| Re: [AD] EVDEV mouse driver |
[ Thread Index |
Date Index
| More lists.liballeg.org/allegro-developers Archives
]
[Sorry for the delay]
> I have added support for event interface (EVDEV)
> for mouse under Linux (e.g. /dev/input/mice or /dev/input/event0).
Nice!
> This is a big (and slow?) driver. Since tablets usually send
> the position in absolute mode, i needed to change lmouse.c
> to be able to manage the mouse position in my driver.
> This means that there are some changes to all Linux mouse
> drivers.
They are fine...
> There is also a change in behavior: when a packet was
> invalid (for ps2 and gmpd, but not for ms) the buttons where
> assumed to be released. Now, for all mice, if a packet is invalid,
> the buttons keep the same state. (If you want the previous behavior,
> i can change it)
...except maybe that one. What's the rationale for changing the behavior ?
Note that I don't know either the rationale for the existing behavior (I'd
say safety though, the invalid packet might be a "button-released" packet),
but I assume it to be still valid unless you have an argument against it.
[No need to modify the patch, I'd do it myself if we settle on my position]
> I don't know how to make the setup utility detect mouse of
> type EVDEV. Two main problems: the device is not /dev/mouse,
> but /dev/input/mouse0, /dev/input/mice or /dev/input/event0.
Then why not hardcode the most commonly used device names (2 or 3) like we do
for /dev/mouse ? On my system, I have /dev/input/mouse, /dev/input/mouse0
and /dev/input/event0 for example.
> Second, a read on an event device returns a whole number of
> 'input_event' structures, and the utility tries to read
> one byte at a time.
Then adapt the code if the device name starts with /dev/input.
> The device file name is set in the config file, but it may
> not always be the same.
Same thing for the regular /dev devices. We only detect the common case.
--
Eric Botcazou