Oh good, that hidapi char should indeed prevent any conflicts - forgot
that Sam had that in there. Once this driver is merged we can add a new
entry to the database right away.
The way the gamepad system works in SDL is that there is SDL_Joystick,
which provides direct access to devices, and SDL_GameController, which
tries to take known hardware from SDL_Joystick and maps them to an
Xbox-style button layout. For us it's generally considered okay for
joysticks to not follow any standards, since you typically only access
them for very specific kinds of devices (flight sim controllers, stuff
like that) and those developers will often want their data as untouched
as humanly possible, while most game developers will only touch
SDL_GameController, which wrangles all the quirks of each device and the
specifics won't matter as much.
-Ethan
On 5/11/20 19:23, François-Xavier Carton wrote:
On Mon, May 11, 2020 at 10:09:28AM -0400, Ethan Lee wrote:
Looks just like the SDL hidapi driver so this should be okay! The only
thing I'm unsure of is the axis inversion, in SDL we read the axes
directly and invert them in the SDL_GameController configuration.
Someone also added a bunch of dead zone work after I wrote the driver,
but unfortunately I don't know the details of that.
The meat of our driver is here...
https://hg.libsdl.org/SDL/file/4298bf108b06/src/joystick/hidapi/SDL_hidapi_gamecube.c#l226
Thanks for the review and pointer to the SDL driver! I think the dead
zone min/max code is a better way to solve the dead zone issue than what
I did in the 3rd commit, as it is an automated solution. I'll implement
something like this.
... and the SDL_GameController config is here:
https://hg.libsdl.org/SDL/file/4298bf108b06/src/joystick/SDL_gamecontrollerdb.h#l614
The '~' char denotes an input value that is flipped after it is read
from the joystick; the vendor/product/version matchup may cause this
config to accidentally flip the axes twice when using the kernel driver.
I tested this on my system and the axes are not getting inverted twice.
This is because the version component of the GUID doesn't match the one
in the db file. The db file has 00016800, whereas "sdl2-jstest --list"
reports 10010000 for my gamepad. Because the GUIDs don't match, the axes
are correct in sdl2-jstest. If I add an entry with my gamepad's GUID,
then the axes are indeed inverted twice.
I don't know why the first part is different (0001 != 1001), but the
second part is different because the SDL hidapi driver adds an 'h'
(0x68) to indicate that it's a hidapi device. See:
https://hg.libsdl.org/SDL/file/4298bf108b06/src/joystick/hidapi/SDL_hidapijoystick.c#l712
So as long as this code is here, the entry in the db file won't get
picked up if the kernel driver is used.
Also, I don't know much about SDL, but looking at sdl2-jstest, it seems
that there are two APIs: joysticks and gamepads. If the db file is only
used for the gamepad API, that means that the axes won't get inverted in
the joystick API if using the SDL hidapi driver. Is that the intended
behaviour?
-François-Xavier
