[Spice-devel] [spice-gtk v2] widget: Do not ignore unsupported keys from keyboard

[Frediano Ziglio]

> 
> Hi,
> 
> On Thu, Jun 09, 2016 at 10:18:32AM +0100, Frediano Ziglio wrote:
> > If Windows layout does not support a given key the resulting virtual code
> > is set to 0xFF. To avoid losing this raw key (causing the key not been
> > sent to remote machine) detect this condition and handle the key.
> > The check for raw scancode is there to understand if we can handle
> > correctly the key in following code.
> > This problem can happen for instance using a 106 key Japanese keyboard
> > while the layout is set in English; in this case keys like CONVERT not
> > forwarded correctly.
> > 
> > Some note on how to reproduce and test this problem.
> > 
> > Physical way:
> > 1- get a 106 key Japanese keyboard for your Windows client machine;
> > 2- setup your client to English keyboard layout;
> > 3- connect to a Linux machine (no matter the distro or version or
> >    keyboard configuration);
> > 4- open "xinput test-xi2 <device>" command on Linux (device is
> >    the "AT" device in this case);
> > 5- press CONVERT or other keys not present on an English keyboard.
> > 
> > Virtual way (Windows machine on a VM):
> > - set machine remote to VNC;
> > - assure Qmeu has lock-key-sync=off option to vnc;
> > - connect to Windows machine with a VNC client (I suggest TigerVNC
> >   as remote-viewer do some keyboard insertion);
> > - do steps 2, 3, 4 above
> > - press the CONVERT key. If you don't have you can simulate, either
> >   - change VNC client code to insert scancode 0x70 instead of another
> >     key and press this key;
> >   - disconnect main VNC client and use some tool to inject keys
> >     (I use a modifier version of vncdotool).
> > 
> > Signed-off-by: Frediano Ziglio <fziglio at redhat.com>
> > ---
> >  src/spice-widget.c | 38 ++++++++++++++++++++------------------
> >  1 file changed, 20 insertions(+), 18 deletions(-)
> > 
> > diff --git a/src/spice-widget.c b/src/spice-widget.c
> > index 51a2055..b5936bc 100644
> > --- a/src/spice-widget.c
> > +++ b/src/spice-widget.c
> > @@ -1406,7 +1406,7 @@ static gboolean key_event(GtkWidget *widget,
> > GdkEventKey *key)
> >  {
> >      SpiceDisplay *display = SPICE_DISPLAY(widget);
> >      SpiceDisplayPrivate *d = display->priv;
> > -    int scancode;
> > +    int scancode = 0;
> >  #ifdef G_OS_WIN32
> >      int native_scancode;
> >      WORD langid = LOWORD(GetKeyboardLayout(0));
> > @@ -1414,8 +1414,23 @@ static gboolean key_event(GtkWidget *widget,
> > GdkEventKey *key)
> >  #endif
> >  
> >  #ifdef G_OS_WIN32
> > +    /* Try to get scancode with gdk_event_get_scancode.
> > +     * This API is available from 3.22 or if backported.
> > +     */
> > +#if HAVE_GDK_EVENT_GET_SCANCODE
> > +    native_scancode = gdk_event_get_scancode((GdkEvent *) key);
> > +    if (native_scancode) {
> > +        scancode = native_scancode & 0x1ff;
> > +        /* Windows always set extended attribute for these keys */
> > +        if (scancode == (0x100|DIK_NUMLOCK) || scancode ==
> > (0x100|DIK_RSHIFT))
> > +            scancode &= 0xff;
> > +    }
> > +#else
> > +    native_scancode = 0;
> > +#endif
> > +
> >      /* on windows, we ought to ignore the reserved key event? */
> > -    if (key->hardware_keycode == 0xff)
> > +    if (!native_scancode && key->hardware_keycode == 0xff)
> >          return false;
> >  
> >      if (!d->keyboard_grab_active) {
> > @@ -1455,23 +1470,10 @@ static gboolean key_event(GtkWidget *widget,
> > GdkEventKey *key)
> >      if (!d->inputs)
> >          return true;
> >  
> > -    scancode = vnc_display_keymap_gdk2xtkbd(d->keycode_map,
> > d->keycode_maplen,
> > -                                            key->hardware_keycode);
> > +    if (!scancode)
> > +        scancode = vnc_display_keymap_gdk2xtkbd(d->keycode_map,
> > d->keycode_maplen,
> > +                                                key->hardware_keycode);
> 
> I think at this point, the if (!scancode) condition is the same as the
> if (!native_scancode) condition below (as we unconditionnally set
> scancode when native_scancode is not 0 in a hunk above that one)...
> 
> >  #ifdef G_OS_WIN32
> > -    /* Try to get scancode with gdk_event_get_scancode.
> > -     * This API is available from 3.22 or if backported.
> > -     */
> > -#if HAVE_GDK_EVENT_GET_SCANCODE
> > -    native_scancode = gdk_event_get_scancode((GdkEvent *) key);
> > -    if (native_scancode) {
> > -        scancode = native_scancode & 0x1ff;
> > -        /* Windows always set extended attribute for these keys */
> > -        if (scancode == (0x100|DIK_NUMLOCK) || scancode ==
> > (0x100|DIK_RSHIFT))
> > -            scancode &= 0xff;
> > -    }
> > -#else
> > -    native_scancode = 0;
> > -#endif
> >      if (!native_scancode) {
> 
> ... so scancode = vnc_display_keymap_gdk2xtkbd(d->keycode_map,
> d->keycode_maplen, key->hardware_keycode);
> can just go there rather than in an additional block.
> 

Won't it fail to compile on no Windows (didn't tested) ?
The vnc_display_keymap_gdk2xtkbd it's always compiled for all
platforms.
Yes, maybe is getting a bit too complicated.

> Acked-by: Christophe Fergeau <cfergeau at redhat.com>
> 
> Christophe
> 

Frediano