)=1, names=['mod2']
787 client   1 @21.297 parse_key_event(<Gdk.EventKey object at 0x000000001cabc9f0 (void at 0x0000000007f5f2a0)>, True)=<GTKKeyEvent object, contents: {'modifiers': ['mod2'], 'keyname': 'KP_5', 'keyval': 65461, 'keycode': 101, 'group': 0, 'string': '5', 'pressed': True}>
804 client   1 @21.297 handle_key_action(ClientWindow(1), <GTKKeyEvent object, contents: {'modifiers': ['mod2'], 'keyname': 'KP_5', 'keyval': 65461, 'keycode': 101, 'group': 0, 'string': '5', 'pressed': True}>) wid=1
804 client   1 @21.297 key_handled_as_shortcut: shortcut(KP_5)=None
804 client   1 @21.298 send_delayed_key() delayed_event=None
804 client   1 @21.298 send_key_action(1, <GTKKeyEvent object, contents: {'modifiers': ['mod2'], 'keyname': 'KP_5', 'keyval': 65461, 'keycode': 101, 'group': 0, 'string': '5', 'pressed': True}>)
805 will try levels: [0, 4, 1, 5, 2, 6, 3, 7]
805 do_get_keycode(101, 'KP_5', True, ['mod2'], 0)=84 (level=1, shift=False, mode=0)
805 adding 'shift' to modifiers
805 process_key_action(['key-action', 1, b'KP_5', True, (b'mod2',), 65461, b'5', 101, 0]) server keycode=84, group=0
805 set_keyboard_layout_group(0) config=KeyboardConfig(gb /  / None), current keyboard group=0
805 filtered_modifiers_set(['mod2'])={'mod2'}
805 filtered_modifiers_set(['mod2', 'shift'])={'mod2', 'shift'}
805 make_keymask_match(['mod2', 'shift']) current mask: {'mod2'}, wanted: {'mod2', 'shift'}, ignoring=84/['KP_5'], keys_pressed={}
806 change_mask(set(), False, remove) failed=[]
806 keynames(shift)={'Shift_R', 'Shift_L'}, keycodes=[62, 50], nuisance=False, nuisance keys={'mod2', 'lock'}
806 change_mask(add) ['mod2', 'shift'] modifier 'shift' using keycode 62
806 change_mask({'shift'}, True, add) failed=[]
806 is_modifier(84) not found
806 handle_key((1, True, 'KP_5', 65461, 84, ['mod2', 'shift'], False, True))
807 handle keycode pressing    84: key 'KP_5'
807 fake_key(84, True)
807 scheduling key repeat timer with delay 500 for KP_5 / 84
873 client   1 @21.385 mask_to_names(<flags 0 of type Gdk.ModifierType>) GetKeyState(VK_NUMLOCK)=1, names=['mod2']
874 client   1 @21.385 parse_key_event(<Gdk.EventKey object at 0x000000001cabca90 (void at 0x0000000007f5f2a0)>, False)=<GTKKeyEvent object, contents: {'modifiers': ['mod2'], 'keyname': 'KP_5', 'keyval': 65461, 'keycode': 101, 'group': 0, 'string': '5', 'pressed': False}>
883 client   1 @21.385 handle_key_action(ClientWindow(1), <GTKKeyEvent object, contents: {'modifiers': ['mod2'], 'keyname': 'KP_5', 'keyval': 65461, 'keycode': 101, 'group': 0, 'string': '5', 'pressed': False}>) wid=1
884 client   1 @21.385 key_handled_as_shortcut: shortcut(KP_5)=None
890 client   1 @21.386 send_delayed_key() delayed_event=None
891 client   1 @21.386 send_key_action(1, <GTKKeyEvent object, contents: {'modifiers': ['mod2'], 'keyname': 'KP_5', 'keyval': 65461, 'keycode': 101, 'group': 0, 'string': '5', 'pressed': False}>)
891 process_key_action(['key-action', 1, b'KP_5', False, (b'mod2',), 65461, b'5', 101, 0]) server keycode=84, group=0
892 set_keyboard_layout_group(0) config=KeyboardConfig(gb /  / None), current keyboard group=0
893 filtered_modifiers_set(['shift', 'mod2'])={'mod2', 'shift'}
893 filtered_modifiers_set(['mod2'])={'mod2'}
893 make_keymask_match(['mod2']) current mask: {'mod2', 'shift'}, wanted: {'mod2'}, ignoring=84/['KP_5'], keys_pressed={84: 'KP_5'}
893 keynames(shift)={'Shift_R', 'Shift_L'}, keycodes=[62, 50], nuisance=False, nuisance keys={'mod2', 'lock'}
894 change_mask(remove) ['mod2'] modifier 'shift' using keycode 62
894 change_mask({'shift'}, False, remove) failed=[]
894 change_mask(set(), True, add) failed=[]
894 is_modifier(84) not found
894 handle_key((1, False, 'KP_5', 65461, 84, ['mod2'], False, True))
894 handle keycode unpressing  84: key 'KP_5'
894 fake_key(84, False)

TIL:

• do_get_keycode(101, 'KP_5', True, ['mod2'], 0)=84 (level=1, shift=False, mode=0)
• adding 'shift' to modifiers

So we're trying to match level=1 and adding shift to get there.

It works for X11 clients because we use a native keymap mode, without translating keycodes.

It works with Fedora servers:

do_get_keycode(101, 'KP_5', True, ['mod2'], 0)=84 (level=0, shift=False, mode=0)

But I don't understand why, as keycode 84 looks like this:

keycode  84 = KP_Begin KP_5 KP_Begin KP_5

It should also match level=1!

In any case, the quick and dirty fix I am likely to apply is to ignore level for the keypad keys since we don't synchronize NumLock.

[stdedos]

In any case, the quick and dirty fix I am likely to apply is to ignore level for the keypad keys since we don't synchronize NumLock.

Are you sure you don't sync NumLock?

In my WindowsLaptopClient/XenialServer case, Numlock is always synced, ending up being disabled on the Server (the laptop has its state off by default, because it has no Numpad)

[carlos4313]

I can confirm that using XPRA_SIMULATE_MODIFIERS=0 is a valid workaround.. not sure what is the impact but it work for now
Thanks

[Antoine Martin]

So there are two parts to this:

• r25917 fixes the bug which made Fedora work! (I guess the slightly different keymap triggers the bug)
• r25918 fixes the lookup for keypad keys
• r25919 fix blooper

Closing, feel free to re-open if I've missed something.

[stdedos]

Replying to stdedos:

In any case, the quick and dirty fix I am likely to apply is to ignore level for the keypad keys since we don't synchronize NumLock.

Are you sure you don't sync NumLock?

In my WindowsLaptopClient/XenialServer case, Numlock is always synced, ending up being disabled on the Server (the laptop has its state off by default, because it has no Numpad)

I probably know what's happening:

You don't sync Numlock changes in seamless server, but you do sync them on shadow