[PATCH 3/6] dix: when converting to XI 1.x events, use the first_valuator's device mode

[Peter Hutterer]

On Mon, Nov 29, 2010 at 11:14:02AM -0500, Chase Douglas wrote:
> On 11/22/2010 10:35 PM, Peter Hutterer wrote:
> > Don't stop processing events when a valuator's mode doesn't match the 0th
> > valuator's mode. Instead, start with the first_valuator in the event and
> > keep stacking them on until the first valuator with a different mode is hit.
> 
> In XI 1.x, a device has one mode. My implementation assumed that the
> generated events must only have valuators that match that mode.
> 
> What is supposed to occur when the XI 1 device is absolute, the first
> two valuators are absolute, the third is relative, and we are generating
> an event with only the third valuator changed?
> 
> My understanding is that we shouldn't expose different mode valuators to
> XI 1.x clients at all. Thus, the event shouldn't be sent at all if it's
> sent as an XI 1.x event. However, I believe this patch would send the
> third axis.

this is a bit of an ambiguous issue in the XI1 specs and I should have
spotted that when reviewing your patches. this effort just isn't needed.
Let's just have a few clarifications first, because there is two
modes that we can perceive.

valuator mode can mean: does the axis send relative events to the client.
i.e. does the event contain -1/-1 or 123/424 when the pointer moved. in XI1,
there is no relative data in any event, all data sent to the client is
absolute.

device mode, which is in the XI specs, is "how does the device react to
input". this is best explained with a touchpad. by default, it is in
relative mode. although the data you get from the device itself is in
absolute coordinates, the driver converts it over to relative coordinates
and then posts this to the server. Thus, if you touch and move, you are
moving the cursor from its current position. If the touchpad was in absolute
mode, the cursor would first jump to the position representing the finger
position.

what happened with the mixed valuator patch set is that we mixed those two
but that's incorrect.
since the events are always absolute anyway, we can keep processing the
event. so even with XI1 clients, if we get the data -1/4833/+10/2828 from
the device the event contains <pointer pos + 1>/4833/<last.valuator[2] +
10>/2828.

the only confusion that may arise in XI1 is that a device that announces
itself to be Relative still has some axes that _behave_ absolute, though
that is not obvious from the data.

Cheers,
  Peter