Comment 1 for bug 365943

I found the cause of the problem and I dramatically reduced the problem (we tested this in the OEM team).

Sometimes the reported movement deltas are too big to actually take place in 20 milliseconds or less (no human hand would ever be able to move this fast).

Furthermore, a comparison between the timestamps collected in priv->move_hist and the ones I'm getting for each event shows that something goes wrong and that one or more events are skipped, hence the deltas are wrong (i.e. bigger than they should really be). The next event is wrong too.

Currently, as you can see in the attached patch, I tell the driver to ignore these events and it seems to correct the driver's behaviour. Knowing why some events are skipped would be even better though.