X server testing

[Thomas Jaeger]

Simon Thum wrote:
> Simon Thum wrote:
>> Peter Hutterer wrote:
>>> On Mon, Jun 22, 2009 at 07:13:12PM -0400, Thomas Jaeger wrote:
>>>> typedef void (*PointerAccelSchemeProc)(
>>>>     DeviceIntPtr /*pDev*/,
>>>>     int /*first_valuator*/,
>>>>     int /*num_valuators*/,
>>>>     int* /*valuators*/,
>>>>     float* /*x_frac*/,
>>>>     float* /*y_frac*/,
>>>>     int /*evtime*/
>>>> );
>>>
>>> punting this to simon (cc'd)
>> A good idea at first glance; I aligned to the scheme modifying
>> valuators and remainder in-place simply because that is how it was
>> done. But, aside for being cleaner in some way, what does it buy us?
>> It's also harder to optimize the no-op case since you always need to
>> copy back.
> Crap, I missed you could still do in-place of course, so this last thing
> is not a valid argument. Still, I'd prefer a more general solution as
> advertised in the previous mail.

Taking non-accelerated code-paths (and crossing sctreens) into account,
it seems the most sensible place to set dev->last.remainder[...] is
positionSprite, so in-place modification doesn't buy us anything.

> Also, if we have a pointer to axes we could strip first_valuator as we
> could as well modify the pointer and num_valuators. But TBH I don't even
> know first_valuators raison d'être. I was able to uncover 2(!) obscure
> sites (vuidFlushAbsEvents in sun_mouse.c and one in jstk.c) where it
> could get non-zero under circumstances. In contrast, the server is
> stuffed with non-obvious logic(to me, at least) depending on it.
There's probably still a few hidden bugs related to first_valuator (I
found two during the last couple weeks).  But some drivers actually
perform the questionable optimization of setting first_valuator to 1 if
the x-coordinate doesn't change, which causes xserver-1.6 (I haven't
tested this on master) to report incorrect coordinates in XI events
(http://sourceforge.net/tracker/?func=detail&atid=1078454&aid=2807915&group_id=229797).
So I'd be very happy with getting rid of first_valuator, but I still
don't quite see how you would want to do that.

Tom