Was: Wrapping up 7.4 (finally)

[Peter Hutterer]

On Fri, Jun 13, 2008 at 01:14:01PM +0200, Simon Thum wrote:
> Peter Hutterer wrote:
>> As said, technically it's an ABI change but that's being anal. Simon 
>> claims no
>> driver is affected, Daniel is happy with it, and it has significant
>> improvements. I'm willing to ACK it.
> IMO, the d*remainders have the right normalization, but the wrong 
> visibility. As I view things, they are an implementation artifact to the 
> acceleration and should have less exposure.
> My completely uninformed suggestion would therefore be to add valuators to 
> the devprivates system (or vice versa, you get the idea) in the next 
> release. Is that overkill?

A better fix may be to reduce the information passed around between input
drivers and the server. The driver doesn't really need to know anything but
the fd and the configure options, everything else is done through defined
APIs. I remember daniel mentioning a similar desire previously.

Our current ABI is ridiculous, we pass these huge structs around that are
never actually used, yet we can't change them for risk of an ABI break.
We basically need to go through all device drivers (the popular ones anyway)
and analyse what is unnessecarily passed around. However, I'm not sure if that
will happen for 1.6, but if you're up for the task, go for it.  

> Alternatively, I'd make last[xy] and d[xy]remaind valuator members work 
> like axisVal, because that's how it should be. OK, I'm yet to see a 5D 
> mouse, and that alone would not solve the visiblity problem.

Master already does last.valuators[MAX_VALUATORS] instead of last_x/y [1]. I
guess remaind needs to be changed as well then.

Cheers,
  Peter

[1]  6c9e9f8a40e20fb1761440acd2755f5fd31f4d44