Bug 45008

what does xev utility reports when you press f10?

I'm having the same problem with gnome-terminal, here is the xev output:

KeyPress event, serial 33, synthetic NO, window 0x1400001,
    root 0x2b5, subw 0x0, time 11437053, (31,432), root:(964,458),
    state 0x0, keycode 76 (keysym 0xffc7, F10), same_screen YES,
    XLookupString gives 0 bytes: 
    XmbLookupString gives 0 bytes: 
    XFilterEvent returns: False

KeyRelease event, serial 33, synthetic NO, window 0x1400001,
    root 0x2b5, subw 0x0, time 11437163, (31,432), root:(964,458),
    state 0x0, keycode 76 (keysym 0xffc7, F10), same_screen YES,
    XLookupString gives 0 bytes: 
    XFilterEvent returns: False

But according to what I see - it is all correct, F10 produces F10.

What is xev output with the previous version of xkb-data?

> KeyPress event, serial 33, synthetic NO, window 0x1400001,
>     root 0x2b5, subw 0x0, time 11437053, (31,432), root:(964,458),
>     state 0x0, keycode 76 (keysym 0xffc7, F10), same_screen YES,
>     XLookupString gives 0 bytes: 
>     XmbLookupString gives 0 bytes: 
>     XFilterEvent returns: False
> 
> KeyRelease event, serial 33, synthetic NO, window 0x1400001,
>     root 0x2b5, subw 0x0, time 11437163, (31,432), root:(964,458),
>     state 0x0, keycode 76 (keysym 0xffc7, F10), same_screen YES,
>     XLookupString gives 0 bytes: 
>     XFilterEvent returns: False

With the previous xkb-data version (2.3), it's essentially the same:

KeyPress event, serial 33, synthetic NO, window 0x1600001,
    root 0x2b5, subw 0x0, time 13172000, (-269,319), root:(664,345),
    state 0x0, keycode 76 (keysym 0xffc7, F10), same_screen YES,
    XLookupString gives 0 bytes: 
    XmbLookupString gives 0 bytes: 
    XFilterEvent returns: False

KeyRelease event, serial 33, synthetic NO, window 0x1600001,
    root 0x2b5, subw 0x0, time 13172097, (-269,319), root:(664,345),
    state 0x0, keycode 76 (keysym 0xffc7, F10), same_screen YES,
    XLookupString gives 0 bytes: 
    XFilterEvent returns: False

Well, in that case God knows why gnome-terminal behaves differently.

Ok, one more thing. Could you please do 'xkbcomp :0 -xkb out.xkb' for both new and old versions? Attach the result here

Created attachment 55894 [details]
xkbcomp output with xkb-data 2.3

Created attachment 55895 [details]
xkbcomp output with xkb-data 2.5

The definitions for F10 are different (I expected that, I did that change) - but they should not matter if the XKB events are identical (and they are!).

What I suspect is that some app (gnome-terminal?) somehow checks the key definition. Why? I have no idea. Something was hardcoded - some workaround, perhaps - there was an issue with functional keys in 2.4. But xev output proves there is nothing to fix in xk-c

It's a bit strange simply closing this... and btw: it's not only gnome-terminal that is affected...

(Especially as it's clearly triggered by the new xkb-data version).

Actually I am ready to reopen if you explain me what exactly would you like me to fix. The x11 (xev) events are absolutely identical - that is the limit of xk-c responsibility, for properly written SW. Is that ok with everybody?

If that's the case, the software is using some other information from XKB config - which is should not, I guess

Returning to the previous version of F10 definition is not an option because it had clearly broken semantics of things like Shift-F10.

So, what would you propose to do?

Well has anyone some idea on whose door we should knock now?

I mean it's probably some GNOME/GTK wide keyboard library or so?

I would start with gnome-terminal perhaps.

I am really sorry I cannot be of much help. When you open the bug in gnome bugzilla, could you please provide the link here in comments - so that I could participate in the discussion.

Thank you and sorry again.

I just tried on my debian, F10 opens g-t menu, aptitide seems not getting it at all. Which is reasonable to some extent...

To reproduce the problem, you need to configure gnome-terminal to pass F10 to the application. Under "Edit → Keyboard Shortcuts", uncheck "Enable the menu shortcut key (F10 by default)".

Yes I can reproduce it. That just convinces me that g-t would be a good starting point - I do not know how g-t implements that feature...

> The definitions for F10 are different (I expected that, I did that change) ->
> but they should not matter if the XKB events are identical (and they are!).

