Possible regression with "Add hysteresis-based noise reduction" in synaptics driver

[Simon Thum]

On 11/23/2011 12:47 AM, Chase Douglas wrote:
> On 11/22/2011 03:19 PM, Simon Thum wrote:
>> Before I forget, it's possible (in theory) for the kernel to supply
>> "fuzz" values, but I haven't seen that in practical use. Those could
>> make it into your property, but I suspect that is what happens.
> 
> I think the reason fuzz params haven't been used much is because they
> are linux specific. Perhaps we should use them where available, though.
We do, see event_get_abs() in eventcomm.c. I wanted to say I *doubt*
that's what happens, just to be sure.


> 
> However, it sounds to me like this is an issue of per-axis defuzzing,
> whereas what is really needed is distance-based defuzzing. I'm basing
> this on Peter's explanation that his circles end up more like squares
> with rounded corners.
That might seem to solve the issue, but I do not think that the noise
present on the axes is co-dependent. So, an euklidean distance may have
less than favourable noise suppression characteristics.

I think it's better to go with a lower default, and if people have a
need they can reduce to zero if they like. I have attached a patch which
provides some more rationale.

IMO it should be a concious (read: with testing) decision made by more
people than me and you to apply the patch, or what else to do about it.

Cheers,

Simon