[RFC XI 2.1 - inputproto] Various fixes in response to Peter Hutterer's review

Kristian Høgsberg krh at bitplanet.net
Thu Dec 2 10:20:37 PST 2010


On Thu, Dec 2, 2010 at 10:46 AM, Daniel Stone <daniel at fooishbar.org> wrote:
...
> OK, so would something like this suit you:
>  * at selection time, clients can choose whether or not to receive
>    not-for-you events
>  * at any point during a touch stream, the current owner of a touch can
>    tell the server that it should not send any more events for that
>    touch stream to non-owning clients
>  * at any point during a touch stream, non-owning clients can opt out
>    of receiving further events from that stream

Are these "not-for-you-events" touches outside any of the clients
windows?  Are we allowing to let the client select for all touch
events, anywhere they occur?  That sounds like a bad idea from a
security / client isolation point of view.  If on the other hand,
that's not the case and clients only receive events that land inside
one of their windows and only when they select for it, I really don't
think we need the roundtrips to stop event delivery.    We're not
waking up a ton of clients, we're waking up the one client that's
actually being touched.

Either way, the requests to stop delivery to yourself or other clients
don't make sense to me.  If we're concerned about waking sleeping
clients unecessary, we need to not wake them at all.  Waking up a
bunch of clients so they can tell us that they don't want to be woken
up is broken.  Especially if the premise is that they need to inspect
the event stream a bit to decided whether or not they need the events.
 In that case, the clients wake up do some amount of work, and then
opt of the stream eventually, but the benefit is negligible, since
most gestures are pretty short lived anyway, and the big cost is
waking the client in the first place.

>  * if a client which has not selected for not-for-you events becomes
>    the owner of a touch stream, an event is sent to that client with
>    the entire current state of the touch (i.e. dev->touch->last*)

How would that happen?

Kristian


More information about the xorg-devel mailing list