[PATCH 2/2] XKB: Add debug key actions for grabs & window tree

[Daniel Stone]

Hi,

On Mon, Jun 20, 2011 at 02:12:56PM +1000, Peter Hutterer wrote:
> On Fri, Jun 17, 2011 at 05:36:09PM +0100, Daniel Stone wrote:
> > On Thu, Jun 16, 2011 at 11:01:58AM +1000, Peter Hutterer wrote:
> > > > +        if (dev->deviceGrab.grab->deviceMask)
> > > > +            DebugF("      xi1 event mask 0x%lx\n",
> > > > +                   dev->deviceGrab.grab->deviceMask);
> > > 
> > > only for implicit passive grabs, otherwise eventMask is the device's grab
> > > mask see the massive comment above struct _GrabRec in inputstr.h
> > 
> > I see.  I've just changed the conditional to
> > if (devGrab->implicitGrab && grab->deviceMask) - does that suit?
> 
> I think it's better to check the grab type and then print out eventMask for
> core, and eventMask/deviceMask depending on implicit. Otherwise you'd still
> print out out "core mask: blah" for XI non-implicit grabs.
> 
> if (grab type == core)
>     print event mask
> else if (grab type == xi)
>     print (implicit) ? device mask : event mask

While looking at this, I found something that confused me further still.

In ActivatePassiveGrab, we set up the masks thusly:
    tempGrab.eventMask = deliveryMask;

    /* get the XI and XI2 device mask */
    inputMasks = wOtherInputMasks(win);
    tempGrab.deviceMask = (inputMasks) ? inputMasks->inputEvents[dev->id]: 0;

    if (inputMasks)
        memcpy(tempGrab.xi2mask, inputMasks->xi2mask,
               sizeof(tempGrab.xi2mask));

So, on an XI1.x passive grab, eventMask will be the proper device mask
we actually want, plus a core MotionNotifyMask (which seems harmless,
but weird).  deviceMask will be the union of all XI1.x selections from
all clients on that particular window, and xi2mask will be the union of
all XI2.x selections from all clients on that particular window.

In DeliverGrabbedEvents, we try the following:
    /* try core event */
    if (sendCore && grab->grabtype == GRABTYPE_CORE)
    {
        rc = EventToCore(event, &core, &count);
        [...]
    }

    if (!deliveries)
    {
        rc = EventToXI2(event, &xi2);
        if (rc == Success)
        {
            int evtype = xi2_get_type(xi2);
            mask = grab->xi2mask[XIAllDevices][evtype/8] |
                grab->xi2mask[XIAllMasterDevices][evtype/8] |
                grab->xi2mask[thisDev->id][evtype/8];
            [...]
        }
    }

    if (!deliveries)
    {
        rc = EventToXI(event, &xi, &count);
        if (rc == Success)
        {
            /* try XI event */
            if (grabinfo->fromPassiveGrab  &&
                    grabinfo->implicitGrab)
                mask = grab->deviceMask;
            else
                mask = grab->eventMask;
            [...]
        }
    }

With the [...] being TryClientEvents and setting deliveries if it was
successful.  So, if we have an XI1.x grab from one client on a window
where another client has a XI2 event selection, it looks like we'll
send an XI2 event to the XI1 client and never send an XI1 event.

Shouldn't both the XI2 and XI1 branches be testing for grabtype ==
GRABTYPE_XI or GRABTYPE_XI2, like the core branch?

If that's true, then I'll push v2 with your suggested if/else change.

Cheers,
Daniel