On Fri, 2017-11-03 at 09:37 +0000, Peter Maydell wrote: > On 3 November 2017 at 09:26, Gerd Hoffmann <kra...@redhat.com> wrote: > > Ignoring autorepeat events (as the patch does, but which the commit > > message doesn't mention) makes sense. But I still fail to see why > > we > > should limit hotkeys to the main window. > > Is there documentation about how the input subsystem wants to > handle autorepeat? (ie how UI frontends should present repeating > keys to it, and how keyboard backends should expect to see > repeating keys delivered from it). There's been discussion on > this in another thread about xenfb and vnc...
Well, that issue didn't got much attention, probably due to the fact that the linux kernel does autorepeat in software anyway so at least for linux guests it doesn't matter at all what qemu is doing here. Currently the input layer doesn't special-case repeat events in any way, so they are just passed through to the device in case the ui submits them. >From a design point of view I think it makes sense to have UIs send only keydown and keyup events and not any repeat events. Then leave it to the device emulation (i.e. ps/2 kbs) to actually generate repeat events. That is the only way the repeat rate programmed by the guest will work as expected. > Also I think it would be nice to have written up what the > "expected" set of hotkeys for UI frontends are, so we can > be consistent. I think SDL and gtk are different about what > their ungrab keycombo is, for instance. (I've been assuming > gtk is the "this is the right behaviour" exemplar to follow.) That is another mess I'll plan to look at, because I want rework the display command line switches parsing. So, the current state is: gtk ungrab is ctrl-alt-g. sdl ungrab is ctrl-alt (by default, but see below). cocoa ungrab is ctl-alt too. For reference: virt-viewer (vnc/spice client) likewise uses ctrl-alt. So it seems gtk is the odd one here. Possibly due to how hotkeys are implemented in gtk, could be a GtkAccelGroup simply doesn't support modifier-only hotkeys. Also: We have a -alt-grab switch which makes sdl use the ctrl-alt- shift modifier combo (for both ungrab and the other hotkeys). We have a -ctrl-grab switch which makes sdl use the right ctrl key (again for both ungrab and hotkeys). While being at it: We have a -no-frame switch which makes sdl disable the window frame, but it works on sdl1 only and nobody complained so far ... I'm tempted to just drop support for -alt-grab, -ctrl-grab and -no- frame switches. cheers, Gerd
