Paolo Bonzini writes ("[PATCH v2] add event queueing to USB HID"):
> For v2 I changed the head/tail implementation of the FIFO buffer
> (which was buggy when the queue became full) to head/count.
> I then removed "have_data", which is the same as count>0
> for pointer devices and the same as "changed" for keyboards.
> This made event merging more correct than in Ian's code and
> easier to understand than my v1.
Thanks. Was my code buggy then ? It's possible; the event merging
case is not easy to trigger in testing.
> I left the "changed" member in USBHIDState, rather than moving it
> to the keyboard, because it is useful to handle the idle period
> (in USB_TOKEN_IN) in a device-independent way. Without it the
> code became more messy.
This leaves the same information recorded in the driver in two places
and is therefore IMO a bad idea. I still think the way I did this is
best: have a common helper function used by the keyboard and pointer
code to deal with the idle handling.
Ian.