This is not completely true. The new definition of CTRL+ALT consumes Shift,
the old one does not.  With the new definition, in a pedantic interpretation
of section 7.2.1 of the protocol specification of the X Keyboard Extension,
applications must not distinguish Shift-F10 and F10 anymore.

To see wether this is related to problem at hand, one could replace the 
definition of CTRL+ALT (file in types/pc) with the following one:

   type "CTRL+ALT" {
	modifiers = Control+Alt+Shift+LevelThree;
        map[None] = Level1;
        map[Shift] = Level2;
        map[LevelThree] = Level3;
        map[Shift+LevelThree] = Level4;
	map[Control+Alt] = Level5;
	preserve[Shift] = Shift;
	preserve[Shift+LevelThree] = Shift;
	preserve[Shift+LevelThree+Alt] = Shift;
	preserve[Shift+LevelThree+Control] = Shift;
	preserve[Shift+LevelThree+Alt+Control] = Shift;
	preserve[Shift+Alt] = Shift;
	preserve[Shift+Alt+Control] = Shift;
	preserve[Shift+Control] = Shift;
        level_name[Level1] = "Base";
        level_name[Level2] = "Shift";
        level_name[Level3] = "Alt Base";
        level_name[Level4] = "Shift Alt";
	level_name[Level5] = "Ctrl+Alt";
    };

I do not have gnome on my machine, so I cannot test myself.  Note that
the above definition is still not nice, but should be ok to check wether
consumed Shift is to blame.

> Returning to the previous version of F10 definition is not an option because it
> had clearly broken semantics of things like Shift-F10.

Do you have a pointer where the problems with the old definition are
described?

> This is not completely true. The new definition of CTRL+ALT consumes Shift,
> the old one does not.  
How does it affect the behavior of the F10 when shift is not pressed? As you can see, the events are exactly same.

> Do you have a pointer where the problems with the old definition are
> described?
https://bugs.freedesktop.org/show_bug.cgi?id=11822

> How does it affect the behavior of the F10 when shift is not pressed? As you
> can see, the events are exactly same.

Basically, as Shift is consumed, one should not distinguish F10 and
Shift-F10.  This means that pressing Shift-F10 should be handled like
pressing F10; or that pressing F10 should be handled like pressing
Shift-F10.  So it is imaginable that consuming Shift assigns two functions
to pressing F10: the usual one for F10, plus the one for Shift-F10.

That is the idea.  Whether it happens like this, I do not know.  One has to
try or read the relevant source code.  In any event, the original report
speaks about two actions triggered by F10...

I've reassigned (http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=656685#19) the Debian bug to the GNOME guys... hope they can point us to the right place.

For the record, this has already been reported in GNOME's bugzilla at https://bugzilla.gnome.org/show_bug.cgi?id=661973.

Could someone take a look at gdk_x11_keymap_translate_keyboard_state () in http://git.gnome.org/browse/gtk+/tree/gdk/x11/gdkkeys-x11.c, and see if they are supposed to be broken? This bug is sitting for six month in the gnome bugzilla now, and it doesn't look like someone knows a definite answer as to what exactly is broken.

I'm reopening this.
My own investigation confirms the finding in comment 17.
The xkeyboard-config changes in 2.4.1 make XKB translate Shift-F10 into F10 with level one, while telling my that the Shift modifier got consumed.

Also note that xev output being identical is pretty irrelevant here, since xev only reports the core key events, not xkb data. that can be extracted from the combination of key events + xkb maps.

    type "CTRL+ALT" {
        modifiers = Control+Alt+Shift+LevelThree;
        map[None] = Level1;
        map[Shift] = Level2;
        map[LevelThree] = Level3;
        map[Shift+LevelThree] = Level4;
        map[Control+Alt] = Level5;
        preserve[Shift] = Shift;
        preserve[Shift+LevelThree] = Shift;
        level_name[Level1] = "Base";
        level_name[Level2] = "Shift";
        level_name[Level3] = "Alt Base";
        level_name[Level4] = "Shift Alt";
        level_name[Level5] = "Ctrl+Alt";
    };

Here is a modified type for CTRL+ALT that works in my testing.
It is similar to the one in comment 17, but avoids blowing up the map with lots of new entries.

With this map, XkbTranslateKeyCode still translates Shift-F10 into F10 at level1, but the Shift modifier is now preserved, so GTK+ uses it when matching  accelerators, and thus F10 and Shift-F10 can once again have different bindings.

Thanks, that idea with preserve looks nice! Pushed.