Re: DGA and PointerMoved()

On Wed, 30 Mar 2005, Marc Aurele La France wrote:
> On Sat, 12 Mar 2005, Marc Aurele La France wrote:
>> On Tue, 8 Mar 2005, Thomas Winischhofer wrote:
>>> Marc Aurele La France wrote:
>>> | On Mon, 7 Mar 2005, Mark Vojkovich wrote:

>>> |>> Is it intentional that pScrn->PointerMoved() is being called (by 
>>> whom?)
>>> |>> when DGA is active?

>>> |>> I only receive the cursor's last coordinates here from the time before
>>> |>> DGA was activated. Any mouse movement while DGA is active has no
>>> |>> influence on the coordinates received here.

>>> |>   There is no cursor in DGA mode.  Clients can still get mouse
>>> |> events, but that's not the same thing.  DGA mouse events reflect
>>> |> relative motion rather than absolute, and the bulk of the cursor
>>> |> paths need to be bypassed.  That is, the cursor isn't anywhere,
>>> |> the mouse has been disconnected from the cursor and the relative
>>> |> mouse events are sent directly to the client.

>>> | Yes, but I don't think that answers Thomas's question.

>>> Erm, yes.

>>> I suspected that the cursor stuff is out of order during DGA usage. My
>>> concern was more in the line of why calls to pScrn->PointerMoved() are
>>> issued during DGA activity. That seems like a bug to me. (The generic
>>> xf86PointerMoved does not contain any checks for DGA activity.)

>>> I discovered this while enhancing MergedFB mode which has a very tricky
>>> driver-side PointerMoved() implementation requirement in order to pan
>>> the real display modes within current meta mode. As soon as I switch,
>>> say, vmware to full screen, I get constant coordinates in my
>>> PointerMoved() routine regardless of mouse movement (but synchronous to
>>> mouse movement, ie no random calls but calls clearly related to actual
>>> mouse movement).

>>> I have looked at the cursor code but it goes through many layers, so I
>>> wasn't able to locate the call yet. In xf86Input.c, a
>>> DGAStealMotionEvent is immediately followed by a "continue" statement,
>>> but that obvisouly doesn't keep the event in some way get through to the
>>> PointerMoved() function.

>>> I work around this now, by checking for DGA myself, but that should be
>>> fixed globally.

>> The attached should do just that.

> Anything more on this?

I guess not.  Committed.

Marc.

+----------------------------------+-----------------------------------+
|  Marc Aurele La France           |  work:   1-780-492-9310           |
|  Computing and Network Services  |  fax:    1-780-492-1729           |
|  352 General Services Building   |  email:  [email protected]          |
|  University of Alberta           +-----------------------------------+
|  Edmonton, Alberta               |                                   |
|  T6G 2H1                         |     Standard disclaimers apply    |
|  CANADA                          |                                   |
+----------------------------------+-----------------------------------+
XFree86 developer and VP.  ATI driver and X server internals.