[ANNOUNCE] predictable pointer acceleration
simon.thum at gmx.de
Thu Aug 14 23:07:34 PDT 2008
Tiago Vignatti wrote:
> . So the pointer accel wouldn't make any difference here at all with
> our current input flow model.
it's solely about the overshoot. Currently, when the server blocks, the
driver delivers one 'big' event with all motion accumulated. This may
depend on driver/hw, but it's the case on my machine.
However, this event got accelerated almost certainly, even if the user
moved well below accelerated speed. More formally, the assumed relation
between mickey and velocity did not hold.
However, this made the cursor not only jump, but jump acceleration times
farther than you had assumed (even worse with polynomial). That is the
overshoot, and it is fixed by having a proper velocity concept.
With associated timestamps, the code could even follow the precise
motion one made while blocked. I expect this to improve the outcome.
However, the input thread should be able to solve the problem entirely.
BTW, I was usually able to provoke blocking by playing videos
simultaneously. So swapping is not the only culprit I guess.
More information about the xorg