[gpm]mixing [im]ps2 + imps2 -> raw -> X

[Alessandro Rubini]

> That makes sense; and I presume in the case of mixing the same protocols,
> having packets 'raw' interlaced from two different inputs messes things
> up.

Actually, I don't think things can mix up. 
<me looking at current source code>
Ok, it can mix up. 

The current implementation is very ugly. Instead of repeating at each
read it should repeat after getMouseData() returns (or just before it
returns, to avoid messing with the clustering loop).

More than mixing up, what can happen here is tha partial packets
are pushed to the fifo. No, I won't post a patch, I'm too busy.

> As opposed to the translated mode which would interlace/merge things
> correctly to present a "single" input to the client (?).

Yes. In that case whole packets are translated at once. And that should
happen for raw repeating as well, IMHO.